2019-06-04 16:11:33 +08:00
|
|
|
// SPDX-License-Identifier: GPL-2.0-only
|
2005-04-17 06:20:36 +08:00
|
|
|
/*
|
|
|
|
* linux/drivers/cpufreq/cpufreq.c
|
|
|
|
*
|
|
|
|
* Copyright (C) 2001 Russell King
|
|
|
|
* (C) 2002 - 2003 Dominik Brodowski <linux@brodo.de>
|
2013-06-19 16:49:33 +08:00
|
|
|
* (C) 2013 Viresh Kumar <viresh.kumar@linaro.org>
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2005-10-31 06:59:54 +08:00
|
|
|
* Oct 2005 - Ashok Raj <ashok.raj@intel.com>
|
2006-02-28 13:43:23 +08:00
|
|
|
* Added handling for CPU hotplug
|
2006-03-05 16:37:23 +08:00
|
|
|
* Feb 2006 - Jacob Shin <jacob.shin@amd.com>
|
|
|
|
* Fix handling for CPU hotplug -- affected CPUs
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
|
2012-10-23 07:29:03 +08:00
|
|
|
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
|
|
|
|
|
2013-08-07 01:23:03 +08:00
|
|
|
#include <linux/cpu.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/cpufreq.h>
|
2019-01-30 13:22:01 +08:00
|
|
|
#include <linux/cpu_cooling.h>
|
2005-04-17 06:20:36 +08:00
|
|
|
#include <linux/delay.h>
|
|
|
|
#include <linux/device.h>
|
2013-08-07 01:23:03 +08:00
|
|
|
#include <linux/init.h>
|
|
|
|
#include <linux/kernel_stat.h>
|
|
|
|
#include <linux/module.h>
|
2006-01-14 07:54:22 +08:00
|
|
|
#include <linux/mutex.h>
|
2019-07-08 18:57:52 +08:00
|
|
|
#include <linux/pm_qos.h>
|
2013-08-07 01:23:03 +08:00
|
|
|
#include <linux/slab.h>
|
2014-03-04 11:00:26 +08:00
|
|
|
#include <linux/suspend.h>
|
2014-12-24 14:09:48 +08:00
|
|
|
#include <linux/syscore_ops.h>
|
2013-08-07 01:23:03 +08:00
|
|
|
#include <linux/tick.h>
|
2022-05-04 16:21:35 +08:00
|
|
|
#include <linux/units.h>
|
2010-04-20 19:17:36 +08:00
|
|
|
#include <trace/events/power.h>
|
|
|
|
|
2015-01-27 16:36:08 +08:00
|
|
|
static LIST_HEAD(cpufreq_policy_list);
|
2015-05-12 14:50:11 +08:00
|
|
|
|
|
|
|
/* Macros to iterate over CPU policies */
|
2016-02-22 02:53:12 +08:00
|
|
|
#define for_each_suitable_policy(__policy, __active) \
|
|
|
|
list_for_each_entry(__policy, &cpufreq_policy_list, policy_list) \
|
|
|
|
if ((__active) == !policy_is_inactive(__policy))
|
2015-05-12 14:50:11 +08:00
|
|
|
|
|
|
|
#define for_each_active_policy(__policy) \
|
|
|
|
for_each_suitable_policy(__policy, true)
|
|
|
|
#define for_each_inactive_policy(__policy) \
|
|
|
|
for_each_suitable_policy(__policy, false)
|
|
|
|
|
2015-01-27 16:36:09 +08:00
|
|
|
/* Iterate over governors */
|
|
|
|
static LIST_HEAD(cpufreq_governor_list);
|
|
|
|
#define for_each_governor(__governor) \
|
|
|
|
list_for_each_entry(__governor, &cpufreq_governor_list, governor_list)
|
|
|
|
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
static char default_governor[CPUFREQ_NAME_LEN];
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2006-08-12 05:59:28 +08:00
|
|
|
* The "cpufreq driver" - the arch- or hardware-dependent low
|
2005-04-17 06:20:36 +08:00
|
|
|
* level driver of CPUFreq support, and its spinlock. This lock
|
|
|
|
* also protects the cpufreq_cpu_data array.
|
|
|
|
*/
|
2013-04-29 06:08:16 +08:00
|
|
|
static struct cpufreq_driver *cpufreq_driver;
|
2008-03-26 06:06:53 +08:00
|
|
|
static DEFINE_PER_CPU(struct cpufreq_policy *, cpufreq_cpu_data);
|
2013-06-19 16:49:33 +08:00
|
|
|
static DEFINE_RWLOCK(cpufreq_driver_lock);
|
|
|
|
|
2020-09-02 04:55:47 +08:00
|
|
|
static DEFINE_STATIC_KEY_FALSE(cpufreq_freq_invariance);
|
|
|
|
bool cpufreq_supports_freq_invariance(void)
|
|
|
|
{
|
|
|
|
return static_branch_likely(&cpufreq_freq_invariance);
|
|
|
|
}
|
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
/* Flag to suspend/resume CPUFreq governors */
|
|
|
|
static bool cpufreq_suspended;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
static inline bool has_target(void)
|
|
|
|
{
|
|
|
|
return cpufreq_driver->target_index || cpufreq_driver->target;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* internal prototypes */
|
2015-01-02 15:04:29 +08:00
|
|
|
static unsigned int __cpufreq_get(struct cpufreq_policy *policy);
|
2016-05-14 07:01:46 +08:00
|
|
|
static int cpufreq_init_governor(struct cpufreq_policy *policy);
|
|
|
|
static void cpufreq_exit_governor(struct cpufreq_policy *policy);
|
|
|
|
static void cpufreq_governor_limits(struct cpufreq_policy *policy);
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
static int cpufreq_set_policy(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_governor *new_gov,
|
|
|
|
unsigned int new_pol);
|
2016-05-12 21:14:12 +08:00
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2006-02-28 13:43:23 +08:00
|
|
|
* Two notifier lists: the "policy" list is involved in the
|
|
|
|
* validation process for a new CPU frequency policy; the
|
2005-04-17 06:20:36 +08:00
|
|
|
* "transition" list for kernel code that needs to handle
|
|
|
|
* changes to devices when the CPU clock speed changes.
|
|
|
|
* The mutex locks both lists.
