korg/linux
0
0
Fork 0
mirror of https://mirrors.bfsu.edu.cn/git/linux.git synced 2024-11-17 01:04:19 +08:00
Code Issues Packages Projects Releases Wiki Activity
a3883df393
linux/kernel/rcu/srcutree.c

1228 lines
41 KiB
C
Raw Normal View History

srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
/*
* Sleepable Read-Copy Update mechanism for mutual exclusion.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, you can access it online at
* http://www.gnu.org/licenses/gpl-2.0.html.
*
* Copyright (C) IBM Corporation, 2006
* Copyright (C) Fujitsu, 2012
*
* Author: Paul McKenney <paulmck@us.ibm.com>
* Lai Jiangshan <laijs@cn.fujitsu.com>
*
* For detailed explanation of Read-Copy Update mechanism see -
* Documentation/RCU/ *.txt
*
*/
#include <linux/export.h>
#include <linux/mutex.h>
#include <linux/percpu.h>
#include <linux/preempt.h>
#include <linux/rcupdate_wait.h>
#include <linux/sched.h>
#include <linux/smp.h>
#include <linux/delay.h>
srcu: Specify auto-expedite holdoff time On small systems, in the absence of readers, expedited SRCU grace periods can complete in less than a microsecond. This means that an eight-CPU system can have all CPUs doing synchronize_srcu() in a tight loop and almost always expedite. This might actually be desirable in some situations, but in general it is a good way to needlessly burn CPU cycles. And in those situations where it is desirable, your friend is the function synchronize_srcu_expedited(). For other situations, this commit adds a kernel parameter that specifies a holdoff between completing the last SRCU grace period and auto-expediting the next. If the next grace period starts before the holdoff expires, auto-expediting is disabled. The holdoff is 50 microseconds by default, and can be tuned to the desired number of nanoseconds. A value of zero disables auto-expediting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 05:03:11 +08:00
#include <linux/module.h>
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
#include <linux/srcu.h>
#include "rcu.h"
srcu: Debloat the <linux/rcu_segcblist.h> header Linus noticed that the <linux/rcu_segcblist.h> has huge inline functions which should not be inline at all. As a first step in cleaning this up, move them all to kernel/rcu/ and only keep an absolute minimum of data type defines in the header: before: -rw-r--r-- 1 mingo mingo 22284 May 2 10:25 include/linux/rcu_segcblist.h after: -rw-r--r-- 1 mingo mingo 3180 May 2 10:22 include/linux/rcu_segcblist.h More can be done, such as uninlining the large functions, which inlining is unjustified even if it's an RCU internal matter. Reported-by: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-02 16:31:18 +08:00
#include "rcu_segcblist.h"
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
srcu: Print non-default exp_holdoff values at boot time This commit makes srcu_bootup_announce() check for non-default values of the auto-expedite holdoff time exp_holdoff and print a message if so. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-29 02:24:22 +08:00
/* Holdoff in nanoseconds for auto-expediting. */
#define DEFAULT_SRCU_EXP_HOLDOFF (25 * 1000)
static ulong exp_holdoff = DEFAULT_SRCU_EXP_HOLDOFF;
srcu: Specify auto-expedite holdoff time On small systems, in the absence of readers, expedited SRCU grace periods can complete in less than a microsecond. This means that an eight-CPU system can have all CPUs doing synchronize_srcu() in a tight loop and almost always expedite. This might actually be desirable in some situations, but in general it is a good way to needlessly burn CPU cycles. And in those situations where it is desirable, your friend is the function synchronize_srcu_expedited(). For other situations, this commit adds a kernel parameter that specifies a holdoff between completing the last SRCU grace period and auto-expediting the next. If the next grace period starts before the holdoff expires, auto-expediting is disabled. The holdoff is 50 microseconds by default, and can be tuned to the desired number of nanoseconds. A value of zero disables auto-expediting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 05:03:11 +08:00
module_param(exp_holdoff, ulong, 0444);
srcu: Prevent sdp->srcu_gp_seq_needed counter wrap If a given CPU never happens to ever start an SRCU grace period, the grace-period sequence counter might wrap. If this CPU were to decide to finally start a grace period, the state of its sdp->srcu_gp_seq_needed might make it appear that it has already requested this grace period, which would prevent starting the grace period. If no other CPU ever started a grace period again, this would look like a grace-period hang. Even if some other CPU took pity and started the needed grace period, the leaf rcu_node structure's ->srcu_data_have_cbs field won't have record of the fact that this CPU has a callback pending, which would look like a very localized grace-period hang. This might seem very unlikely, but SRCU grace periods can take less than a microsecond on small systems, which means that overflow can happen in much less than an hour on a 32-bit embedded system. And embedded systems are especially likely to have long-term idle CPUs. Therefore, it makes sense to prevent this scenario from happening. This commit therefore scans each srcu_data structure occasionally, with frequency controlled by the srcutree.counter_wrap_check kernel boot parameter. This parameter can be set to something like 255 in order to exercise the counter-wrap-prevention code. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-04 06:35:32 +08:00
/* Overflow-check frequency. N bits roughly says every 2**N grace periods. */
static ulong counter_wrap_check = (ULONG_MAX >> 2);
module_param(counter_wrap_check, ulong, 0444);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
static void srcu_invoke_callbacks(struct work_struct *work);
static void srcu_reschedule(struct srcu_struct *sp, unsigned long delay);
/*
* Initialize SRCU combining tree. Note that statically allocated
* srcu_struct structures might already have srcu_read_lock() and
* srcu_read_unlock() running against them. So if the is_static parameter
* is set, don't initialize ->srcu_lock_count[] and ->srcu_unlock_count[].
*/
static void init_srcu_struct_nodes(struct srcu_struct *sp, bool is_static)
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
{
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
int cpu;
int i;
int level = 0;
int levelspread[RCU_NUM_LVLS];
struct srcu_data *sdp;
struct srcu_node *snp;
struct srcu_node *snp_first;
/* Work out the overall tree geometry. */
sp->level[0] = &sp->node[0];
for (i = 1; i < rcu_num_lvls; i++)
sp->level[i] = sp->level[i - 1] + num_rcu_lvl[i - 1];
rcu_init_levelspread(levelspread, num_rcu_lvl);
/* Each pass through this loop initializes one srcu_node structure. */
rcu_for_each_node_breadth_first(sp, snp) {
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_init(&ACCESS_PRIVATE(snp, lock));
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
WARN_ON_ONCE(ARRAY_SIZE(snp->srcu_have_cbs) !=
ARRAY_SIZE(snp->srcu_data_have_cbs));
for (i = 0; i < ARRAY_SIZE(snp->srcu_have_cbs); i++) {
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
snp->srcu_have_cbs[i] = 0;
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
snp->srcu_data_have_cbs[i] = 0;
}
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
snp->srcu_gp_seq_needed_exp = 0;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
snp->grplo = -1;
snp->grphi = -1;
if (snp == &sp->node[0]) {
/* Root node, special case. */
snp->srcu_parent = NULL;
continue;
}
/* Non-root node. */
if (snp == sp->level[level + 1])
level++;
snp->srcu_parent = sp->level[level - 1] +
(snp - sp->level[level]) /
levelspread[level - 1];
}
/*
* Initialize the per-CPU srcu_data array, which feeds into the
* leaves of the srcu_node tree.
*/
WARN_ON_ONCE(ARRAY_SIZE(sdp->srcu_lock_count) !=
ARRAY_SIZE(sdp->srcu_unlock_count));
level = rcu_num_lvls - 1;
snp_first = sp->level[level];
for_each_possible_cpu(cpu) {
sdp = per_cpu_ptr(sp->sda, cpu);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_init(&ACCESS_PRIVATE(sdp, lock));
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
rcu_segcblist_init(&sdp->srcu_cblist);
sdp->srcu_cblist_invoking = false;
sdp->srcu_gp_seq_needed = sp->srcu_gp_seq;
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
sdp->srcu_gp_seq_needed_exp = sp->srcu_gp_seq;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
sdp->mynode = &snp_first[cpu / levelspread[level]];
for (snp = sdp->mynode; snp != NULL; snp = snp->srcu_parent) {
if (snp->grplo < 0)
snp->grplo = cpu;
snp->grphi = cpu;
}
sdp->cpu = cpu;
INIT_DELAYED_WORK(&sdp->work, srcu_invoke_callbacks);
sdp->sp = sp;
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
sdp->grpmask = 1 << (cpu - sdp->mynode->grplo);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (is_static)
continue;
/* Dynamically allocated, better be no srcu_read_locks()! */
for (i = 0; i < ARRAY_SIZE(sdp->srcu_lock_count); i++) {
sdp->srcu_lock_count[i] = 0;
sdp->srcu_unlock_count[i] = 0;
}
}
}
/*
* Initialize non-compile-time initialized fields, including the
* associated srcu_node and srcu_data structures. The is_static
* parameter is passed through to init_srcu_struct_nodes(), and
* also tells us that ->sda has already been wired up to srcu_data.
*/
static int init_srcu_struct_fields(struct srcu_struct *sp, bool is_static)
{
mutex_init(&sp->srcu_cb_mutex);
mutex_init(&sp->srcu_gp_mutex);
sp->srcu_idx = 0;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
sp->srcu_gp_seq = 0;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
sp->srcu_barrier_seq = 0;
mutex_init(&sp->srcu_barrier_mutex);
atomic_set(&sp->srcu_barrier_cpu_cnt, 0);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
INIT_DELAYED_WORK(&sp->work, process_srcu);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (!is_static)
sp->sda = alloc_percpu(struct srcu_data);
init_srcu_struct_nodes(sp, is_static);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
sp->srcu_gp_seq_needed_exp = 0;
srcu: Specify auto-expedite holdoff time On small systems, in the absence of readers, expedited SRCU grace periods can complete in less than a microsecond. This means that an eight-CPU system can have all CPUs doing synchronize_srcu() in a tight loop and almost always expedite. This might actually be desirable in some situations, but in general it is a good way to needlessly burn CPU cycles. And in those situations where it is desirable, your friend is the function synchronize_srcu_expedited(). For other situations, this commit adds a kernel parameter that specifies a holdoff between completing the last SRCU grace period and auto-expediting the next. If the next grace period starts before the holdoff expires, auto-expediting is disabled. The holdoff is 50 microseconds by default, and can be tuned to the desired number of nanoseconds. A value of zero disables auto-expediting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 05:03:11 +08:00
sp->srcu_last_gp_end = ktime_get_mono_fast_ns();
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
smp_store_release(&sp->srcu_gp_seq_needed, 0); /* Init done. */
return sp->sda ? 0 : -ENOMEM;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
#ifdef CONFIG_DEBUG_LOCK_ALLOC
int __init_srcu_struct(struct srcu_struct *sp, const char *name,
struct lock_class_key *key)
{
/* Don't re-initialize a lock while it is held. */
debug_check_no_locks_freed((void *)sp, sizeof(*sp));
lockdep_init_map(&sp->dep_map, name, key, 0);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_init(&ACCESS_PRIVATE(sp, lock));
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
return init_srcu_struct_fields(sp, false);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(__init_srcu_struct);
#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
/**
* init_srcu_struct - initialize a sleep-RCU structure
* @sp: structure to initialize.