|
|
|
|
*/
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
static BLOCKING_NOTIFIER_HEAD(cpufreq_policy_notifier_list);
|
2018-05-25 18:19:58 +08:00
|
|
|
SRCU_NOTIFIER_HEAD_STATIC(cpufreq_transition_notifier_list);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
static int off __read_mostly;
|
2012-10-26 06:51:32 +08:00
|
|
|
static int cpufreq_disabled(void)
|
2012-03-14 07:18:39 +08:00
|
|
|
{
|
|
|
|
return off;
|
|
|
|
}
|
|
|
|
void disable_cpufreq(void)
|
|
|
|
{
|
|
|
|
off = 1;
|
|
|
|
}
|
2009-01-18 14:37:11 +08:00
|
|
|
static DEFINE_MUTEX(cpufreq_governor_mutex);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-03-27 23:58:58 +08:00
|
|
|
bool have_governor_per_policy(void)
|
|
|
|
{
|
2013-10-02 16:43:18 +08:00
|
|
|
return !!(cpufreq_driver->flags & CPUFREQ_HAVE_GOVERNOR_PER_POLICY);
|
2013-03-27 23:58:58 +08:00
|
|
|
}
|
2013-05-16 13:09:56 +08:00
|
|
|
EXPORT_SYMBOL_GPL(have_governor_per_policy);
|
2013-03-27 23:58:58 +08:00
|
|
|
|
2020-02-03 23:45:17 +08:00
|
|
|
static struct kobject *cpufreq_global_kobject;
|
|
|
|
|
2013-05-16 13:09:57 +08:00
|
|
|
struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
if (have_governor_per_policy())
|
|
|
|
return &policy->kobj;
|
|
|
|
else
|
|
|
|
return cpufreq_global_kobject;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
|
|
|
|
|
2013-05-17 19:26:32 +08:00
|
|
|
static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
|
|
|
|
{
|
2019-11-21 10:44:28 +08:00
|
|
|
struct kernel_cpustat kcpustat;
|
2013-05-17 19:26:32 +08:00
|
|
|
u64 cur_wall_time;
|
2019-11-21 10:44:28 +08:00
|
|
|
u64 idle_time;
|
2013-05-17 19:26:32 +08:00
|
|
|
u64 busy_time;
|
|
|
|
|
2017-01-31 11:09:19 +08:00
|
|
|
cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
|
2013-05-17 19:26:32 +08:00
|
|
|
|
2019-11-21 10:44:28 +08:00
|
|
|
kcpustat_cpu_fetch(&kcpustat, cpu);
|
|
|
|
|
|
|
|
busy_time = kcpustat.cpustat[CPUTIME_USER];
|
|
|
|
busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
|
|
|
|
busy_time += kcpustat.cpustat[CPUTIME_IRQ];
|
|
|
|
busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
|
|
|
|
busy_time += kcpustat.cpustat[CPUTIME_STEAL];
|
|
|
|
busy_time += kcpustat.cpustat[CPUTIME_NICE];
|
2013-05-17 19:26:32 +08:00
|
|
|
|
|
|
|
idle_time = cur_wall_time - busy_time;
|
|
|
|
if (wall)
|
2017-01-31 11:09:19 +08:00
|
|
|
*wall = div_u64(cur_wall_time, NSEC_PER_USEC);
|
2013-05-17 19:26:32 +08:00
|
|
|
|
2017-01-31 11:09:19 +08:00
|
|
|
return div_u64(idle_time, NSEC_PER_USEC);
|
2013-05-17 19:26:32 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
|
|
|
|
{
|
|
|
|
u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
|
|
|
|
|
|
|
|
if (idle_time == -1ULL)
|
|
|
|
return get_cpu_idle_time_jiffy(cpu, wall);
|
|
|
|
else if (!io_busy)
|
|
|
|
idle_time += get_cpu_iowait_time_us(cpu, wall);
|
|
|
|
|
|
|
|
return idle_time;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(get_cpu_idle_time);
|
|
|
|
|
2013-10-03 22:59:07 +08:00
|
|
|
/*
|
|
|
|
* This is a generic cpufreq init() routine which can be used by cpufreq
|
|
|
|
* drivers of SMP systems. It will do following:
|
|
|
|
* - validate & show freq table passed
|
|
|
|
* - set policies transition latency
|
|
|
|
* - policy->cpus with all possible CPUs
|
|
|
|
*/
|
2019-07-16 12:06:08 +08:00
|
|
|
void cpufreq_generic_init(struct cpufreq_policy *policy,
|
2013-10-03 22:59:07 +08:00
|
|
|
struct cpufreq_frequency_table *table,
|
|
|
|
unsigned int transition_latency)
|
|
|
|
{
|
2018-02-26 13:08:45 +08:00
|
|
|
policy->freq_table = table;
|
2013-10-03 22:59:07 +08:00
|
|
|
policy->cpuinfo.transition_latency = transition_latency;
|
|
|
|
|
|
|
|
/*
|
2015-05-23 01:18:22 +08:00
|
|
|
* The driver only supports the SMP configuration where all processors
|
2013-10-03 22:59:07 +08:00
|
|
|
* share the clock and voltage and clock.
|
|
|
|
*/
|
|
|
|
cpumask_setall(policy->cpus);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_generic_init);
|
|
|
|
|
2015-09-16 08:17:49 +08:00
|
|
|
struct cpufreq_policy *cpufreq_cpu_get_raw(unsigned int cpu)
|
2014-01-09 23:08:43 +08:00
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
|
|
|
|
|
2015-05-08 14:23:45 +08:00
|
|
|
return policy && cpumask_test_cpu(cpu, policy->cpus) ? policy : NULL;
|
|
|
|
}
|
2015-09-16 08:17:49 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_cpu_get_raw);
|
2015-05-08 14:23:45 +08:00
|
|
|
|
|
|
|
unsigned int cpufreq_generic_get(unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get_raw(cpu);
|
|
|
|
|
2014-01-09 23:08:43 +08:00
|
|
|
if (!policy || IS_ERR(policy->clk)) {
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_err("%s: No %s associated to cpu: %d\n",
|
|
|
|
__func__, policy ? "clk" : "policy", cpu);
|
2014-01-09 23:08:43 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
return clk_get_rate(policy->clk) / 1000;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_generic_get);
|
|
|
|
|
2015-02-19 19:32:03 +08:00
|
|
|
/**
|
2019-03-05 18:44:04 +08:00
|
|
|
* cpufreq_cpu_get - Return policy for a CPU and mark it as busy.
|
|
|
|
* @cpu: CPU to find the policy for.
|
2015-02-19 19:32:03 +08:00
|
|
|
*
|
2019-03-05 18:44:04 +08:00
|
|
|
* Call cpufreq_cpu_get_raw() to obtain a cpufreq policy for @cpu and increment
|
|
|
|
* the kobject reference counter of that policy. Return a valid policy on
|
|
|
|
* success or NULL on failure.
|
2015-02-19 19:32:03 +08:00
|
|
|
*
|
2019-03-05 18:44:04 +08:00
|
|
|
* The policy returned by this function has to be released with the help of
|
|
|
|
* cpufreq_cpu_put() to balance its kobject reference counter properly.
|
2015-02-19 19:32:03 +08:00
|
|
|
*/
|
2013-08-07 01:23:11 +08:00
|
|
|
struct cpufreq_policy *cpufreq_cpu_get(unsigned int cpu)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-08-07 01:23:11 +08:00
|
|
|
struct cpufreq_policy *policy = NULL;
|
2005-04-17 06:20:36 +08:00
|
|
|
unsigned long flags;
|
|
|
|
|
2015-02-19 19:32:05 +08:00
|
|
|
if (WARN_ON(cpu >= nr_cpu_ids))
|
2013-08-07 01:23:11 +08:00
|
|
|
return NULL;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/* get the cpufreq driver */
|
2013-04-29 06:08:16 +08:00
|
|
|
read_lock_irqsave(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
if (cpufreq_driver) {
|
|
|
|
/* get the CPU */
|
2015-05-08 14:23:45 +08:00
|
|
|
policy = cpufreq_cpu_get_raw(cpu);
|
2013-08-07 01:23:11 +08:00
|
|
|
if (policy)
|
|
|
|
kobject_get(&policy->kobj);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:11 +08:00
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-08-07 01:23:05 +08:00
|
|
|
return policy;
|
2012-07-21 02:14:38 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_cpu_get);
|
|
|
|
|
2015-02-19 19:32:03 +08:00
|
|
|
/**
|
2019-03-05 18:44:04 +08:00
|
|
|
* cpufreq_cpu_put - Decrement kobject usage counter for cpufreq policy.
|
|
|
|
* @policy: cpufreq policy returned by cpufreq_cpu_get().
|
2015-02-19 19:32:03 +08:00
|
|
|
*/
|
2013-08-07 01:23:05 +08:00
|
|
|
void cpufreq_cpu_put(struct cpufreq_policy *policy)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-08-07 01:23:11 +08:00
|
|
|
kobject_put(&policy->kobj);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_cpu_put);
|
|
|
|
|
2019-03-26 19:16:58 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_cpu_release - Unlock a policy and decrement its usage counter.
|
|
|
|
* @policy: cpufreq policy returned by cpufreq_cpu_acquire().
|
|
|
|
*/
|
2019-03-26 19:19:52 +08:00
|
|
|
void cpufreq_cpu_release(struct cpufreq_policy *policy)
|
2019-03-26 19:16:58 +08:00
|
|
|
{
|
|
|
|
if (WARN_ON(!policy))
|
|
|
|
return;
|
|
|
|
|
|
|
|
lockdep_assert_held(&policy->rwsem);
|
|
|
|
|
|
|
|
up_write(&policy->rwsem);
|
|
|
|
|
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_cpu_acquire - Find policy for a CPU, mark it as busy and lock it.
|
|
|
|
* @cpu: CPU to find the policy for.
|
|
|
|
*
|
|
|
|
* Call cpufreq_cpu_get() to get a reference on the cpufreq policy for @cpu and
|
|
|
|
* if the policy returned by it is not NULL, acquire its rwsem for writing.
|
|
|
|
* Return the policy if it is active or release it and return NULL otherwise.
|
|
|
|
*
|
|
|
|
* The policy returned by this function has to be released with the help of
|
|
|
|
* cpufreq_cpu_release() in order to release its rwsem and balance its usage
|
|
|
|
* counter properly.
|
|
|
|
*/
|
2019-03-26 19:19:52 +08:00
|
|
|
struct cpufreq_policy *cpufreq_cpu_acquire(unsigned int cpu)
|
2019-03-26 19:16:58 +08:00
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
|
|
|
|
|
|
|
if (!policy)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
down_write(&policy->rwsem);
|
|
|
|
|
|
|
|
if (policy_is_inactive(policy)) {
|
|
|
|
cpufreq_cpu_release(policy);
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return policy;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*********************************************************************
|
|
|
|
* EXTERNALLY AFFECTING FREQUENCY CHANGES *
|
|
|
|
*********************************************************************/
|
|
|
|
|
2020-11-19 03:02:42 +08:00
|
|
|
/**
|
|
|
|
* adjust_jiffies - Adjust the system "loops_per_jiffy".
|
|
|
|
* @val: CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
|
|
|
|
* @ci: Frequency change information.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* This function alters the system "loops_per_jiffy" for the clock
|
|
|
|
* speed change. Note that loops_per_jiffy cannot be updated on SMP
|
2006-02-28 13:43:23 +08:00
|
|
|
* systems as each CPU might be scaled differently. So, use the arch
|
2005-04-17 06:20:36 +08:00
|
|
|
* per-CPU loops_per_jiffy value wherever possible.
|
|
|
|
*/
|
2006-01-15 05:20:43 +08:00
|
|
|
static void adjust_jiffies(unsigned long val, struct cpufreq_freqs *ci)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2015-01-02 15:04:34 +08:00
|
|
|
#ifndef CONFIG_SMP
|
|
|
|
static unsigned long l_p_j_ref;
|
|
|
|
static unsigned int l_p_j_ref_freq;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ci->flags & CPUFREQ_CONST_LOOPS)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (!l_p_j_ref_freq) {
|
|
|
|
l_p_j_ref = loops_per_jiffy;
|
|
|
|
l_p_j_ref_freq = ci->old;
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("saving %lu as reference value for loops_per_jiffy; freq is %u kHz\n",
|
|
|
|
l_p_j_ref, l_p_j_ref_freq);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2014-03-19 13:54:58 +08:00
|
|
|
if (val == CPUFREQ_POSTCHANGE && ci->old != ci->new) {
|
2006-10-26 18:50:58 +08:00
|
|
|
loops_per_jiffy = cpufreq_scale(l_p_j_ref, l_p_j_ref_freq,
|
|
|
|
ci->new);
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("scaling loops_per_jiffy to %lu for frequency %u kHz\n",
|
|
|
|
loops_per_jiffy, ci->new);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
#endif
|
2015-01-02 15:04:34 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2018-05-10 17:30:29 +08:00
|
|
|
/**
|
2020-11-19 03:02:42 +08:00
|
|
|
* cpufreq_notify_transition - Notify frequency transition and adjust jiffies.
|
2018-05-10 17:30:29 +08:00
|
|
|
* @policy: cpufreq policy to enable fast frequency switching for.
|
|
|
|
* @freqs: contain details of the frequency update.
|
|
|
|
* @state: set to CPUFREQ_PRECHANGE or CPUFREQ_POSTCHANGE.
|
|
|
|
*
|
2020-11-19 03:02:42 +08:00
|
|
|
* This function calls the transition notifiers and adjust_jiffies().
|
|
|
|
*
|
|
|
|
* It is called twice on all CPU frequency changes that have external effects.
|
2018-05-10 17:30:29 +08:00
|
|
|
*/
|
|
|
|
static void cpufreq_notify_transition(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_freqs *freqs,
|
|
|
|
unsigned int state)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2019-04-29 17:33:58 +08:00
|
|
|
int cpu;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
BUG_ON(irqs_disabled());
|
|
|
|
|
2013-01-18 00:22:21 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return;
|
|
|
|
|
2019-04-29 17:33:58 +08:00
|
|
|
freqs->policy = policy;
|
2013-04-29 06:08:16 +08:00
|
|
|
freqs->flags = cpufreq_driver->flags;
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("notification %u of frequency transition to %u kHz\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
state, freqs->new);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
switch (state) {
|
|
|
|
case CPUFREQ_PRECHANGE:
|
2018-05-10 17:30:29 +08:00
|
|
|
/*
|
|
|
|
* Detect if the driver reported a value as "old frequency"
|
2006-02-01 07:53:55 +08:00
|
|
|
* which is not equal to what the cpufreq core thinks is
|
|
|
|
* "old frequency".
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2019-06-28 13:16:55 +08:00
|
|
|
if (policy->cur && policy->cur != freqs->old) {
|
|
|
|
pr_debug("Warning: CPU frequency is %u, cpufreq assumed %u kHz\n",
|
|
|
|
freqs->old, policy->cur);
|
|
|
|
freqs->old = policy->cur;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2018-05-10 17:30:29 +08:00
|
|
|
|
2019-04-29 17:33:58 +08:00
|
|
|
srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
|
|
|
|
CPUFREQ_PRECHANGE, freqs);
|
2018-05-10 17:30:29 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
adjust_jiffies(CPUFREQ_PRECHANGE, freqs);
|
|
|
|
break;
|
2006-02-01 07:53:55 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
case CPUFREQ_POSTCHANGE:
|
|
|
|
adjust_jiffies(CPUFREQ_POSTCHANGE, freqs);
|
2018-05-10 17:30:29 +08:00
|
|
|
pr_debug("FREQ: %u - CPUs: %*pbl\n", freqs->new,
|
|
|
|
cpumask_pr_args(policy->cpus));
|
|
|
|
|
2019-04-29 17:33:58 +08:00
|
|
|
for_each_cpu(cpu, policy->cpus)
|
|
|
|
trace_cpu_frequency(freqs->new, cpu);
|
|
|
|
|
|
|
|
srcu_notifier_call_chain(&cpufreq_transition_notifier_list,
|
|
|
|
CPUFREQ_POSTCHANGE, freqs);
|
2018-05-10 17:30:29 +08:00
|
|
|
|
2016-06-01 04:14:44 +08:00
|
|
|
cpufreq_stats_record_transition(policy, freqs->new);
|
2018-05-10 17:30:29 +08:00
|
|
|
policy->cur = freqs->new;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
}
|
2013-06-19 16:49:33 +08:00
|
|
|
|
2013-12-02 13:34:12 +08:00
|
|
|
/* Do post notifications when there are chances that transition has failed */
|
2014-03-24 16:05:46 +08:00
|
|
|
static void cpufreq_notify_post_transition(struct cpufreq_policy *policy,
|
2013-12-02 13:34:12 +08:00
|
|
|
struct cpufreq_freqs *freqs, int transition_failed)
|
|
|
|
{
|
|
|
|
cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
|
|
|
|
if (!transition_failed)
|
|
|
|
return;
|
|
|
|
|
|
|
|
swap(freqs->old, freqs->new);
|
|
|
|
cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
|
|
|
|
cpufreq_notify_transition(policy, freqs, CPUFREQ_POSTCHANGE);
|
|
|
|
}
|
|
|
|
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
void cpufreq_freq_transition_begin(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_freqs *freqs)
|
|
|
|
{
|
cpufreq: Catch double invocations of cpufreq_freq_transition_begin/end
Some cpufreq drivers were redundantly invoking the _begin() and _end()
APIs around frequency transitions, and this double invocation (one from
the cpufreq core and the other from the cpufreq driver) used to result
in a self-deadlock, leading to system hangs during boot. (The _begin()
API makes contending callers wait until the previous invocation is
complete. Hence, the cpufreq driver would end up waiting on itself!).
Now all such drivers have been fixed, but debugging this issue was not
very straight-forward (even lockdep didn't catch this). So let us add a
debug infrastructure to the cpufreq core to catch such issues more easily
in the future.
We add a new field called 'transition_task' to the policy structure, to keep
track of the task which is performing the frequency transition. Using this
field, we make note of this task during _begin() and print a warning if we
find a case where the same task is calling _begin() again, before completing
the previous frequency transition using the corresponding _end().
We have left out ASYNC_NOTIFICATION drivers from this debug infrastructure
for 2 reasons:
1. At the moment, we have no way to avoid a particular scenario where this
debug infrastructure can emit false-positive warnings for such drivers.
The scenario is depicted below:
Task A Task B
/* 1st freq transition */
Invoke _begin() {
...
...
}
Change the frequency
/* 2nd freq transition */
Invoke _begin() {
... //waiting for B to
... //finish _end() for
... //the 1st transition
... | Got interrupt for successful
... | change of frequency (1st one).
... |
... | /* 1st freq transition */
... | Invoke _end() {
... | ...
... V }
...
...
}
This scenario is actually deadlock-free because, once Task A changes the
frequency, it is Task B's responsibility to invoke the corresponding
_end() for the 1st frequency transition. Hence it is perfectly legal for
Task A to go ahead and attempt another frequency transition in the meantime.
(Of course it won't be able to proceed until Task B finishes the 1st _end(),
but this doesn't cause a deadlock or a hang).
The debug infrastructure cannot handle this scenario and will treat it as
a deadlock and print a warning. To avoid this, we exclude such drivers
from the purview of this code.