*
* Must invoke this on a given srcu_struct before passing that srcu_struct
* to any other function. Each srcu_struct represents a separate domain
* of SRCU protection.
*/
int init_srcu_struct(struct srcu_struct *sp)
{
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_init(&ACCESS_PRIVATE(sp, lock));
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
return init_srcu_struct_fields(sp, false);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(init_srcu_struct);
#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
/*
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* First-use initialization of statically allocated srcu_struct
* structure. Wiring up the combining tree is more than can be
* done with compile-time initialization, so this check is added
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
* to each update-side SRCU primitive. Use sp->lock, which -is-
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* compile-time initialized, to resolve races involving multiple
* CPUs trying to garner first-use privileges.
*/
static void check_init_srcu_struct(struct srcu_struct *sp)
{
unsigned long flags;
WARN_ON_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INIT);
/* The smp_load_acquire() pairs with the smp_store_release(). */
if (!rcu_seq_state(smp_load_acquire(&sp->srcu_gp_seq_needed))) /*^^^*/
return; /* Already initialized. */
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irqsave_rcu_node(sp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (!rcu_seq_state(sp->srcu_gp_seq_needed)) {
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(sp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
return;
}
init_srcu_struct_fields(sp, true);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(sp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
/*
* Returns approximate total of the readers' ->srcu_lock_count[] values
* for the rank of per-CPU counters specified by idx.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
*/
static unsigned long srcu_readers_lock_idx(struct srcu_struct *sp, int idx)
{
int cpu;
unsigned long sum = 0;
for_each_possible_cpu(cpu) {
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
struct srcu_data *cpuc = per_cpu_ptr(sp->sda, cpu);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
sum += READ_ONCE(cpuc->srcu_lock_count[idx]);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
return sum;
}
/*
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* Returns approximate total of the readers' ->srcu_unlock_count[] values
* for the rank of per-CPU counters specified by idx.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
*/
static unsigned long srcu_readers_unlock_idx(struct srcu_struct *sp, int idx)
{
int cpu;
unsigned long sum = 0;
for_each_possible_cpu(cpu) {
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
struct srcu_data *cpuc = per_cpu_ptr(sp->sda, cpu);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
sum += READ_ONCE(cpuc->srcu_unlock_count[idx]);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
return sum;
}
/*
* Return true if the number of pre-existing readers is determined to
* be zero.
*/
static bool srcu_readers_active_idx_check(struct srcu_struct *sp, int idx)
{
unsigned long unlocks;
unlocks = srcu_readers_unlock_idx(sp, idx);
/*
* Make sure that a lock is always counted if the corresponding
* unlock is counted. Needs to be a smp_mb() as the read side may
* contain a read from a variable that is written to before the
* synchronize_srcu() in the write side. In this case smp_mb()s
* A and B act like the store buffering pattern.
*
* This smp_mb() also pairs with smp_mb() C to prevent accesses
* after the synchronize_srcu() from being executed before the
* grace period ends.
*/
smp_mb(); /* A */
/*
* If the locks are the same as the unlocks, then there must have
* been no readers on this index at some time in between. This does
* not mean that there are no more readers, as one could have read
* the current index but not have incremented the lock counter yet.
*
srcu: Eliminate possibility of destructive counter overflow Earlier versions of Tree SRCU were subject to a counter overflow bug that could theoretically result in too-short grace periods. This commit eliminates this problem by adding an update-side memory barrier. The short explanation is that if the updater sums the unlock counts too late to see a given __srcu_read_unlock() increment, that CPU's next __srcu_read_lock() must see the new value of ->srcu_idx, thus incrementing the other bank of counters. This eliminates the possibility of destructive counter overflow as long as the srcu_read_lock() nesting level does not exceed floor(ULONG_MAX/NR_CPUS/2), which should be an eminently reasonable nesting limit, especially on 64-bit systems. Reported-by: Lance Roy <ldr709@gmail.com> Suggested-by: Lance Roy <ldr709@gmail.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-13 06:16:50 +08:00
* So suppose that the updater is preempted here for so long
* that more than ULONG_MAX non-nested readers come and go in
* the meantime. It turns out that this cannot result in overflow
* because if a reader modifies its unlock count after we read it
* above, then that reader's next load of ->srcu_idx is guaranteed
* to get the new value, which will cause it to operate on the
* other bank of counters, where it cannot contribute to the
* overflow of these counters. This means that there is a maximum
* of 2*NR_CPUS increments, which cannot overflow given current
* systems, especially not on 64-bit systems.
*
* OK, how about nesting? This does impose a limit on nesting
* of floor(ULONG_MAX/NR_CPUS/2), which should be sufficient,
* especially on 64-bit systems.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
*/
return srcu_readers_lock_idx(sp, idx) == unlocks;
}
/**
* srcu_readers_active - returns true if there are readers. and false
* otherwise
* @sp: which srcu_struct to count active readers (holding srcu_read_lock).
*
* Note that this is not an atomic primitive, and can therefore suffer
* severe errors when invoked on an active srcu_struct. That said, it
* can be useful as an error check at cleanup time.
*/
static bool srcu_readers_active(struct srcu_struct *sp)
{
int cpu;
unsigned long sum = 0;
for_each_possible_cpu(cpu) {
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
struct srcu_data *cpuc = per_cpu_ptr(sp->sda, cpu);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
sum += READ_ONCE(cpuc->srcu_lock_count[0]);
sum += READ_ONCE(cpuc->srcu_lock_count[1]);
sum -= READ_ONCE(cpuc->srcu_unlock_count[0]);
sum -= READ_ONCE(cpuc->srcu_unlock_count[1]);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
return sum;
}
#define SRCU_INTERVAL 1
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
/*
* Return grace-period delay, zero if there are expedited grace
* periods pending, SRCU_INTERVAL otherwise.
*/
static unsigned long srcu_get_delay(struct srcu_struct *sp)
{
if (ULONG_CMP_LT(READ_ONCE(sp->srcu_gp_seq),
READ_ONCE(sp->srcu_gp_seq_needed_exp)))
return 0;
return SRCU_INTERVAL;
}
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
/**
* cleanup_srcu_struct - deconstruct a sleep-RCU structure
* @sp: structure to clean up.
*
* Must invoke this after you are finished using a given srcu_struct that
* was initialized via init_srcu_struct(), else you leak memory.
*/
void cleanup_srcu_struct(struct srcu_struct *sp)
{
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
int cpu;
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (WARN_ON(!srcu_get_delay(sp)))
return; /* Leakage unless caller handles error. */
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
if (WARN_ON(srcu_readers_active(sp)))
return; /* Leakage unless caller handles error. */
flush_delayed_work(&sp->work);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
for_each_possible_cpu(cpu)
flush_delayed_work(&per_cpu_ptr(sp->sda, cpu)->work);
if (WARN_ON(rcu_seq_state(READ_ONCE(sp->srcu_gp_seq)) != SRCU_STATE_IDLE) ||
WARN_ON(srcu_readers_active(sp))) {
pr_info("cleanup_srcu_struct: Active srcu_struct %p state: %d\n", sp, rcu_seq_state(READ_ONCE(sp->srcu_gp_seq)));
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
return; /* Caller forgot to stop doing call_srcu()? */
}
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
free_percpu(sp->sda);
sp->sda = NULL;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(cleanup_srcu_struct);
/*
* Counts the new reader in the appropriate per-CPU element of the
srcu: Allow use of Tiny/Tree SRCU from both process and interrupt context Linu Cherian reported a WARN in cleanup_srcu_struct() when shutting down a guest running iperf on a VFIO assigned device. This happens because irqfd_wakeup() calls srcu_read_lock(&kvm->irq_srcu) in interrupt context, while a worker thread does the same inside kvm_set_irq(). If the interrupt happens while the worker thread is executing __srcu_read_lock(), updates to the Classic SRCU ->lock_count[] field or the Tree SRCU ->srcu_lock_count[] field can be lost. The docs say you are not supposed to call srcu_read_lock() and srcu_read_unlock() from irq context, but KVM interrupt injection happens from (host) interrupt context and it would be nice if SRCU supported the use case. KVM is using SRCU here not really for the "sleepable" part, but rather due to its IPI-free fast detection of grace periods. It is therefore not desirable to switch back to RCU, which would effectively revert commit 719d93cd5f5c ("kvm/irqchip: Speed up KVM_SET_GSI_ROUTING", 2014-01-16). However, the docs are overly conservative. You can have an SRCU instance only has users in irq context, and you can mix process and irq context as long as process context users disable interrupts. In addition, __srcu_read_unlock() actually uses this_cpu_dec() on both Tree SRCU and Classic SRCU. For those two implementations, only srcu_read_lock() is unsafe. When Classic SRCU's __srcu_read_unlock() was changed to use this_cpu_dec(), in commit 5a41344a3d83 ("srcu: Simplify __srcu_read_unlock() via this_cpu_dec()", 2012-11-29), __srcu_read_lock() did two increments. Therefore it kept __this_cpu_inc(), with preempt_disable/enable in the caller. Tree SRCU however only does one increment, so on most architectures it is more efficient for __srcu_read_lock() to use this_cpu_inc(), and any performance differences appear to be down in the noise. Unlike Classic and Tree SRCU, Tiny SRCU does increments and decrements on a single variable. Therefore, as Peter Zijlstra pointed out, Tiny SRCU's implementation already supports mixed-context use of srcu_read_lock() and srcu_read_unlock(), at least as long as uses of srcu_read_lock() and srcu_read_unlock() in each handler are nested and paired properly. In other words, it is still illegal to (say) invoke srcu_read_lock() in an interrupt handler and to invoke the matching srcu_read_unlock() in a softirq handler. Therefore, the only change required for Tiny SRCU is to its comments. Fixes: 719d93cd5f5c ("kvm/irqchip: Speed up KVM_SET_GSI_ROUTING") Reported-by: Linu Cherian <linuc.decode@gmail.com> Suggested-by: Linu Cherian <linuc.decode@gmail.com> Cc: kvm@vger.kernel.org Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Paolo Bonzini <pbonzini@redhat.com>
2017-05-31 20:03:10 +08:00
* srcu_struct.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
* Returns an index that must be passed to the matching srcu_read_unlock().
*/
int __srcu_read_lock(struct srcu_struct *sp)
{
int idx;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
idx = READ_ONCE(sp->srcu_idx) & 0x1;
srcu: Allow use of Tiny/Tree SRCU from both process and interrupt context Linu Cherian reported a WARN in cleanup_srcu_struct() when shutting down a guest running iperf on a VFIO assigned device. This happens because irqfd_wakeup() calls srcu_read_lock(&kvm->irq_srcu) in interrupt context, while a worker thread does the same inside kvm_set_irq(). If the interrupt happens while the worker thread is executing __srcu_read_lock(), updates to the Classic SRCU ->lock_count[] field or the Tree SRCU ->srcu_lock_count[] field can be lost. The docs say you are not supposed to call srcu_read_lock() and srcu_read_unlock() from irq context, but KVM interrupt injection happens from (host) interrupt context and it would be nice if SRCU supported the use case. KVM is using SRCU here not really for the "sleepable" part, but rather due to its IPI-free fast detection of grace periods. It is therefore not desirable to switch back to RCU, which would effectively revert commit 719d93cd5f5c ("kvm/irqchip: Speed up KVM_SET_GSI_ROUTING", 2014-01-16). However, the docs are overly conservative. You can have an SRCU instance only has users in irq context, and you can mix process and irq context as long as process context users disable interrupts. In addition, __srcu_read_unlock() actually uses this_cpu_dec() on both Tree SRCU and Classic SRCU. For those two implementations, only srcu_read_lock() is unsafe. When Classic SRCU's __srcu_read_unlock() was changed to use this_cpu_dec(), in commit 5a41344a3d83 ("srcu: Simplify __srcu_read_unlock() via this_cpu_dec()", 2012-11-29), __srcu_read_lock() did two increments. Therefore it kept __this_cpu_inc(), with preempt_disable/enable in the caller. Tree SRCU however only does one increment, so on most architectures it is more efficient for __srcu_read_lock() to use this_cpu_inc(), and any performance differences appear to be down in the noise. Unlike Classic and Tree SRCU, Tiny SRCU does increments and decrements on a single variable. Therefore, as Peter Zijlstra pointed out, Tiny SRCU's implementation already supports mixed-context use of srcu_read_lock() and srcu_read_unlock(), at least as long as uses of srcu_read_lock() and srcu_read_unlock() in each handler are nested and paired properly. In other words, it is still illegal to (say) invoke srcu_read_lock() in an interrupt handler and to invoke the matching srcu_read_unlock() in a softirq handler. Therefore, the only change required for Tiny SRCU is to its comments. Fixes: 719d93cd5f5c ("kvm/irqchip: Speed up KVM_SET_GSI_ROUTING") Reported-by: Linu Cherian <linuc.decode@gmail.com> Suggested-by: Linu Cherian <linuc.decode@gmail.com> Cc: kvm@vger.kernel.org Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Paolo Bonzini <pbonzini@redhat.com>
2017-05-31 20:03:10 +08:00
this_cpu_inc(sp->sda->srcu_lock_count[idx]);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
smp_mb(); /* B */ /* Avoid leaking the critical section. */
return idx;
}
EXPORT_SYMBOL_GPL(__srcu_read_lock);
/*
* Removes the count for the old reader from the appropriate per-CPU
* element of the srcu_struct. Note that this may well be a different
* CPU than that which was incremented by the corresponding srcu_read_lock().
*/
void __srcu_read_unlock(struct srcu_struct *sp, int idx)
{
smp_mb(); /* C */ /* Avoid leaking the critical section. */
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
this_cpu_inc(sp->sda->srcu_unlock_count[idx]);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(__srcu_read_unlock);
/*
* We use an adaptive strategy for synchronize_srcu() and especially for
* synchronize_srcu_expedited(). We spin for a fixed time period
* (defined below) to allow SRCU readers to exit their read-side critical
* sections. If there are still some readers after a few microseconds,
* we repeatedly block for 1-millisecond time periods.
*/
#define SRCU_RETRY_CHECK_DELAY 5
/*
* Start an SRCU grace period.
*/
static void srcu_gp_start(struct srcu_struct *sp)
{
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
struct srcu_data *sdp = this_cpu_ptr(sp->sda);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
int state;
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
lockdep_assert_held(&sp->lock);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
WARN_ON_ONCE(ULONG_CMP_GE(sp->srcu_gp_seq, sp->srcu_gp_seq_needed));
rcu_segcblist_advance(&sdp->srcu_cblist,
rcu_seq_current(&sp->srcu_gp_seq));
(void)rcu_segcblist_accelerate(&sdp->srcu_cblist,
rcu_seq_snap(&sp->srcu_gp_seq));
srcu: Expedite first synchronize_srcu() when idle Classic SRCU in effect expedites the first synchronize_srcu() when SRCU is idle, and Mike Galbraith demonstrated that some use cases do in fact rely on this behavior. In particular, Mike showed that Steven Rostedt's hotplug stress script takes 55 seconds with Classic SRCU and more than 16 -minutes- when running Tree SRCU. Assuming that each Tree SRCU's call to synchronize_srcu() takes four milliseconds, this implies that Steven's test invokes synchronize_srcu() in isolation, but more than once per 200 microseconds. Mike used ftrace to demonstrate that the time between successive calls to synchronize_srcu() ranged from 118 to 342 microseconds, with one outlier at 80 milliseconds. This data clearly indicates that Tree SRCU needs to expedite the first invocation of synchronize_srcu() during an SRCU idle period. This commit therefor introduces a srcu_might_be_idle() function that probabilistically checks whether or not SRCU is idle. This function is used by synchronize_rcu() as an additional criterion in deciding whether or not to expedite. (Hat trick to Peter Zijlstra for his earlier suggestion that this might in fact be a problem. Which for all I know might have motivated Mike to look into it.) Reported-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 02:34:40 +08:00
smp_mb(); /* Order prior store to ->srcu_gp_seq_needed vs. GP start. */
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
rcu_seq_start(&sp->srcu_gp_seq);
state = rcu_seq_state(READ_ONCE(sp->srcu_gp_seq));
WARN_ON_ONCE(state != SRCU_STATE_SCAN1);
}
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
/*
* Track online CPUs to guide callback workqueue placement.
*/
DEFINE_PER_CPU(bool, srcu_online);
void srcu_online_cpu(unsigned int cpu)
{
WRITE_ONCE(per_cpu(srcu_online, cpu), true);
}
void srcu_offline_cpu(unsigned int cpu)
{
WRITE_ONCE(per_cpu(srcu_online, cpu), false);
}
/*
* Place the workqueue handler on the specified CPU if online, otherwise
* just run it whereever. This is useful for placing workqueue handlers
* that are to invoke the specified CPU's callbacks.
*/
static bool srcu_queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
struct delayed_work *dwork,
unsigned long delay)
{
bool ret;
preempt_disable();
if (READ_ONCE(per_cpu(srcu_online, cpu)))
ret = queue_delayed_work_on(cpu, wq, dwork, delay);
else
ret = queue_delayed_work(wq, dwork, delay);
preempt_enable();
return ret;
}
/*
* Schedule callback invocation for the specified srcu_data structure,
* if possible, on the corresponding CPU.
*/
static void srcu_schedule_cbs_sdp(struct srcu_data *sdp, unsigned long delay)
{
srcu_queue_delayed_work_on(sdp->cpu, system_power_efficient_wq,
&sdp->work, delay);
}
/*
* Schedule callback invocation for all srcu_data structures associated
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
* with the specified srcu_node structure that have callbacks for the
* just-completed grace period, the one corresponding to idx. If possible,
* schedule this invocation on the corresponding CPUs.
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
*/
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
static void srcu_schedule_cbs_snp(struct srcu_struct *sp, struct srcu_node *snp,
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
unsigned long mask, unsigned long delay)
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
{
int cpu;
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
for (cpu = snp->grplo; cpu <= snp->grphi; cpu++) {
if (!(mask & (1 << (cpu - snp->grplo))))
continue;
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
srcu_schedule_cbs_sdp(per_cpu_ptr(sp->sda, cpu), delay);
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
}
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
/*
* Note the end of an SRCU grace period. Initiates callback invocation
* and starts a new grace period if needed.
*
* The ->srcu_cb_mutex acquisition does not protect any data, but
* instead prevents more than one grace period from starting while we
* are initiating callback invocation. This allows the ->srcu_have_cbs[]
* array to have a finite number of elements.