2. Luckily, we don't _need_ this infrastructure for ASYNC_NOTIFICATION drivers
at all! The cpufreq core does not automatically invoke the _begin() and
_end() APIs during frequency transitions in such drivers. Thus, the driver
alone is responsible for invoking _begin()/_end() and hence there shouldn't
be any conflicts which lead to double invocations. So, we can skip these
drivers, since the probability that such drivers will hit this problem is
extremely low, as outlined above.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-05-05 15:22:39 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Catch double invocations of _begin() which lead to self-deadlock.
|
|
|
|
* ASYNC_NOTIFICATION drivers are left out because the cpufreq core
|
|
|
|
* doesn't invoke _begin() on their behalf, and hence the chances of
|
|
|
|
* double invocations are very low. Moreover, there are scenarios
|
|
|
|
* where these checks can emit false-positive warnings in these
|
|
|
|
* drivers; so we avoid that by skipping them altogether.
|
|
|
|
*/
|
|
|
|
WARN_ON(!(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION)
|
|
|
|
&& current == policy->transition_task);
|
|
|
|
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
wait:
|
|
|
|
wait_event(policy->transition_wait, !policy->transition_ongoing);
|
|
|
|
|
|
|
|
spin_lock(&policy->transition_lock);
|
|
|
|
|
|
|
|
if (unlikely(policy->transition_ongoing)) {
|
|
|
|
spin_unlock(&policy->transition_lock);
|
|
|
|
goto wait;
|
|
|
|
}
|
|
|
|
|
|
|
|
policy->transition_ongoing = true;
|
cpufreq: Catch double invocations of cpufreq_freq_transition_begin/end
Some cpufreq drivers were redundantly invoking the _begin() and _end()
APIs around frequency transitions, and this double invocation (one from
the cpufreq core and the other from the cpufreq driver) used to result
in a self-deadlock, leading to system hangs during boot. (The _begin()
API makes contending callers wait until the previous invocation is
complete. Hence, the cpufreq driver would end up waiting on itself!).
Now all such drivers have been fixed, but debugging this issue was not
very straight-forward (even lockdep didn't catch this). So let us add a
debug infrastructure to the cpufreq core to catch such issues more easily
in the future.
We add a new field called 'transition_task' to the policy structure, to keep
track of the task which is performing the frequency transition. Using this
field, we make note of this task during _begin() and print a warning if we
find a case where the same task is calling _begin() again, before completing
the previous frequency transition using the corresponding _end().
We have left out ASYNC_NOTIFICATION drivers from this debug infrastructure
for 2 reasons:
1. At the moment, we have no way to avoid a particular scenario where this
debug infrastructure can emit false-positive warnings for such drivers.
The scenario is depicted below:
Task A Task B
/* 1st freq transition */
Invoke _begin() {
...
...
}
Change the frequency
/* 2nd freq transition */
Invoke _begin() {
... //waiting for B to
... //finish _end() for
... //the 1st transition
... | Got interrupt for successful
... | change of frequency (1st one).
... |
... | /* 1st freq transition */
... | Invoke _end() {
... | ...
... V }
...
...
}
This scenario is actually deadlock-free because, once Task A changes the
frequency, it is Task B's responsibility to invoke the corresponding
_end() for the 1st frequency transition. Hence it is perfectly legal for
Task A to go ahead and attempt another frequency transition in the meantime.
(Of course it won't be able to proceed until Task B finishes the 1st _end(),
but this doesn't cause a deadlock or a hang).
The debug infrastructure cannot handle this scenario and will treat it as
a deadlock and print a warning. To avoid this, we exclude such drivers
from the purview of this code.
2. Luckily, we don't _need_ this infrastructure for ASYNC_NOTIFICATION drivers
at all! The cpufreq core does not automatically invoke the _begin() and
_end() APIs during frequency transitions in such drivers. Thus, the driver
alone is responsible for invoking _begin()/_end() and hence there shouldn't
be any conflicts which lead to double invocations. So, we can skip these
drivers, since the probability that such drivers will hit this problem is
extremely low, as outlined above.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-05-05 15:22:39 +08:00
|
|
|
policy->transition_task = current;
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
|
|
|
|
spin_unlock(&policy->transition_lock);
|
|
|
|
|
|
|
|
cpufreq_notify_transition(policy, freqs, CPUFREQ_PRECHANGE);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_freq_transition_begin);
|
|
|
|
|
|
|
|
void cpufreq_freq_transition_end(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_freqs *freqs, int transition_failed)
|
|
|
|
{
|
2018-09-08 00:09:55 +08:00
|
|
|
if (WARN_ON(!policy->transition_ongoing))
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
cpufreq_notify_post_transition(policy, freqs, transition_failed);
|
|
|
|
|
cpufreq: move invariance setter calls in cpufreq core
To properly scale its per-entity load-tracking signals, the task scheduler
needs to be given a frequency scale factor, i.e. some image of the current
frequency the CPU is running at. Currently, this scale can be computed
either by using counters (APERF/MPERF on x86, AMU on arm64), or by
piggy-backing on the frequency selection done by cpufreq.
For the latter, drivers have to explicitly set the scale factor
themselves, despite it being purely boiler-plate code: the required
information depends entirely on the kind of frequency switch callback
implemented by the driver, i.e. either of: target_index(), target(),
fast_switch() and setpolicy().
The fitness of those callbacks with regard to driving the Frequency
Invariance Engine (FIE) is studied below:
target_index()
==============
Documentation states that the chosen frequency "must be determined by
freq_table[index].frequency". It isn't clear if it *has* to be that
frequency, or if it can use that frequency value to do some computation
that ultimately leads to a different frequency selection. All drivers
go for the former, while the vexpress-spc-cpufreq has an atypical
implementation which is handled separately.
Therefore, the hook works on the assumption the core can use
freq_table[index].frequency.
target()
=======
This has been flagged as deprecated since:
commit 9c0ebcf78fde ("cpufreq: Implement light weight ->target_index() routine")
It also doesn't have that many users:
gx-suspmod.c:439: .target = cpufreq_gx_target,
s3c24xx-cpufreq.c:428: .target = s3c_cpufreq_target,
intel_pstate.c:2528: .target = intel_cpufreq_target,
cppc_cpufreq.c:401: .target = cppc_cpufreq_set_target,
cpufreq-nforce2.c:371: .target = nforce2_target,
sh-cpufreq.c:163: .target = sh_cpufreq_target,
pcc-cpufreq.c:573: .target = pcc_cpufreq_target,
Similarly to the path taken for target_index() calls in the cpufreq core
during a frequency change, all of the drivers above will mark the end of a
frequency change by a call to cpufreq_freq_transition_end().
Therefore, cpufreq_freq_transition_end() can be used as the location for
the arch_set_freq_scale() call to potentially inform the scheduler of the
frequency change.
This change maintains the previous functionality for the drivers that
implement the target_index() callback, while also adding support for the
few drivers that implement the deprecated target() callback.
fast_switch()
=============
This callback *has* to return the frequency that was selected.
setpolicy()
===========
This callback does not have any designated way of informing what was the
end choice. But there are only two drivers using setpolicy(), and none
of them have current FIE support:
drivers/cpufreq/longrun.c:281: .setpolicy = longrun_set_policy,
drivers/cpufreq/intel_pstate.c:2215: .setpolicy = intel_pstate_set_policy,
The intel_pstate is known to use counter-driven frequency invariance.
Conclusion
==========
Given that the significant majority of current FIE enabled drivers use
callbacks that lend themselves to triggering the setting of the FIE scale
factor in a generic way, move the invariance setter calls to cpufreq core.
As a result of setting the frequency scale factor in cpufreq core, after
callbacks that lend themselves to trigger it, remove this functionality
from the driver side.
To be noted that despite marking a successful frequency change, many
cpufreq drivers will consider the new frequency as the requested
frequency, although this is might not be the one granted by the hardware.
Therefore, the call to arch_set_freq_scale() is a "best effort" one, and
it is up to the architecture if the new frequency is used in the new
frequency scale factor setting (determined by the implementation of
arch_set_freq_scale()) or eventually used by the scheduler (determined
by the implementation of arch_scale_freq_capacity()). The architecture
is in a better position to decide if it has better methods to obtain
more accurate information regarding the current frequency and use that
information instead (for example, the use of counters).
Also, the implementation to arch_set_freq_scale() will now have to handle
error conditions (current frequency == 0) in order to prevent the
overhead in cpufreq core when the default arch_set_freq_scale()
implementation is used.