*/
static void srcu_gp_end(struct srcu_struct *sp)
{
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
unsigned long cbdelay;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
bool cbs;
srcu: Prevent sdp->srcu_gp_seq_needed counter wrap If a given CPU never happens to ever start an SRCU grace period, the grace-period sequence counter might wrap. If this CPU were to decide to finally start a grace period, the state of its sdp->srcu_gp_seq_needed might make it appear that it has already requested this grace period, which would prevent starting the grace period. If no other CPU ever started a grace period again, this would look like a grace-period hang. Even if some other CPU took pity and started the needed grace period, the leaf rcu_node structure's ->srcu_data_have_cbs field won't have record of the fact that this CPU has a callback pending, which would look like a very localized grace-period hang. This might seem very unlikely, but SRCU grace periods can take less than a microsecond on small systems, which means that overflow can happen in much less than an hour on a 32-bit embedded system. And embedded systems are especially likely to have long-term idle CPUs. Therefore, it makes sense to prevent this scenario from happening. This commit therefore scans each srcu_data structure occasionally, with frequency controlled by the srcutree.counter_wrap_check kernel boot parameter. This parameter can be set to something like 255 in order to exercise the counter-wrap-prevention code. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-04 06:35:32 +08:00
int cpu;
unsigned long flags;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
unsigned long gpseq;
int idx;
int idxnext;
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
unsigned long mask;
srcu: Prevent sdp->srcu_gp_seq_needed counter wrap If a given CPU never happens to ever start an SRCU grace period, the grace-period sequence counter might wrap. If this CPU were to decide to finally start a grace period, the state of its sdp->srcu_gp_seq_needed might make it appear that it has already requested this grace period, which would prevent starting the grace period. If no other CPU ever started a grace period again, this would look like a grace-period hang. Even if some other CPU took pity and started the needed grace period, the leaf rcu_node structure's ->srcu_data_have_cbs field won't have record of the fact that this CPU has a callback pending, which would look like a very localized grace-period hang. This might seem very unlikely, but SRCU grace periods can take less than a microsecond on small systems, which means that overflow can happen in much less than an hour on a 32-bit embedded system. And embedded systems are especially likely to have long-term idle CPUs. Therefore, it makes sense to prevent this scenario from happening. This commit therefore scans each srcu_data structure occasionally, with frequency controlled by the srcutree.counter_wrap_check kernel boot parameter. This parameter can be set to something like 255 in order to exercise the counter-wrap-prevention code. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-04 06:35:32 +08:00
struct srcu_data *sdp;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
struct srcu_node *snp;
/* Prevent more than one additional grace period. */
mutex_lock(&sp->srcu_cb_mutex);
/* End the current grace period. */
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
idx = rcu_seq_state(sp->srcu_gp_seq);
WARN_ON_ONCE(idx != SRCU_STATE_SCAN2);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
cbdelay = srcu_get_delay(sp);
srcu: Specify auto-expedite holdoff time On small systems, in the absence of readers, expedited SRCU grace periods can complete in less than a microsecond. This means that an eight-CPU system can have all CPUs doing synchronize_srcu() in a tight loop and almost always expedite. This might actually be desirable in some situations, but in general it is a good way to needlessly burn CPU cycles. And in those situations where it is desirable, your friend is the function synchronize_srcu_expedited(). For other situations, this commit adds a kernel parameter that specifies a holdoff between completing the last SRCU grace period and auto-expediting the next. If the next grace period starts before the holdoff expires, auto-expediting is disabled. The holdoff is 50 microseconds by default, and can be tuned to the desired number of nanoseconds. A value of zero disables auto-expediting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 05:03:11 +08:00
sp->srcu_last_gp_end = ktime_get_mono_fast_ns();
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
rcu_seq_end(&sp->srcu_gp_seq);
gpseq = rcu_seq_current(&sp->srcu_gp_seq);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (ULONG_CMP_LT(sp->srcu_gp_seq_needed_exp, gpseq))
sp->srcu_gp_seq_needed_exp = gpseq;
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
mutex_unlock(&sp->srcu_gp_mutex);
/* A new grace period can start at this point. But only one. */
/* Initiate callback invocation as needed. */
idx = rcu_seq_ctr(gpseq) % ARRAY_SIZE(snp->srcu_have_cbs);
idxnext = (idx + 1) % ARRAY_SIZE(snp->srcu_have_cbs);
rcu_for_each_node_breadth_first(sp, snp) {
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irq_rcu_node(snp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
cbs = false;
if (snp >= sp->level[rcu_num_lvls - 1])
cbs = snp->srcu_have_cbs[idx] == gpseq;
snp->srcu_have_cbs[idx] = gpseq;
rcu_seq_set_state(&snp->srcu_have_cbs[idx], 1);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (ULONG_CMP_LT(snp->srcu_gp_seq_needed_exp, gpseq))
snp->srcu_gp_seq_needed_exp = gpseq;
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
mask = snp->srcu_data_have_cbs[idx];
snp->srcu_data_have_cbs[idx] = 0;
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(snp);
if (cbs)
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
srcu_schedule_cbs_snp(sp, snp, mask, cbdelay);
srcu: Prevent sdp->srcu_gp_seq_needed counter wrap If a given CPU never happens to ever start an SRCU grace period, the grace-period sequence counter might wrap. If this CPU were to decide to finally start a grace period, the state of its sdp->srcu_gp_seq_needed might make it appear that it has already requested this grace period, which would prevent starting the grace period. If no other CPU ever started a grace period again, this would look like a grace-period hang. Even if some other CPU took pity and started the needed grace period, the leaf rcu_node structure's ->srcu_data_have_cbs field won't have record of the fact that this CPU has a callback pending, which would look like a very localized grace-period hang. This might seem very unlikely, but SRCU grace periods can take less than a microsecond on small systems, which means that overflow can happen in much less than an hour on a 32-bit embedded system. And embedded systems are especially likely to have long-term idle CPUs. Therefore, it makes sense to prevent this scenario from happening. This commit therefore scans each srcu_data structure occasionally, with frequency controlled by the srcutree.counter_wrap_check kernel boot parameter. This parameter can be set to something like 255 in order to exercise the counter-wrap-prevention code. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-04 06:35:32 +08:00
/* Occasionally prevent srcu_data counter wrap. */
if (!(gpseq & counter_wrap_check))
for (cpu = snp->grplo; cpu <= snp->grphi; cpu++) {
sdp = per_cpu_ptr(sp->sda, cpu);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irqsave_rcu_node(sdp, flags);
srcu: Prevent sdp->srcu_gp_seq_needed counter wrap If a given CPU never happens to ever start an SRCU grace period, the grace-period sequence counter might wrap. If this CPU were to decide to finally start a grace period, the state of its sdp->srcu_gp_seq_needed might make it appear that it has already requested this grace period, which would prevent starting the grace period. If no other CPU ever started a grace period again, this would look like a grace-period hang. Even if some other CPU took pity and started the needed grace period, the leaf rcu_node structure's ->srcu_data_have_cbs field won't have record of the fact that this CPU has a callback pending, which would look like a very localized grace-period hang. This might seem very unlikely, but SRCU grace periods can take less than a microsecond on small systems, which means that overflow can happen in much less than an hour on a 32-bit embedded system. And embedded systems are especially likely to have long-term idle CPUs. Therefore, it makes sense to prevent this scenario from happening. This commit therefore scans each srcu_data structure occasionally, with frequency controlled by the srcutree.counter_wrap_check kernel boot parameter. This parameter can be set to something like 255 in order to exercise the counter-wrap-prevention code. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-04 06:35:32 +08:00
if (ULONG_CMP_GE(gpseq,
sdp->srcu_gp_seq_needed + 100))
sdp->srcu_gp_seq_needed = gpseq;
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(sdp, flags);
srcu: Prevent sdp->srcu_gp_seq_needed counter wrap If a given CPU never happens to ever start an SRCU grace period, the grace-period sequence counter might wrap. If this CPU were to decide to finally start a grace period, the state of its sdp->srcu_gp_seq_needed might make it appear that it has already requested this grace period, which would prevent starting the grace period. If no other CPU ever started a grace period again, this would look like a grace-period hang. Even if some other CPU took pity and started the needed grace period, the leaf rcu_node structure's ->srcu_data_have_cbs field won't have record of the fact that this CPU has a callback pending, which would look like a very localized grace-period hang. This might seem very unlikely, but SRCU grace periods can take less than a microsecond on small systems, which means that overflow can happen in much less than an hour on a 32-bit embedded system. And embedded systems are especially likely to have long-term idle CPUs. Therefore, it makes sense to prevent this scenario from happening. This commit therefore scans each srcu_data structure occasionally, with frequency controlled by the srcutree.counter_wrap_check kernel boot parameter. This parameter can be set to something like 255 in order to exercise the counter-wrap-prevention code. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-04 06:35:32 +08:00
}
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
/* Callback initiation done, allow grace periods after next. */
mutex_unlock(&sp->srcu_cb_mutex);
/* Start a new grace period if needed. */
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
gpseq = rcu_seq_current(&sp->srcu_gp_seq);
if (!rcu_seq_state(gpseq) &&
ULONG_CMP_LT(gpseq, sp->srcu_gp_seq_needed)) {
srcu_gp_start(sp);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
/* Throttle expedited grace periods: Should be rare! */
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
srcu_reschedule(sp, rcu_seq_ctr(gpseq) & 0x3ff
? 0 : SRCU_INTERVAL);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
} else {
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
}
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
/*
* Funnel-locking scheme to scalably mediate many concurrent expedited
* grace-period requests. This function is invoked for the first known
* expedited request for a grace period that has already been requested,
* but without expediting. To start a completely new grace period,
* whether expedited or not, use srcu_funnel_gp_start() instead.
*/
static void srcu_funnel_exp_start(struct srcu_struct *sp, struct srcu_node *snp,
unsigned long s)
{
unsigned long flags;
for (; snp != NULL; snp = snp->srcu_parent) {
if (rcu_seq_done(&sp->srcu_gp_seq, s) ||
ULONG_CMP_GE(READ_ONCE(snp->srcu_gp_seq_needed_exp), s))
return;
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irqsave_rcu_node(snp, flags);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (ULONG_CMP_GE(snp->srcu_gp_seq_needed_exp, s)) {
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(snp, flags);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
return;
}
WRITE_ONCE(snp->srcu_gp_seq_needed_exp, s);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(snp, flags);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
}
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irqsave_rcu_node(sp, flags);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (!ULONG_CMP_LT(sp->srcu_gp_seq_needed_exp, s))
sp->srcu_gp_seq_needed_exp = s;
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(sp, flags);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
}
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
/*
* Funnel-locking scheme to scalably mediate many concurrent grace-period
* requests. The winner has to do the work of actually starting grace
* period s. Losers must either ensure that their desired grace-period
* number is recorded on at least their leaf srcu_node structure, or they
* must take steps to invoke their own callbacks.
*/
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
static void srcu_funnel_gp_start(struct srcu_struct *sp, struct srcu_data *sdp,
unsigned long s, bool do_norm)
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
{
unsigned long flags;
int idx = rcu_seq_ctr(s) % ARRAY_SIZE(sdp->mynode->srcu_have_cbs);
struct srcu_node *snp = sdp->mynode;
unsigned long snp_seq;
/* Each pass through the loop does one level of the srcu_node tree. */
for (; snp != NULL; snp = snp->srcu_parent) {
if (rcu_seq_done(&sp->srcu_gp_seq, s) && snp != sdp->mynode)
return; /* GP already done and CBs recorded. */
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irqsave_rcu_node(snp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (ULONG_CMP_GE(snp->srcu_have_cbs[idx], s)) {
snp_seq = snp->srcu_have_cbs[idx];
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
if (snp == sdp->mynode && snp_seq == s)
snp->srcu_data_have_cbs[idx] |= sdp->grpmask;
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(snp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (snp == sdp->mynode && snp_seq != s) {
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
srcu_schedule_cbs_sdp(sdp, do_norm
? SRCU_INTERVAL
: 0);
return;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (!do_norm)
srcu_funnel_exp_start(sp, snp, s);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
return;
}
snp->srcu_have_cbs[idx] = s;
srcu: Exact tracking of srcu_data structures containing callbacks The current Tree SRCU implementation schedules a workqueue for every srcu_data covered by a given leaf srcu_node structure having callbacks, even if only one of those srcu_data structures actually contains callbacks. This is clearly inefficient for workloads that don't feature callbacks everywhere all the time. This commit therefore adds an array of masks that are used by the leaf srcu_node structures to track exactly which srcu_data structures contain callbacks. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 07:01:46 +08:00
if (snp == sdp->mynode)
snp->srcu_data_have_cbs[idx] |= sdp->grpmask;
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (!do_norm && ULONG_CMP_LT(snp->srcu_gp_seq_needed_exp, s))
snp->srcu_gp_seq_needed_exp = s;
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(snp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
/* Top of tree, must ensure the grace period will be started. */
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irqsave_rcu_node(sp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (ULONG_CMP_LT(sp->srcu_gp_seq_needed, s)) {
/*
* Record need for grace period s. Pair with load
* acquire setting up for initialization.