Signed-off-by: Ionela Voinescu <ionela.voinescu@arm.com>
Suggested-by: Valentin Schneider <valentin.schneider@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-09-02 04:55:46 +08:00
|
|
|
arch_set_freq_scale(policy->related_cpus,
|
|
|
|
policy->cur,
|
|
|
|
policy->cpuinfo.max_freq);
|
|
|
|
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
policy->transition_ongoing = false;
|
cpufreq: Catch double invocations of cpufreq_freq_transition_begin/end
Some cpufreq drivers were redundantly invoking the _begin() and _end()
APIs around frequency transitions, and this double invocation (one from
the cpufreq core and the other from the cpufreq driver) used to result
in a self-deadlock, leading to system hangs during boot. (The _begin()
API makes contending callers wait until the previous invocation is
complete. Hence, the cpufreq driver would end up waiting on itself!).
Now all such drivers have been fixed, but debugging this issue was not
very straight-forward (even lockdep didn't catch this). So let us add a
debug infrastructure to the cpufreq core to catch such issues more easily
in the future.
We add a new field called 'transition_task' to the policy structure, to keep
track of the task which is performing the frequency transition. Using this
field, we make note of this task during _begin() and print a warning if we
find a case where the same task is calling _begin() again, before completing
the previous frequency transition using the corresponding _end().
We have left out ASYNC_NOTIFICATION drivers from this debug infrastructure
for 2 reasons:
1. At the moment, we have no way to avoid a particular scenario where this
debug infrastructure can emit false-positive warnings for such drivers.
The scenario is depicted below:
Task A Task B
/* 1st freq transition */
Invoke _begin() {
...
...
}
Change the frequency
/* 2nd freq transition */
Invoke _begin() {
... //waiting for B to
... //finish _end() for
... //the 1st transition
... | Got interrupt for successful
... | change of frequency (1st one).
... |
... | /* 1st freq transition */
... | Invoke _end() {
... | ...
... V }
...
...
}
This scenario is actually deadlock-free because, once Task A changes the
frequency, it is Task B's responsibility to invoke the corresponding
_end() for the 1st frequency transition. Hence it is perfectly legal for
Task A to go ahead and attempt another frequency transition in the meantime.
(Of course it won't be able to proceed until Task B finishes the 1st _end(),
but this doesn't cause a deadlock or a hang).
The debug infrastructure cannot handle this scenario and will treat it as
a deadlock and print a warning. To avoid this, we exclude such drivers
from the purview of this code.
2. Luckily, we don't _need_ this infrastructure for ASYNC_NOTIFICATION drivers
at all! The cpufreq core does not automatically invoke the _begin() and
_end() APIs during frequency transitions in such drivers. Thus, the driver
alone is responsible for invoking _begin()/_end() and hence there shouldn't
be any conflicts which lead to double invocations. So, we can skip these
drivers, since the probability that such drivers will hit this problem is
extremely low, as outlined above.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-05-05 15:22:39 +08:00
|
|
|
policy->transition_task = NULL;
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
|
|
|
|
wake_up(&policy->transition_wait);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_freq_transition_end);
|
|
|
|
|
2016-03-30 09:47:49 +08:00
|
|
|
/*
|
|
|
|
* Fast frequency switching status count. Positive means "enabled", negative
|
|
|
|
* means "disabled" and 0 means "not decided yet".
|
|
|
|
*/
|
|
|
|
static int cpufreq_fast_switch_count;
|
|
|
|
static DEFINE_MUTEX(cpufreq_fast_switch_lock);
|
|
|
|
|
|
|
|
static void cpufreq_list_transition_notifiers(void)
|
|
|
|
{
|
|
|
|
struct notifier_block *nb;
|
|
|
|
|
|
|
|
pr_info("Registered transition notifiers:\n");
|
|
|
|
|
|
|
|
mutex_lock(&cpufreq_transition_notifier_list.mutex);
|
|
|
|
|
|
|
|
for (nb = cpufreq_transition_notifier_list.head; nb; nb = nb->next)
|
2019-03-26 03:32:28 +08:00
|
|
|
pr_info("%pS\n", nb->notifier_call);
|
2016-03-30 09:47:49 +08:00
|
|
|
|
|
|
|
mutex_unlock(&cpufreq_transition_notifier_list.mutex);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_enable_fast_switch - Enable fast frequency switching for policy.
|
|
|
|
* @policy: cpufreq policy to enable fast frequency switching for.
|
|
|
|
*
|
|
|
|
* Try to enable fast frequency switching for @policy.
|
|
|
|
*
|
|
|
|
* The attempt will fail if there is at least one transition notifier registered
|
|
|
|
* at this point, as fast frequency switching is quite fundamentally at odds
|
|
|
|
* with transition notifiers. Thus if successful, it will make registration of
|
|
|
|
* transition notifiers fail going forward.
|
|
|
|
*/
|
|
|
|
void cpufreq_enable_fast_switch(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
lockdep_assert_held(&policy->rwsem);
|
|
|
|
|
|
|
|
if (!policy->fast_switch_possible)
|
|
|
|
return;
|
|
|
|
|
|
|
|
mutex_lock(&cpufreq_fast_switch_lock);
|
|
|
|
if (cpufreq_fast_switch_count >= 0) {
|
|
|
|
cpufreq_fast_switch_count++;
|
|
|
|
policy->fast_switch_enabled = true;
|
|
|
|
} else {
|
|
|
|
pr_warn("CPU%u: Fast frequency switching not enabled\n",
|
|
|
|
policy->cpu);
|
|
|
|
cpufreq_list_transition_notifiers();
|
|
|
|
}
|
|
|
|
mutex_unlock(&cpufreq_fast_switch_lock);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_enable_fast_switch);
|
|
|
|
|
2016-04-08 05:38:46 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_disable_fast_switch - Disable fast frequency switching for policy.
|
|
|
|
* @policy: cpufreq policy to disable fast frequency switching for.
|
|
|
|
*/
|
|
|
|
void cpufreq_disable_fast_switch(struct cpufreq_policy *policy)
|
2016-03-30 09:47:49 +08:00
|
|
|
{
|
|
|
|
mutex_lock(&cpufreq_fast_switch_lock);
|
|
|
|
if (policy->fast_switch_enabled) {
|
|
|
|
policy->fast_switch_enabled = false;
|
|
|
|
if (!WARN_ON(cpufreq_fast_switch_count <= 0))
|
|
|
|
cpufreq_fast_switch_count--;
|
|
|
|
}
|
|
|
|
mutex_unlock(&cpufreq_fast_switch_lock);
|
|
|
|
}
|
2016-04-08 05:38:46 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_disable_fast_switch);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2021-06-29 14:27:07 +08:00
|
|
|
static unsigned int __resolve_freq(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq, unsigned int relation)
|
2016-07-14 04:25:25 +08:00
|
|
|
{
|
2021-06-29 14:27:08 +08:00
|
|
|
unsigned int idx;
|
2016-07-22 05:39:26 +08:00
|
|
|
|
2021-06-29 14:27:08 +08:00
|
|
|
target_freq = clamp_val(target_freq, policy->min, policy->max);
|
2016-07-22 05:39:26 +08:00
|
|
|
|
2022-08-16 20:01:57 +08:00
|
|
|
if (!policy->freq_table)
|
2021-06-29 14:27:08 +08:00
|
|
|
return target_freq;
|
2016-07-22 05:39:26 +08:00
|
|
|
|
2021-06-29 14:27:08 +08:00
|
|
|
idx = cpufreq_frequency_table_target(policy, target_freq, relation);
|
|
|
|
policy->cached_resolved_idx = idx;
|
|
|
|
policy->cached_target_freq = target_freq;
|
|
|
|
return policy->freq_table[idx].frequency;
|
2016-07-14 04:25:25 +08:00
|
|
|
}
|
2021-06-29 14:27:07 +08:00
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_driver_resolve_freq - Map a target frequency to a driver-supported
|
|
|
|
* one.
|
|
|
|
* @policy: associated policy to interrogate
|
|
|
|
* @target_freq: target frequency to resolve.
|
|
|
|
*
|
|
|
|
* The target to driver frequency mapping is cached in the policy.
|
|
|
|
*
|
|
|
|
* Return: Lowest driver-supported frequency greater than or equal to the
|
|
|
|
* given target_freq, subject to policy (min/max) and driver limitations.