*/
smp_store_release(&sp->srcu_gp_seq_needed, s); /*^^^*/
}
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (!do_norm && ULONG_CMP_LT(sp->srcu_gp_seq_needed_exp, s))
sp->srcu_gp_seq_needed_exp = s;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
/* If grace period not already done and none in progress, start it. */
if (!rcu_seq_done(&sp->srcu_gp_seq, s) &&
rcu_seq_state(sp->srcu_gp_seq) == SRCU_STATE_IDLE) {
WARN_ON_ONCE(ULONG_CMP_GE(sp->srcu_gp_seq, sp->srcu_gp_seq_needed));
srcu_gp_start(sp);
queue_delayed_work(system_power_efficient_wq, &sp->work,
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
srcu_get_delay(sp));
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(sp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
/*
* Wait until all readers counted by array index idx complete, but
* loop an additional time if there is an expedited grace period pending.
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* The caller must ensure that ->srcu_idx is not changed while checking.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
*/
static bool try_check_zero(struct srcu_struct *sp, int idx, int trycount)
{
for (;;) {
if (srcu_readers_active_idx_check(sp, idx))
return true;
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (--trycount + !srcu_get_delay(sp) <= 0)
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
return false;
udelay(SRCU_RETRY_CHECK_DELAY);
}
}
/*
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* Increment the ->srcu_idx counter so that future SRCU readers will
* use the other rank of the ->srcu_(un)lock_count[] arrays. This allows
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
* us to wait for pre-existing readers in a starvation-free manner.
*/
static void srcu_flip(struct srcu_struct *sp)
{
srcu: Eliminate possibility of destructive counter overflow Earlier versions of Tree SRCU were subject to a counter overflow bug that could theoretically result in too-short grace periods. This commit eliminates this problem by adding an update-side memory barrier. The short explanation is that if the updater sums the unlock counts too late to see a given __srcu_read_unlock() increment, that CPU's next __srcu_read_lock() must see the new value of ->srcu_idx, thus incrementing the other bank of counters. This eliminates the possibility of destructive counter overflow as long as the srcu_read_lock() nesting level does not exceed floor(ULONG_MAX/NR_CPUS/2), which should be an eminently reasonable nesting limit, especially on 64-bit systems. Reported-by: Lance Roy <ldr709@gmail.com> Suggested-by: Lance Roy <ldr709@gmail.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-13 06:16:50 +08:00
/*
* Ensure that if this updater saw a given reader's increment
* from __srcu_read_lock(), that reader was using an old value
* of ->srcu_idx. Also ensure that if a given reader sees the
* new value of ->srcu_idx, this updater's earlier scans cannot
* have seen that reader's increments (which is OK, because this
* grace period need not wait on that reader).
*/
smp_mb(); /* E */ /* Pairs with B and C. */
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
WRITE_ONCE(sp->srcu_idx, sp->srcu_idx + 1);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
/*
* Ensure that if the updater misses an __srcu_read_unlock()
* increment, that task's next __srcu_read_lock() will see the
* above counter update. Note that both this memory barrier
* and the one in srcu_readers_active_idx_check() provide the
* guarantee for __srcu_read_lock().
*/
smp_mb(); /* D */ /* Pairs with C. */
}
srcu: Expedite first synchronize_srcu() when idle Classic SRCU in effect expedites the first synchronize_srcu() when SRCU is idle, and Mike Galbraith demonstrated that some use cases do in fact rely on this behavior. In particular, Mike showed that Steven Rostedt's hotplug stress script takes 55 seconds with Classic SRCU and more than 16 -minutes- when running Tree SRCU. Assuming that each Tree SRCU's call to synchronize_srcu() takes four milliseconds, this implies that Steven's test invokes synchronize_srcu() in isolation, but more than once per 200 microseconds. Mike used ftrace to demonstrate that the time between successive calls to synchronize_srcu() ranged from 118 to 342 microseconds, with one outlier at 80 milliseconds. This data clearly indicates that Tree SRCU needs to expedite the first invocation of synchronize_srcu() during an SRCU idle period. This commit therefor introduces a srcu_might_be_idle() function that probabilistically checks whether or not SRCU is idle. This function is used by synchronize_rcu() as an additional criterion in deciding whether or not to expedite. (Hat trick to Peter Zijlstra for his earlier suggestion that this might in fact be a problem. Which for all I know might have motivated Mike to look into it.) Reported-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 02:34:40 +08:00
/*
* If SRCU is likely idle, return true, otherwise return false.
*
* Note that it is OK for several current from-idle requests for a new
* grace period from idle to specify expediting because they will all end
* up requesting the same grace period anyhow. So no loss.
*
* Note also that if any CPU (including the current one) is still invoking
* callbacks, this function will nevertheless say "idle". This is not
* ideal, but the overhead of checking all CPUs' callback lists is even
* less ideal, especially on large systems. Furthermore, the wakeup
* can happen before the callback is fully removed, so we have no choice
* but to accept this type of error.
*
* This function is also subject to counter-wrap errors, but let's face
* it, if this function was preempted for enough time for the counters
* to wrap, it really doesn't matter whether or not we expedite the grace
* period. The extra overhead of a needlessly expedited grace period is
* negligible when amoritized over that time period, and the extra latency
* of a needlessly non-expedited grace period is similarly negligible.
*/
static bool srcu_might_be_idle(struct srcu_struct *sp)
{
srcu: Specify auto-expedite holdoff time On small systems, in the absence of readers, expedited SRCU grace periods can complete in less than a microsecond. This means that an eight-CPU system can have all CPUs doing synchronize_srcu() in a tight loop and almost always expedite. This might actually be desirable in some situations, but in general it is a good way to needlessly burn CPU cycles. And in those situations where it is desirable, your friend is the function synchronize_srcu_expedited(). For other situations, this commit adds a kernel parameter that specifies a holdoff between completing the last SRCU grace period and auto-expediting the next. If the next grace period starts before the holdoff expires, auto-expediting is disabled. The holdoff is 50 microseconds by default, and can be tuned to the desired number of nanoseconds. A value of zero disables auto-expediting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 05:03:11 +08:00
unsigned long curseq;
srcu: Expedite first synchronize_srcu() when idle Classic SRCU in effect expedites the first synchronize_srcu() when SRCU is idle, and Mike Galbraith demonstrated that some use cases do in fact rely on this behavior. In particular, Mike showed that Steven Rostedt's hotplug stress script takes 55 seconds with Classic SRCU and more than 16 -minutes- when running Tree SRCU. Assuming that each Tree SRCU's call to synchronize_srcu() takes four milliseconds, this implies that Steven's test invokes synchronize_srcu() in isolation, but more than once per 200 microseconds. Mike used ftrace to demonstrate that the time between successive calls to synchronize_srcu() ranged from 118 to 342 microseconds, with one outlier at 80 milliseconds. This data clearly indicates that Tree SRCU needs to expedite the first invocation of synchronize_srcu() during an SRCU idle period. This commit therefor introduces a srcu_might_be_idle() function that probabilistically checks whether or not SRCU is idle. This function is used by synchronize_rcu() as an additional criterion in deciding whether or not to expedite. (Hat trick to Peter Zijlstra for his earlier suggestion that this might in fact be a problem. Which for all I know might have motivated Mike to look into it.) Reported-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 02:34:40 +08:00
unsigned long flags;
struct srcu_data *sdp;
srcu: Specify auto-expedite holdoff time On small systems, in the absence of readers, expedited SRCU grace periods can complete in less than a microsecond. This means that an eight-CPU system can have all CPUs doing synchronize_srcu() in a tight loop and almost always expedite. This might actually be desirable in some situations, but in general it is a good way to needlessly burn CPU cycles. And in those situations where it is desirable, your friend is the function synchronize_srcu_expedited(). For other situations, this commit adds a kernel parameter that specifies a holdoff between completing the last SRCU grace period and auto-expediting the next. If the next grace period starts before the holdoff expires, auto-expediting is disabled. The holdoff is 50 microseconds by default, and can be tuned to the desired number of nanoseconds. A value of zero disables auto-expediting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 05:03:11 +08:00
unsigned long t;
srcu: Expedite first synchronize_srcu() when idle Classic SRCU in effect expedites the first synchronize_srcu() when SRCU is idle, and Mike Galbraith demonstrated that some use cases do in fact rely on this behavior. In particular, Mike showed that Steven Rostedt's hotplug stress script takes 55 seconds with Classic SRCU and more than 16 -minutes- when running Tree SRCU. Assuming that each Tree SRCU's call to synchronize_srcu() takes four milliseconds, this implies that Steven's test invokes synchronize_srcu() in isolation, but more than once per 200 microseconds. Mike used ftrace to demonstrate that the time between successive calls to synchronize_srcu() ranged from 118 to 342 microseconds, with one outlier at 80 milliseconds. This data clearly indicates that Tree SRCU needs to expedite the first invocation of synchronize_srcu() during an SRCU idle period. This commit therefor introduces a srcu_might_be_idle() function that probabilistically checks whether or not SRCU is idle. This function is used by synchronize_rcu() as an additional criterion in deciding whether or not to expedite. (Hat trick to Peter Zijlstra for his earlier suggestion that this might in fact be a problem. Which for all I know might have motivated Mike to look into it.) Reported-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 02:34:40 +08:00
/* If the local srcu_data structure has callbacks, not idle. */
local_irq_save(flags);
sdp = this_cpu_ptr(sp->sda);
if (rcu_segcblist_pend_cbs(&sdp->srcu_cblist)) {
local_irq_restore(flags);
return false; /* Callbacks already present, so not idle. */
}
local_irq_restore(flags);
/*
* No local callbacks, so probabalistically probe global state.
* Exact information would require acquiring locks, which would
* kill scalability, hence the probabalistic nature of the probe.
*/
srcu: Specify auto-expedite holdoff time On small systems, in the absence of readers, expedited SRCU grace periods can complete in less than a microsecond. This means that an eight-CPU system can have all CPUs doing synchronize_srcu() in a tight loop and almost always expedite. This might actually be desirable in some situations, but in general it is a good way to needlessly burn CPU cycles. And in those situations where it is desirable, your friend is the function synchronize_srcu_expedited(). For other situations, this commit adds a kernel parameter that specifies a holdoff between completing the last SRCU grace period and auto-expediting the next. If the next grace period starts before the holdoff expires, auto-expediting is disabled. The holdoff is 50 microseconds by default, and can be tuned to the desired number of nanoseconds. A value of zero disables auto-expediting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 05:03:11 +08:00
/* First, see if enough time has passed since the last GP. */
t = ktime_get_mono_fast_ns();
if (exp_holdoff == 0 ||
time_in_range_open(t, sp->srcu_last_gp_end,
sp->srcu_last_gp_end + exp_holdoff))
return false; /* Too soon after last GP. */
/* Next, check for probable idleness. */
srcu: Expedite first synchronize_srcu() when idle Classic SRCU in effect expedites the first synchronize_srcu() when SRCU is idle, and Mike Galbraith demonstrated that some use cases do in fact rely on this behavior. In particular, Mike showed that Steven Rostedt's hotplug stress script takes 55 seconds with Classic SRCU and more than 16 -minutes- when running Tree SRCU. Assuming that each Tree SRCU's call to synchronize_srcu() takes four milliseconds, this implies that Steven's test invokes synchronize_srcu() in isolation, but more than once per 200 microseconds. Mike used ftrace to demonstrate that the time between successive calls to synchronize_srcu() ranged from 118 to 342 microseconds, with one outlier at 80 milliseconds. This data clearly indicates that Tree SRCU needs to expedite the first invocation of synchronize_srcu() during an SRCU idle period. This commit therefor introduces a srcu_might_be_idle() function that probabilistically checks whether or not SRCU is idle. This function is used by synchronize_rcu() as an additional criterion in deciding whether or not to expedite. (Hat trick to Peter Zijlstra for his earlier suggestion that this might in fact be a problem. Which for all I know might have motivated Mike to look into it.) Reported-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 02:34:40 +08:00
curseq = rcu_seq_current(&sp->srcu_gp_seq);
smp_mb(); /* Order ->srcu_gp_seq with ->srcu_gp_seq_needed. */
if (ULONG_CMP_LT(curseq, READ_ONCE(sp->srcu_gp_seq_needed)))
return false; /* Grace period in progress, so not idle. */
smp_mb(); /* Order ->srcu_gp_seq with prior access. */
if (curseq != rcu_seq_current(&sp->srcu_gp_seq))
return false; /* GP # changed, so not idle. */
return true; /* With reasonable probability, idle! */
}
srcu: Add DEBUG_OBJECTS_RCU_HEAD functionality This commit adds DEBUG_OBJECTS_RCU_HEAD checking to detect call_srcu() counterparts to double-free bugs. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-29 06:39:34 +08:00
/*
* SRCU callback function to leak a callback.
*/
static void srcu_leak_callback(struct rcu_head *rhp)
{
}
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
/*
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* Enqueue an SRCU callback on the srcu_data structure associated with
* the current CPU and the specified srcu_struct structure, initiating
* grace-period processing if it is not already running.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
*
* Note that all CPUs must agree that the grace period extended beyond
* all pre-existing SRCU read-side critical section. On systems with
* more than one CPU, this means that when "func()" is invoked, each CPU
* is guaranteed to have executed a full memory barrier since the end of
* its last corresponding SRCU read-side critical section whose beginning
* preceded the call to call_rcu(). It also means that each CPU executing
* an SRCU read-side critical section that continues beyond the start of
* "func()" must have executed a memory barrier after the call_rcu()
* but before the beginning of that SRCU read-side critical section.
* Note that these guarantees include CPUs that are offline, idle, or
* executing in user mode, as well as CPUs that are executing in the kernel.
*
* Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
* resulting SRCU callback function "func()", then both CPU A and CPU
* B are guaranteed to execute a full memory barrier during the time
* interval between the call to call_rcu() and the invocation of "func()".
* This guarantee applies even if CPU A and CPU B are the same CPU (but
* again only if the system has more than one CPU).
*
* Of course, these guarantees apply only for invocations of call_srcu(),
* srcu_read_lock(), and srcu_read_unlock() that are all passed the same
* srcu_struct structure.
*/
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
void __call_srcu(struct srcu_struct *sp, struct rcu_head *rhp,
rcu_callback_t func, bool do_norm)
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
{
unsigned long flags;
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
bool needexp = false;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
bool needgp = false;
unsigned long s;
struct srcu_data *sdp;
check_init_srcu_struct(sp);
srcu: Add DEBUG_OBJECTS_RCU_HEAD functionality This commit adds DEBUG_OBJECTS_RCU_HEAD checking to detect call_srcu() counterparts to double-free bugs. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-29 06:39:34 +08:00
if (debug_rcu_head_queue(rhp)) {
/* Probable double call_srcu(), so leak the callback. */
WRITE_ONCE(rhp->func, srcu_leak_callback);
WARN_ONCE(1, "call_srcu(): Leaked duplicate callback\n");
return;
}
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
rhp->func = func;
local_irq_save(flags);
sdp = this_cpu_ptr(sp->sda);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_rcu_node(sdp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
rcu_segcblist_enqueue(&sdp->srcu_cblist, rhp, false);
rcu_segcblist_advance(&sdp->srcu_cblist,
rcu_seq_current(&sp->srcu_gp_seq));
s = rcu_seq_snap(&sp->srcu_gp_seq);
(void)rcu_segcblist_accelerate(&sdp->srcu_cblist, s);
if (ULONG_CMP_LT(sdp->srcu_gp_seq_needed, s)) {
sdp->srcu_gp_seq_needed = s;
needgp = true;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
if (!do_norm && ULONG_CMP_LT(sdp->srcu_gp_seq_needed_exp, s)) {
sdp->srcu_gp_seq_needed_exp = s;
needexp = true;
}
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irqrestore_rcu_node(sdp, flags);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (needgp)
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
srcu_funnel_gp_start(sp, sdp, s, do_norm);
else if (needexp)
srcu_funnel_exp_start(sp, sdp->mynode, s);
}
srcu: Shrink srcu.h by moving docbook and private function The call_srcu() docbook entry is currently in include/linux/srcu.h, which causes needless processing for each include point. This commit therefore moves this entry to kernel/rcu/srcutree.c, which the compiler reads only once. In addition, the srcu_batches_completed() function is used only within RCU and its torture-test suites. This commit therefore also moves this function's declaration from include/linux/srcutiny.h, include/linux/srcutree.h, and include/linux/srcuclassic.h to kernel/rcu/rcu.h. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-05 02:31:04 +08:00
/**
* call_srcu() - Queue a callback for invocation after an SRCU grace period
* @sp: srcu_struct in queue the callback
* @head: structure to be used for queueing the SRCU callback.
* @func: function to be invoked after the SRCU grace period
*
* The callback function will be invoked some time after a full SRCU
* grace period elapses, in other words after all pre-existing SRCU
* read-side critical sections have completed. However, the callback
* function might well execute concurrently with other SRCU read-side
* critical sections that started after call_srcu() was invoked. SRCU
* read-side critical sections are delimited by srcu_read_lock() and
* srcu_read_unlock(), and may be nested.
*
* The callback will be invoked from process context, but must nevertheless
* be fast and must not block.
*/
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
void call_srcu(struct srcu_struct *sp, struct rcu_head *rhp,
rcu_callback_t func)
{
__call_srcu(sp, rhp, func, true);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(call_srcu);
/*
* Helper function for synchronize_srcu() and synchronize_srcu_expedited().
*/
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
static void __synchronize_srcu(struct srcu_struct *sp, bool do_norm)
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
{
struct rcu_synchronize rcu;
RCU_LOCKDEP_WARN(lock_is_held(&sp->dep_map) ||
lock_is_held(&rcu_bh_lock_map) ||
lock_is_held(&rcu_lock_map) ||
lock_is_held(&rcu_sched_lock_map),
"Illegal synchronize_srcu() in same-type SRCU (or in RCU) read-side critical section");
if (rcu_scheduler_active == RCU_SCHEDULER_INACTIVE)
return;
might_sleep();
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
check_init_srcu_struct(sp);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
init_completion(&rcu.completion);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
init_rcu_head_on_stack(&rcu.head);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
__call_srcu(sp, &rcu.head, wakeme_after_rcu, do_norm);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
wait_for_completion(&rcu.completion);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
destroy_rcu_head_on_stack(&rcu.head);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
/**
* synchronize_srcu_expedited - Brute-force SRCU grace period
* @sp: srcu_struct with which to synchronize.
*
* Wait for an SRCU grace period to elapse, but be more aggressive about
* spinning rather than blocking when waiting.
*
* Note that synchronize_srcu_expedited() has the same deadlock and
* memory-ordering properties as does synchronize_srcu().
*/
void synchronize_srcu_expedited(struct srcu_struct *sp)
{
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
__synchronize_srcu(sp, rcu_gp_is_normal());
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(synchronize_srcu_expedited);
/**
* synchronize_srcu - wait for prior SRCU read-side critical-section completion
* @sp: srcu_struct with which to synchronize.
*
* Wait for the count to drain to zero of both indexes. To avoid the
* possible starvation of synchronize_srcu(), it waits for the count of
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* the index=((->srcu_idx & 1) ^ 1) to drain to zero at first,
* and then flip the srcu_idx and wait for the count of the other index.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
*
* Can block; must be called from process context.
*
* Note that it is illegal to call synchronize_srcu() from the corresponding
* SRCU read-side critical section; doing so will result in deadlock.
* However, it is perfectly legal to call synchronize_srcu() on one
* srcu_struct from some other srcu_struct's read-side critical section,
* as long as the resulting graph of srcu_structs is acyclic.
*
* There are memory-ordering constraints implied by synchronize_srcu().
* On systems with more than one CPU, when synchronize_srcu() returns,
* each CPU is guaranteed to have executed a full memory barrier since
* the end of its last corresponding SRCU-sched read-side critical section
* whose beginning preceded the call to synchronize_srcu(). In addition,
* each CPU having an SRCU read-side critical section that extends beyond
* the return from synchronize_srcu() is guaranteed to have executed a
* full memory barrier after the beginning of synchronize_srcu() and before
* the beginning of that SRCU read-side critical section. Note that these
* guarantees include CPUs that are offline, idle, or executing in user mode,
* as well as CPUs that are executing in the kernel.