|
|
|
|
*/
|
|
|
|
unsigned int cpufreq_driver_resolve_freq(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq)
|
|
|
|
{
|
2021-09-08 22:05:29 +08:00
|
|
|
return __resolve_freq(policy, target_freq, CPUFREQ_RELATION_LE);
|
2021-06-29 14:27:07 +08:00
|
|
|
}
|
2016-07-22 10:24:38 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_driver_resolve_freq);
|
2016-07-14 04:25:25 +08:00
|
|
|
|
2017-07-19 18:12:42 +08:00
|
|
|
unsigned int cpufreq_policy_transition_delay_us(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
unsigned int latency;
|
|
|
|
|
|
|
|
if (policy->transition_delay_us)
|
|
|
|
return policy->transition_delay_us;
|
|
|
|
|
|
|
|
latency = policy->cpuinfo.transition_latency / NSEC_PER_USEC;
|
2017-08-17 11:42:27 +08:00
|
|
|
if (latency) {
|
|
|
|
/*
|
|
|
|
* For platforms that can change the frequency very fast (< 10
|
|
|
|
* us), the above formula gives a decent transition delay. But
|
|
|
|
* for platforms where transition_latency is in milliseconds, it
|
|
|
|
* ends up giving unrealistic values.
|
|
|
|
*
|
|
|
|
* Cap the default transition delay to 10 ms, which seems to be
|
|
|
|
* a reasonable amount of time after which we should reevaluate
|
|
|
|
* the frequency.
|
|
|
|
*/
|
|
|
|
return min(latency * LATENCY_MULTIPLIER, (unsigned int)10000);
|
|
|
|
}
|
2017-07-19 18:12:42 +08:00
|
|
|
|
|
|
|
return LATENCY_MULTIPLIER;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_policy_transition_delay_us);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*********************************************************************
|
|
|
|
* SYSFS INTERFACE *
|
|
|
|
*********************************************************************/
|
2014-02-27 00:42:42 +08:00
|
|
|
static ssize_t show_boost(struct kobject *kobj,
|
2019-01-25 15:23:07 +08:00
|
|
|
struct kobj_attribute *attr, char *buf)
|
2013-12-20 22:24:49 +08:00
|
|
|
{
|
|
|
|
return sprintf(buf, "%d\n", cpufreq_driver->boost_enabled);
|
|
|
|
}
|
|
|
|
|
2019-01-25 15:23:07 +08:00
|
|
|
static ssize_t store_boost(struct kobject *kobj, struct kobj_attribute *attr,
|
|
|
|
const char *buf, size_t count)
|
2013-12-20 22:24:49 +08:00
|
|
|
{
|
|
|
|
int ret, enable;
|
|
|
|
|
|
|
|
ret = sscanf(buf, "%d", &enable);
|
|
|
|
if (ret != 1 || enable < 0 || enable > 1)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (cpufreq_boost_trigger_state(enable)) {
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_err("%s: Cannot %s BOOST!\n",
|
|
|
|
__func__, enable ? "enable" : "disable");
|
2013-12-20 22:24:49 +08:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("%s: cpufreq BOOST %s\n",
|
|
|
|
__func__, enable ? "enabled" : "disabled");
|
2013-12-20 22:24:49 +08:00
|
|
|
|
|
|
|
return count;
|
|
|
|
}
|
|
|
|
define_one_global_rw(boost);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-01-02 15:04:26 +08:00
|
|
|
static struct cpufreq_governor *find_governor(const char *str_governor)
|
2006-07-07 03:30:26 +08:00
|
|
|
{
|
|
|
|
struct cpufreq_governor *t;
|
|
|
|
|
2015-01-27 16:36:09 +08:00
|
|
|
for_each_governor(t)
|
2014-09-29 21:50:11 +08:00
|
|
|
if (!strncasecmp(str_governor, t->name, CPUFREQ_NAME_LEN))
|
2006-07-07 03:30:26 +08:00
|
|
|
return t;
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2020-06-29 16:24:58 +08:00
|
|
|
static struct cpufreq_governor *get_governor(const char *str_governor)
|
|
|
|
{
|
|
|
|
struct cpufreq_governor *t;
|
|
|
|
|
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
|
|
|
t = find_governor(str_governor);
|
|
|
|
if (!t)
|
|
|
|
goto unlock;
|
|
|
|
|
|
|
|
if (!try_module_get(t->owner))
|
|
|
|
t = NULL;
|
|
|
|
|
|
|
|
unlock:
|
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
|
|
|
|
|
|
|
return t;
|
|
|
|
}
|
|
|
|
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
static unsigned int cpufreq_parse_policy(char *str_governor)
|
2019-04-29 15:24:18 +08:00
|
|
|
{
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
if (!strncasecmp(str_governor, "performance", CPUFREQ_NAME_LEN))
|
|
|
|
return CPUFREQ_POLICY_PERFORMANCE;
|
|
|
|
|
|
|
|
if (!strncasecmp(str_governor, "powersave", CPUFREQ_NAME_LEN))
|
|
|
|
return CPUFREQ_POLICY_POWERSAVE;
|
|
|
|
|
|
|
|
return CPUFREQ_POLICY_UNKNOWN;
|
2019-04-29 15:24:18 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2019-06-20 11:05:48 +08:00
|
|
|
* cpufreq_parse_governor - parse a governor string only for has_target()
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
* @str_governor: Governor name.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
static struct cpufreq_governor *cpufreq_parse_governor(char *str_governor)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2019-04-29 15:24:18 +08:00
|
|
|
struct cpufreq_governor *t;
|
2017-11-23 08:23:16 +08:00
|
|
|
|
2020-06-29 16:24:58 +08:00
|
|
|
t = get_governor(str_governor);
|
|
|
|
if (t)
|
|
|
|
return t;
|
2006-07-07 03:32:01 +08:00
|
|
|
|
2020-06-29 16:24:58 +08:00
|
|
|
if (request_module("cpufreq_%s", str_governor))
|
|
|
|
return NULL;
|
2017-11-23 08:23:16 +08:00
|
|
|
|
2020-06-29 16:24:58 +08:00
|
|
|
return get_governor(str_governor);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2006-10-26 18:50:58 +08:00
|
|
|
* cpufreq_per_cpu_attr_read() / show_##file_name() -
|
|
|
|
* print out cpufreq information
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* Write out information from cpufreq_driver->policy[cpu]; object must be
|
|
|
|
* "unsigned int".
|
|
|
|
*/
|
|
|
|
|
2006-02-28 13:43:23 +08:00
|
|
|
#define show_one(file_name, object) \
|
|
|
|
static ssize_t show_##file_name \
|
2008-03-06 03:28:32 +08:00
|
|
|
(struct cpufreq_policy *policy, char *buf) \
|
2006-02-28 13:43:23 +08:00
|
|
|
{ \
|
2009-01-18 14:37:11 +08:00
|
|
|
return sprintf(buf, "%u\n", policy->object); \
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
show_one(cpuinfo_min_freq, cpuinfo.min_freq);
|
|
|
|
show_one(cpuinfo_max_freq, cpuinfo.max_freq);
|
2009-02-04 08:17:41 +08:00
|
|
|
show_one(cpuinfo_transition_latency, cpuinfo.transition_latency);
|
2005-04-17 06:20:36 +08:00
|
|
|
show_one(scaling_min_freq, min);
|
|
|
|
show_one(scaling_max_freq, max);
|
2014-10-13 23:37:40 +08:00
|
|
|
|
2017-06-24 13:11:52 +08:00
|
|
|
__weak unsigned int arch_freq_get_on_cpu(int cpu)
|
|
|
|
{
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2015-01-02 15:04:24 +08:00
|
|
|
static ssize_t show_scaling_cur_freq(struct cpufreq_policy *policy, char *buf)
|
2014-10-13 23:37:40 +08:00
|
|
|
{
|
|
|
|
ssize_t ret;
|
2017-06-24 13:11:52 +08:00
|
|
|
unsigned int freq;
|
2014-10-13 23:37:40 +08:00
|
|
|
|
2017-06-24 13:11:52 +08:00
|
|
|
freq = arch_freq_get_on_cpu(policy->cpu);
|
|
|
|
if (freq)
|
|
|
|
ret = sprintf(buf, "%u\n", freq);
|
2020-08-27 13:24:15 +08:00
|
|
|
else if (cpufreq_driver->setpolicy && cpufreq_driver->get)
|
2014-10-13 23:37:40 +08:00
|
|
|
ret = sprintf(buf, "%u\n", cpufreq_driver->get(policy->cpu));
|
|
|
|
else
|
|
|
|
ret = sprintf(buf, "%u\n", policy->cur);
|
|
|
|
return ret;
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2005-04-17 06:20:36 +08:00
|
|
|
* cpufreq_per_cpu_attr_write() / store_##file_name() - sysfs write access
|
|
|
|
*/
|
|
|
|
#define store_one(file_name, object) \
|
|
|
|
static ssize_t store_##file_name \
|
2008-03-06 03:28:32 +08:00
|
|
|
(struct cpufreq_policy *policy, const char *buf, size_t count) \
|
2005-04-17 06:20:36 +08:00
|
|
|
{ \
|
2019-07-05 18:51:24 +08:00
|
|
|
unsigned long val; \
|
|
|
|
int ret; \
|
2005-04-17 06:20:36 +08:00
|
|
|
\
|
2019-07-05 18:51:24 +08:00
|
|
|
ret = sscanf(buf, "%lu", &val); \
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret != 1) \
|
|
|
|
return -EINVAL; \
|
|
|
|
\
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
ret = freq_qos_update_request(policy->object##_freq_req, val);\
|
2019-07-05 18:51:24 +08:00
|
|
|
return ret >= 0 ? count : ret; \
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2009-01-18 14:37:11 +08:00
|
|
|
store_one(scaling_min_freq, min);
|
|
|
|
store_one(scaling_max_freq, max);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2005-04-17 06:20:36 +08:00
|
|
|
* show_cpuinfo_cur_freq - current CPU frequency as detected by hardware
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t show_cpuinfo_cur_freq(struct cpufreq_policy *policy,
|
|
|
|
char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2015-01-02 15:04:29 +08:00
|
|
|
unsigned int cur_freq = __cpufreq_get(policy);
|
2017-03-15 07:12:16 +08:00
|
|
|
|
|
|
|
if (cur_freq)
|
|
|
|
return sprintf(buf, "%u\n", cur_freq);
|
|
|
|
|
|
|
|
return sprintf(buf, "<unknown>\n");
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2005-04-17 06:20:36 +08:00
|
|
|
* show_scaling_governor - show the current policy for the specified CPU
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t show_scaling_governor(struct cpufreq_policy *policy, char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2009-01-18 14:37:11 +08:00
|
|
|
if (policy->policy == CPUFREQ_POLICY_POWERSAVE)
|
2005-04-17 06:20:36 +08:00
|
|
|
return sprintf(buf, "powersave\n");
|
|
|
|
else if (policy->policy == CPUFREQ_POLICY_PERFORMANCE)
|
|
|
|
return sprintf(buf, "performance\n");
|
|
|
|
else if (policy->governor)
|
2012-10-23 07:23:43 +08:00
|
|
|
return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n",
|
2009-01-18 14:37:11 +08:00
|
|
|
policy->governor->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2005-04-17 06:20:36 +08:00
|
|
|
* store_scaling_governor - store policy for the specified CPU
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t store_scaling_governor(struct cpufreq_policy *policy,
|
|
|
|
const char *buf, size_t count)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
char str_governor[16];
|
2013-09-07 03:54:06 +08:00
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2009-01-18 14:37:11 +08:00
|
|
|
ret = sscanf(buf, "%15s", str_governor);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret != 1)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2019-04-29 15:24:18 +08:00
|
|
|
if (cpufreq_driver->setpolicy) {
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
unsigned int new_pol;
|
|
|
|
|
|
|
|
new_pol = cpufreq_parse_policy(str_governor);
|
|
|
|
if (!new_pol)
|
2019-04-29 15:24:18 +08:00
|
|
|
return -EINVAL;
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
|
|
|
|
ret = cpufreq_set_policy(policy, NULL, new_pol);
|
2019-04-29 15:24:18 +08:00
|
|
|
} else {
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
struct cpufreq_governor *new_gov;
|
|
|
|
|
|
|
|
new_gov = cpufreq_parse_governor(str_governor);
|
|
|
|
if (!new_gov)
|
2019-04-29 15:24:18 +08:00
|
|
|
return -EINVAL;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
ret = cpufreq_set_policy(policy, new_gov,
|
|
|
|
CPUFREQ_POLICY_UNKNOWN);
|
2017-11-23 21:27:07 +08:00
|
|
|
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
module_put(new_gov->owner);
|
|
|
|
}
|
2017-11-23 21:27:07 +08:00
|
|
|
|
2015-08-03 11:06:18 +08:00
|
|
|
return ret ? ret : count;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2005-04-17 06:20:36 +08:00
|
|
|
* show_scaling_driver - show the cpufreq driver currently loaded
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t show_scaling_driver(struct cpufreq_policy *policy, char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-04-29 06:08:16 +08:00
|
|
|
return scnprintf(buf, CPUFREQ_NAME_PLEN, "%s\n", cpufreq_driver->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2005-04-17 06:20:36 +08:00
|
|
|
* show_scaling_available_governors - show the available CPUfreq governors
|
|
|
|
*/
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t show_scaling_available_governors(struct cpufreq_policy *policy,
|
|
|
|
char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
ssize_t i = 0;
|
|
|
|
struct cpufreq_governor *t;
|
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
if (!has_target()) {
|
2005-04-17 06:20:36 +08:00
|
|
|
i += sprintf(buf, "performance powersave");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2020-06-29 16:24:58 +08:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2015-01-27 16:36:09 +08:00
|
|
|
for_each_governor(t) {
|
2009-01-18 14:37:11 +08:00
|
|
|
if (i >= (ssize_t) ((PAGE_SIZE / sizeof(char))
|
|
|
|
- (CPUFREQ_NAME_LEN + 2)))
|
2020-06-29 16:24:58 +08:00
|
|
|
break;
|
2012-10-23 07:23:43 +08:00
|
|
|
i += scnprintf(&buf[i], CPUFREQ_NAME_PLEN, "%s ", t->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2020-06-29 16:24:58 +08:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2006-02-03 06:03:42 +08:00
|
|
|
out:
|
2005-04-17 06:20:36 +08:00
|
|
|
i += sprintf(&buf[i], "\n");
|
|
|
|
return i;
|
|
|
|
}
|
2008-04-19 04:31:12 +08:00
|
|
|
|
2013-06-27 15:08:54 +08:00
|
|
|
ssize_t cpufreq_show_cpus(const struct cpumask *mask, char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
ssize_t i = 0;
|
|
|
|
unsigned int cpu;
|
|
|
|
|
2009-01-04 21:18:06 +08:00
|
|
|
for_each_cpu(cpu, mask) {
|
2022-05-26 19:51:19 +08:00
|
|
|
i += scnprintf(&buf[i], (PAGE_SIZE - i - 2), "%u ", cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (i >= (PAGE_SIZE - 5))
|
2009-01-18 14:37:11 +08:00
|
|
|
break;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2022-05-26 19:51:19 +08:00
|
|
|
|
|
|
|
/* Remove the extra space at the end */
|
|
|
|
i--;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
i += sprintf(&buf[i], "\n");
|
|
|
|
return i;
|
|
|
|
}
|
2013-06-27 15:08:54 +08:00
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_show_cpus);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2008-04-19 04:31:12 +08:00
|
|
|
* show_related_cpus - show the CPUs affected by each transition even if
|
|
|
|
* hw coordination is in use
|
|
|
|
*/
|
|
|
|
static ssize_t show_related_cpus(struct cpufreq_policy *policy, char *buf)
|
|
|
|
{
|
2013-06-27 15:08:54 +08:00
|
|
|
return cpufreq_show_cpus(policy->related_cpus, buf);
|
2008-04-19 04:31:12 +08:00
|
|
|
}
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2008-04-19 04:31:12 +08:00
|
|
|
* show_affected_cpus - show the CPUs affected by each transition
|
|
|
|
*/
|
|
|
|
static ssize_t show_affected_cpus(struct cpufreq_policy *policy, char *buf)
|
|
|
|
{
|
2013-06-27 15:08:54 +08:00
|
|
|
return cpufreq_show_cpus(policy->cpus, buf);
|
2008-04-19 04:31:12 +08:00
|
|
|
}
|
|
|
|
|
2007-10-27 01:18:21 +08:00
|
|
|
static ssize_t store_scaling_setspeed(struct cpufreq_policy *policy,
|
2008-03-06 03:28:32 +08:00
|
|
|
const char *buf, size_t count)
|
2007-10-27 01:18:21 +08:00
|
|
|
{
|
|
|
|
unsigned int freq = 0;
|
|
|
|
unsigned int ret;