*
* Furthermore, if CPU A invoked synchronize_srcu(), which returned
* to its caller on CPU B, then both CPU A and CPU B are guaranteed
* to have executed a full memory barrier during the execution of
* synchronize_srcu(). This guarantee applies even if CPU A and CPU B
* are the same CPU, but again only if the system has more than one CPU.
*
* Of course, these memory-ordering guarantees apply only when
* synchronize_srcu(), srcu_read_lock(), and srcu_read_unlock() are
* passed the same srcu_struct structure.
srcu: Expedite first synchronize_srcu() when idle Classic SRCU in effect expedites the first synchronize_srcu() when SRCU is idle, and Mike Galbraith demonstrated that some use cases do in fact rely on this behavior. In particular, Mike showed that Steven Rostedt's hotplug stress script takes 55 seconds with Classic SRCU and more than 16 -minutes- when running Tree SRCU. Assuming that each Tree SRCU's call to synchronize_srcu() takes four milliseconds, this implies that Steven's test invokes synchronize_srcu() in isolation, but more than once per 200 microseconds. Mike used ftrace to demonstrate that the time between successive calls to synchronize_srcu() ranged from 118 to 342 microseconds, with one outlier at 80 milliseconds. This data clearly indicates that Tree SRCU needs to expedite the first invocation of synchronize_srcu() during an SRCU idle period. This commit therefor introduces a srcu_might_be_idle() function that probabilistically checks whether or not SRCU is idle. This function is used by synchronize_rcu() as an additional criterion in deciding whether or not to expedite. (Hat trick to Peter Zijlstra for his earlier suggestion that this might in fact be a problem. Which for all I know might have motivated Mike to look into it.) Reported-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 02:34:40 +08:00
*
* If SRCU is likely idle, expedite the first request. This semantic
* was provided by Classic SRCU, and is relied upon by its users, so TREE
* SRCU must also provide it. Note that detecting idleness is heuristic
* and subject to both false positives and negatives.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
*/
void synchronize_srcu(struct srcu_struct *sp)
{
srcu: Expedite first synchronize_srcu() when idle Classic SRCU in effect expedites the first synchronize_srcu() when SRCU is idle, and Mike Galbraith demonstrated that some use cases do in fact rely on this behavior. In particular, Mike showed that Steven Rostedt's hotplug stress script takes 55 seconds with Classic SRCU and more than 16 -minutes- when running Tree SRCU. Assuming that each Tree SRCU's call to synchronize_srcu() takes four milliseconds, this implies that Steven's test invokes synchronize_srcu() in isolation, but more than once per 200 microseconds. Mike used ftrace to demonstrate that the time between successive calls to synchronize_srcu() ranged from 118 to 342 microseconds, with one outlier at 80 milliseconds. This data clearly indicates that Tree SRCU needs to expedite the first invocation of synchronize_srcu() during an SRCU idle period. This commit therefor introduces a srcu_might_be_idle() function that probabilistically checks whether or not SRCU is idle. This function is used by synchronize_rcu() as an additional criterion in deciding whether or not to expedite. (Hat trick to Peter Zijlstra for his earlier suggestion that this might in fact be a problem. Which for all I know might have motivated Mike to look into it.) Reported-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-26 02:34:40 +08:00
if (srcu_might_be_idle(sp) || rcu_gp_is_expedited())
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
synchronize_srcu_expedited(sp);
else
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
__synchronize_srcu(sp, true);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(synchronize_srcu);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
/*
* Callback function for srcu_barrier() use.
*/
static void srcu_barrier_cb(struct rcu_head *rhp)
{
struct srcu_data *sdp;
struct srcu_struct *sp;
sdp = container_of(rhp, struct srcu_data, srcu_barrier_head);
sp = sdp->sp;
if (atomic_dec_and_test(&sp->srcu_barrier_cpu_cnt))
complete(&sp->srcu_barrier_completion);
}
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
/**
* srcu_barrier - Wait until all in-flight call_srcu() callbacks complete.
* @sp: srcu_struct on which to wait for in-flight callbacks.
*/
void srcu_barrier(struct srcu_struct *sp)
{
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
int cpu;
struct srcu_data *sdp;
unsigned long s = rcu_seq_snap(&sp->srcu_barrier_seq);
check_init_srcu_struct(sp);
mutex_lock(&sp->srcu_barrier_mutex);
if (rcu_seq_done(&sp->srcu_barrier_seq, s)) {
smp_mb(); /* Force ordering following return. */
mutex_unlock(&sp->srcu_barrier_mutex);
return; /* Someone else did our work for us. */
}
rcu_seq_start(&sp->srcu_barrier_seq);
init_completion(&sp->srcu_barrier_completion);
/* Initial count prevents reaching zero until all CBs are posted. */
atomic_set(&sp->srcu_barrier_cpu_cnt, 1);
/*
* Each pass through this loop enqueues a callback, but only
* on CPUs already having callbacks enqueued. Note that if
* a CPU already has callbacks enqueue, it must have already
* registered the need for a future grace period, so all we
* need do is enqueue a callback that will use the same
* grace period as the last callback already in the queue.
*/
for_each_possible_cpu(cpu) {
sdp = per_cpu_ptr(sp->sda, cpu);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irq_rcu_node(sdp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
atomic_inc(&sp->srcu_barrier_cpu_cnt);
sdp->srcu_barrier_head.func = srcu_barrier_cb;
srcu: Add DEBUG_OBJECTS_RCU_HEAD functionality This commit adds DEBUG_OBJECTS_RCU_HEAD checking to detect call_srcu() counterparts to double-free bugs. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-29 06:39:34 +08:00
debug_rcu_head_queue(&sdp->srcu_barrier_head);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (!rcu_segcblist_entrain(&sdp->srcu_cblist,
srcu: Add DEBUG_OBJECTS_RCU_HEAD functionality This commit adds DEBUG_OBJECTS_RCU_HEAD checking to detect call_srcu() counterparts to double-free bugs. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-29 06:39:34 +08:00
&sdp->srcu_barrier_head, 0)) {
debug_rcu_head_unqueue(&sdp->srcu_barrier_head);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
atomic_dec(&sp->srcu_barrier_cpu_cnt);
srcu: Add DEBUG_OBJECTS_RCU_HEAD functionality This commit adds DEBUG_OBJECTS_RCU_HEAD checking to detect call_srcu() counterparts to double-free bugs. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-29 06:39:34 +08:00
}
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sdp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
/* Remove the initial count, at which point reaching zero can happen. */
if (atomic_dec_and_test(&sp->srcu_barrier_cpu_cnt))
complete(&sp->srcu_barrier_completion);
wait_for_completion(&sp->srcu_barrier_completion);
rcu_seq_end(&sp->srcu_barrier_seq);
mutex_unlock(&sp->srcu_barrier_mutex);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(srcu_barrier);
/**
* srcu_batches_completed - return batches completed.
* @sp: srcu_struct on which to report batch completion.
*
* Report the number of batches, correlated with, but not necessarily
* precisely the same as, the number of grace periods that have elapsed.
*/
unsigned long srcu_batches_completed(struct srcu_struct *sp)
{
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
return sp->srcu_idx;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(srcu_batches_completed);
/*
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* Core SRCU state machine. Push state bits of ->srcu_gp_seq
* to SRCU_STATE_SCAN2, and invoke srcu_gp_end() when scan has
* completed in that state.
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
*/
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
static void srcu_advance_state(struct srcu_struct *sp)
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
{
int idx;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
mutex_lock(&sp->srcu_gp_mutex);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
/*
* Because readers might be delayed for an extended period after
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
* fetching ->srcu_idx for their index, at any point in time there
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
* might well be readers using both idx=0 and idx=1. We therefore
* need to wait for readers to clear from both index values before
* invoking a callback.
*
* The load-acquire ensures that we see the accesses performed
* by the prior grace period.
*/
idx = rcu_seq_state(smp_load_acquire(&sp->srcu_gp_seq)); /* ^^^ */
if (idx == SRCU_STATE_IDLE) {
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (ULONG_CMP_GE(sp->srcu_gp_seq, sp->srcu_gp_seq_needed)) {
WARN_ON_ONCE(rcu_seq_state(sp->srcu_gp_seq));
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
mutex_unlock(&sp->srcu_gp_mutex);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
return;
}
idx = rcu_seq_state(READ_ONCE(sp->srcu_gp_seq));
if (idx == SRCU_STATE_IDLE)
srcu_gp_start(sp);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (idx != SRCU_STATE_IDLE) {
mutex_unlock(&sp->srcu_gp_mutex);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
return; /* Someone else started the grace period. */
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
if (rcu_seq_state(READ_ONCE(sp->srcu_gp_seq)) == SRCU_STATE_SCAN1) {
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
idx = 1 ^ (sp->srcu_idx & 1);
if (!try_check_zero(sp, idx, 1)) {
mutex_unlock(&sp->srcu_gp_mutex);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
return; /* readers present, retry later. */
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
}
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
srcu_flip(sp);
rcu_seq_set_state(&sp->srcu_gp_seq, SRCU_STATE_SCAN2);
}
if (rcu_seq_state(READ_ONCE(sp->srcu_gp_seq)) == SRCU_STATE_SCAN2) {
/*
* SRCU read-side critical sections are normally short,
* so check at least twice in quick succession after a flip.
*/
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
idx = 1 ^ (sp->srcu_idx & 1);
if (!try_check_zero(sp, idx, 2)) {
mutex_unlock(&sp->srcu_gp_mutex);
return; /* readers present, retry later. */
}
srcu_gp_end(sp); /* Releases ->srcu_gp_mutex. */
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
}
/*
* Invoke a limited number of SRCU callbacks that have passed through
* their grace period. If there are more to do, SRCU will reschedule
* the workqueue. Note that needed memory barriers have been executed
* in this task's context by srcu_readers_active_idx_check().
*/
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
static void srcu_invoke_callbacks(struct work_struct *work)
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
{
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
bool more;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
struct rcu_cblist ready_cbs;
struct rcu_head *rhp;
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
struct srcu_data *sdp;
struct srcu_struct *sp;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
sdp = container_of(work, struct srcu_data, work.work);
sp = sdp->sp;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
rcu_cblist_init(&ready_cbs);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irq_rcu_node(sdp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
rcu_segcblist_advance(&sdp->srcu_cblist,
rcu_seq_current(&sp->srcu_gp_seq));
if (sdp->srcu_cblist_invoking ||
!rcu_segcblist_ready_cbs(&sdp->srcu_cblist)) {
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sdp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
return; /* Someone else on the job or nothing to do. */
}
/* We are on the job! Extract and invoke ready callbacks. */
sdp->srcu_cblist_invoking = true;
rcu_segcblist_extract_done_cbs(&sdp->srcu_cblist, &ready_cbs);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sdp);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
rhp = rcu_cblist_dequeue(&ready_cbs);
for (; rhp != NULL; rhp = rcu_cblist_dequeue(&ready_cbs)) {
srcu: Add DEBUG_OBJECTS_RCU_HEAD functionality This commit adds DEBUG_OBJECTS_RCU_HEAD checking to detect call_srcu() counterparts to double-free bugs. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-29 06:39:34 +08:00
debug_rcu_head_unqueue(rhp);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
local_bh_disable();
rhp->func(rhp);
local_bh_enable();
}
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
/*
* Update counts, accelerate new callbacks, and if needed,
* schedule another round of callback invocation.
*/
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irq_rcu_node(sdp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
rcu_segcblist_insert_count(&sdp->srcu_cblist, &ready_cbs);
(void)rcu_segcblist_accelerate(&sdp->srcu_cblist,
rcu_seq_snap(&sp->srcu_gp_seq));
sdp->srcu_cblist_invoking = false;
more = rcu_segcblist_ready_cbs(&sdp->srcu_cblist);
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sdp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (more)
srcu_schedule_cbs_sdp(sdp, 0);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
/*
* Finished one round of SRCU grace period. Start another if there are
* more SRCU callbacks queued, otherwise put SRCU into not-running state.
*/
static void srcu_reschedule(struct srcu_struct *sp, unsigned long delay)
{
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
bool pushgp = true;
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_lock_irq_rcu_node(sp);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (ULONG_CMP_GE(sp->srcu_gp_seq, sp->srcu_gp_seq_needed)) {
if (!WARN_ON_ONCE(rcu_seq_state(sp->srcu_gp_seq))) {
/* All requests fulfilled, time to go idle. */
pushgp = false;
}
} else if (!rcu_seq_state(sp->srcu_gp_seq)) {
/* Outstanding request and no GP. Start one. */
srcu_gp_start(sp);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
srcu: Use rnp->lock wrappers to replace explicit memory barriers This commit uses TREE RCU's rnp->lock wrappers to replace a few explicit memory barriers. This change also has the advantage of making SRCU's memory-ordering properties be implemented in roughly the same way as they are in Tree RCU. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-05-10 06:00:14 +08:00
raw_spin_unlock_irq_rcu_node(sp);
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
if (pushgp)
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
queue_delayed_work(system_power_efficient_wq, &sp->work, delay);
}
/*
* This is the work-queue function that handles SRCU grace periods.
*/
void process_srcu(struct work_struct *work)
{
struct srcu_struct *sp;
sp = container_of(work, struct srcu_struct, work.work);
srcu: Parallelize callback handling Peter Zijlstra proposed using SRCU to reduce mmap_sem contention [1,2], however, there are workloads that could result in a high volume of concurrent invocations of call_srcu(), which with current SRCU would result in excessive lock contention on the srcu_struct structure's ->queue_lock, which protects SRCU's callback lists. This commit therefore moves SRCU to per-CPU callback lists, thus greatly reducing contention. Because a given SRCU instance no longer has a single centralized callback list, starting grace periods and invoking callbacks are both more complex than in the single-list Classic SRCU implementation. Starting grace periods and handling callbacks are now handled using an srcu_node tree that is in some ways similar to the rcu_node trees used by RCU-bh, RCU-preempt, and RCU-sched (for example, the srcu_node tree shape is controlled by exactly the same Kconfig options and boot parameters that control the shape of the rcu_node tree). In addition, the old per-CPU srcu_array structure is now named srcu_data and contains an rcu_segcblist structure named ->srcu_cblist for its callbacks (and a spinlock to protect this). The srcu_struct gets an srcu_gp_seq that is used to associate callback segments with the corresponding completion-time grace-period number. These completion-time grace-period numbers are propagated up the srcu_node tree so that the grace-period workqueue handler can determine whether additional grace periods are needed on the one hand and where to look for callbacks that are ready to be invoked. The srcu_barrier() function must now wait on all instances of the per-CPU ->srcu_cblist. Because each ->srcu_cblist is protected by ->lock, srcu_barrier() can remotely add the needed callbacks. In theory, it could also remotely start grace periods, but in practice doing so is complex and racy. And interestingly enough, it is never necessary for srcu_barrier() to start a grace period because srcu_barrier() only enqueues a callback when a callback is already present--and it turns out that a grace period has to have already been started for this pre-existing callback. Furthermore, it is only the callback that srcu_barrier() needs to wait on, not any particular grace period. Therefore, a new rcu_segcblist_entrain() function enqueues the srcu_barrier() function's callback into the same segment occupied by the last pre-existing callback in the list. The special case where all the pre-existing callbacks are on a different list (because they are in the process of being invoked) is handled by enqueuing srcu_barrier()'s callback into the RCU_DONE_TAIL segment, relying on the done-callbacks check that takes place after all callbacks are inovked. Note that the readers use the same algorithm as before. Note that there is a separate srcu_idx that tells the readers what counter to increment. This unfortunately cannot be combined with srcu_gp_seq because they need to be incremented at different times. This commit introduces some ugly #ifdefs in rcutorture. These will go away when I feel good enough about Tree SRCU to ditch Classic SRCU. Some crude performance comparisons, courtesy of a quickly hacked rcuperf asynchronous-grace-period capability: Callback Queuing Overhead ------------------------- # CPUS Classic SRCU Tree SRCU ------ ------------ --------- 2 0.349 us 0.342 us 16 31.66 us 0.4 us 41 --------- 0.417 us The times are the 90th percentiles, a statistic that was chosen to reject the overheads of the occasional srcu_barrier() call needed to avoid OOMing the test machine. The rcuperf test hangs when running Classic SRCU at 41 CPUs, hence the line of dashes. Despite the hacks to both the rcuperf code and that statistics, this is a convincing demonstration of Tree SRCU's performance and scalability advantages. [1] https://lwn.net/Articles/309030/ [2] https://patchwork.kernel.org/patch/5108281/ Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> [ paulmck: Fix initialization if synchronize_srcu_expedited() called first. ]
2017-04-06 00:01:53 +08:00
srcu_advance_state(sp);
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
srcu_reschedule(sp, srcu_get_delay(sp));
srcu: Introduce CLASSIC_SRCU Kconfig option The TREE_SRCU rewrite is large and a bit on the non-simple side, so this commit helps reduce risk by allowing the old v4.11 SRCU algorithm to be selected using a new CLASSIC_SRCU Kconfig option that depends on RCU_EXPERT. The default is to use the new TREE_SRCU and TINY_SRCU algorithms, in order to help get these the testing that they need. However, if your users do not require the update-side scalability that is to be provided by TREE_SRCU, select RCU_EXPERT and then CLASSIC_SRCU to revert back to the old classic SRCU algorithm. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-03-26 08:23:44 +08:00
}
EXPORT_SYMBOL_GPL(process_srcu);
srcu: Make rcutorture writer stalls print SRCU GP state In the past, SRCU was simple enough that there was little point in making the rcutorture writer stall messages print the SRCU grace-period number state. With the advent of Tree SRCU, this has changed. This commit therefore makes Classic, Tiny, and Tree SRCU report this state to rcutorture as needed. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 08:17:35 +08:00
void srcutorture_get_gp_data(enum rcutorture_type test_type,
srcu: Expedited grace periods with reduced memory contention Commit f60d231a87c5 ("srcu: Crude control of expedited grace periods") introduced a per-srcu_struct atomic counter to track outstanding requests for grace periods. This works, but represents a memory-contention bottleneck. This commit therefore uses the srcu_node combining tree to remove this bottleneck. This commit adds new ->srcu_gp_seq_needed_exp fields to the srcu_data, srcu_node, and srcu_struct structures, which track the farthest-in-the-future grace period that must be expedited, which in turn requires that all nearer-term grace periods also be expedited. Requests for expediting start with the srcu_data structure, run up through the srcu_node tree, and end at the srcu_struct structure. Note that it may be necessary to expedite a grace period that just now started, and this is handled by a new srcu_funnel_exp_start() function, which is invoked when the grace period itself is already in its way, but when that grace period was not marked as expedited. A new srcu_get_delay() function returns zero if there is at least one expedited SRCU grace period in flight, or SRCU_INTERVAL otherwise. This function is used to calculate delays: Normal grace periods are allowed to extend in order to cover more requests with a given grace-period computation, which decreases per-request overhead. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-25 07:02:09 +08:00
struct srcu_struct *sp, int *flags,
unsigned long *gpnum, unsigned long *completed)
srcu: Make rcutorture writer stalls print SRCU GP state In the past, SRCU was simple enough that there was little point in making the rcutorture writer stall messages print the SRCU grace-period number state. With the advent of Tree SRCU, this has changed. This commit therefore makes Classic, Tiny, and Tree SRCU report this state to rcutorture as needed. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Mike Galbraith <efault@gmx.de>
2017-04-19 08:17:35 +08:00
{
if (test_type != SRCU_FLAVOR)
return;
*flags = 0;
*completed = rcu_seq_ctr(sp->srcu_gp_seq);
*gpnum = rcu_seq_ctr(sp->srcu_gp_seq_needed);
}
EXPORT_SYMBOL_GPL(srcutorture_get_gp_data);
srcu: Make Classic and Tree SRCU announce themselves at bootup Currently, the only way to tell whether a given kernel is running Classic, Tiny, or Tree SRCU is to look at the .config file, which can easily be lost or associated with the wrong kernel. This commit therefore has Classic and Tree SRCU identify themselves at boot time. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-22 02:16:32 +08:00
static int __init srcu_bootup_announce(void)
{
pr_info("Hierarchical SRCU implementation.\n");
srcu: Print non-default exp_holdoff values at boot time This commit makes srcu_bootup_announce() check for non-default values of the auto-expedite holdoff time exp_holdoff and print a message if so. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-29 02:24:22 +08:00
if (exp_holdoff != DEFAULT_SRCU_EXP_HOLDOFF)
pr_info("\tNon-default auto-expedite holdoff of %lu ns.\n", exp_holdoff);
srcu: Make Classic and Tree SRCU announce themselves at bootup Currently, the only way to tell whether a given kernel is running Classic, Tiny, or Tree SRCU is to look at the .config file, which can easily be lost or associated with the wrong kernel. This commit therefore has Classic and Tree SRCU identify themselves at boot time. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2017-04-22 02:16:32 +08:00
return 0;
}
early_initcall(srcu_bootup_announce);
Reference in New Issue Copy Permalink