|
|
|
|
|
2008-06-06 13:46:33 +08:00
|
|
|
if (!policy->governor || !policy->governor->store_setspeed)
|
2007-10-27 01:18:21 +08:00
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
ret = sscanf(buf, "%u", &freq);
|
|
|
|
if (ret != 1)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
policy->governor->store_setspeed(policy, freq);
|
|
|
|
|
|
|
|
return count;
|
|
|
|
}
|
|
|
|
|
|
|
|
static ssize_t show_scaling_setspeed(struct cpufreq_policy *policy, char *buf)
|
|
|
|
{
|
2008-06-06 13:46:33 +08:00
|
|
|
if (!policy->governor || !policy->governor->show_setspeed)
|
2007-10-27 01:18:21 +08:00
|
|
|
return sprintf(buf, "<unsupported>\n");
|
|
|
|
|
|
|
|
return policy->governor->show_setspeed(policy, buf);
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2012-10-23 07:23:33 +08:00
|
|
|
* show_bios_limit - show the current cpufreq HW/BIOS limitation
|
2009-11-19 19:31:01 +08:00
|
|
|
*/
|
|
|
|
static ssize_t show_bios_limit(struct cpufreq_policy *policy, char *buf)
|
|
|
|
{
|
|
|
|
unsigned int limit;
|
|
|
|
int ret;
|
2019-04-16 10:40:27 +08:00
|
|
|
ret = cpufreq_driver->bios_limit(policy->cpu, &limit);
|
|
|
|
if (!ret)
|
|
|
|
return sprintf(buf, "%u\n", limit);
|
2009-11-19 19:31:01 +08:00
|
|
|
return sprintf(buf, "%u\n", policy->cpuinfo.max_freq);
|
|
|
|
}
|
|
|
|
|
2010-04-01 03:56:46 +08:00
|
|
|
cpufreq_freq_attr_ro_perm(cpuinfo_cur_freq, 0400);
|
|
|
|
cpufreq_freq_attr_ro(cpuinfo_min_freq);
|
|
|
|
cpufreq_freq_attr_ro(cpuinfo_max_freq);
|
|
|
|
cpufreq_freq_attr_ro(cpuinfo_transition_latency);
|
|
|
|
cpufreq_freq_attr_ro(scaling_available_governors);
|
|
|
|
cpufreq_freq_attr_ro(scaling_driver);
|
|
|
|
cpufreq_freq_attr_ro(scaling_cur_freq);
|
|
|
|
cpufreq_freq_attr_ro(bios_limit);
|
|
|
|
cpufreq_freq_attr_ro(related_cpus);
|
|
|
|
cpufreq_freq_attr_ro(affected_cpus);
|
|
|
|
cpufreq_freq_attr_rw(scaling_min_freq);
|
|
|
|
cpufreq_freq_attr_rw(scaling_max_freq);
|
|
|
|
cpufreq_freq_attr_rw(scaling_governor);
|
|
|
|
cpufreq_freq_attr_rw(scaling_setspeed);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2021-12-28 21:19:12 +08:00
|
|
|
static struct attribute *cpufreq_attrs[] = {
|
2005-04-17 06:20:36 +08:00
|
|
|
&cpuinfo_min_freq.attr,
|
|
|
|
&cpuinfo_max_freq.attr,
|
2009-02-04 08:17:41 +08:00
|
|
|
&cpuinfo_transition_latency.attr,
|
2005-04-17 06:20:36 +08:00
|
|
|
&scaling_min_freq.attr,
|
|
|
|
&scaling_max_freq.attr,
|
|
|
|
&affected_cpus.attr,
|
2008-04-19 04:31:12 +08:00
|
|
|
&related_cpus.attr,
|
2005-04-17 06:20:36 +08:00
|
|
|
&scaling_governor.attr,
|
|
|
|
&scaling_driver.attr,
|
|
|
|
&scaling_available_governors.attr,
|
2007-10-27 01:18:21 +08:00
|
|
|
&scaling_setspeed.attr,
|
2005-04-17 06:20:36 +08:00
|
|
|
NULL
|
|
|
|
};
|
2021-12-28 21:19:12 +08:00
|
|
|
ATTRIBUTE_GROUPS(cpufreq);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2009-01-18 14:37:11 +08:00
|
|
|
#define to_policy(k) container_of(k, struct cpufreq_policy, kobj)
|
|
|
|
#define to_attr(a) container_of(a, struct freq_attr, attr)
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2009-01-18 14:37:11 +08:00
|
|
|
static ssize_t show(struct kobject *kobj, struct attribute *attr, char *buf)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-03-06 03:28:32 +08:00
|
|
|
struct cpufreq_policy *policy = to_policy(kobj);
|
|
|
|
struct freq_attr *fattr = to_attr(attr);
|
2022-05-16 11:02:50 +08:00
|
|
|
ssize_t ret = -EBUSY;
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2019-11-07 13:08:17 +08:00
|
|
|
if (!fattr->show)
|
|
|
|
return -EIO;
|
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_read(&policy->rwsem);
|
2022-05-16 11:02:50 +08:00
|
|
|
if (likely(!policy_is_inactive(policy)))
|
|
|
|
ret = fattr->show(policy, buf);
|
2013-10-18 21:40:15 +08:00
|
|
|
up_read(&policy->rwsem);
|
2013-10-02 16:43:09 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-03-06 03:28:32 +08:00
|
|
|
static ssize_t store(struct kobject *kobj, struct attribute *attr,
|
|
|
|
const char *buf, size_t count)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-03-06 03:28:32 +08:00
|
|
|
struct cpufreq_policy *policy = to_policy(kobj);
|
|
|
|
struct freq_attr *fattr = to_attr(attr);
|
2022-05-16 11:02:50 +08:00
|
|
|
ssize_t ret = -EBUSY;
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2019-11-07 13:08:17 +08:00
|
|
|
if (!fattr->store)
|
|
|
|
return -EIO;
|
|
|
|
|
2022-05-26 19:51:21 +08:00
|
|
|
down_write(&policy->rwsem);
|
|
|
|
if (likely(!policy_is_inactive(policy)))
|
|
|
|
ret = fattr->store(policy, buf, count);
|
|
|
|
up_write(&policy->rwsem);
|
2013-09-07 03:53:43 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2008-03-06 03:28:32 +08:00
|
|
|
static void cpufreq_sysfs_release(struct kobject *kobj)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2008-03-06 03:28:32 +08:00
|
|
|
struct cpufreq_policy *policy = to_policy(kobj);
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("last reference is dropped\n");
|
2005-04-17 06:20:36 +08:00
|
|
|
complete(&policy->kobj_unregister);
|
|
|
|
}
|
|
|
|
|
2010-01-19 09:58:23 +08:00
|
|
|
static const struct sysfs_ops sysfs_ops = {
|
2005-04-17 06:20:36 +08:00
|
|
|
.show = show,
|
|
|
|
.store = store,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct kobj_type ktype_cpufreq = {
|
|
|
|
.sysfs_ops = &sysfs_ops,
|
2021-12-28 21:19:12 +08:00
|
|
|
.default_groups = cpufreq_groups,
|
2005-04-17 06:20:36 +08:00
|
|
|
.release = cpufreq_sysfs_release,
|
|
|
|
};
|
|
|
|
|
2021-11-29 16:02:48 +08:00
|
|
|
static void add_cpu_dev_symlink(struct cpufreq_policy *policy, unsigned int cpu,
|
|
|
|
struct device *dev)
|
2015-06-10 08:13:21 +08:00
|
|
|
{
|
2019-07-08 18:57:52 +08:00
|
|
|
if (unlikely(!dev))
|
2017-03-28 01:33:09 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
if (cpumask_test_and_set_cpu(cpu, policy->real_cpus))
|
|
|
|
return;
|
|
|
|
|
cpufreq: create link to policy only for registered CPUs
If a cpufreq driver is registered very early in the boot stage (e.g.
registered from postcore_initcall()), then cpufreq core may generate
kernel warnings for it.
In this case, the CPUs are brought online, then the cpufreq driver is
registered, and then the CPU topology devices are registered. However,
by the time cpufreq_add_dev() gets called, the cpu device isn't stored
in the per-cpu variable (cpu_sys_devices,) which is read by
get_cpu_device().
So the cpufreq core fails to get device for the CPU, for which
cpufreq_add_dev() was called in the first place and we will hit a
WARN_ON(!cpu_dev).
Even if we reuse the 'dev' parameter passed to cpufreq_add_dev() to
avoid that warning, there might be other CPUs online that share the
policy with the cpu for which cpufreq_add_dev() is called. Eventually
get_cpu_device() will return NULL for them as well, and we will hit the
same WARN_ON() again.
In order to fix these issues, change cpufreq core to create links to the
policy for a cpu only when cpufreq_add_dev() is called for that CPU.
Reuse the 'real_cpus' mask to track that as well.
Note that cpufreq_remove_dev() already handles removal of the links for
individual CPUs and cpufreq_add_dev() has aligned with that now.
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2016-09-12 14:37:05 +08:00
|
|
|
dev_dbg(dev, "%s: Adding symlink\n", __func__);
|
2017-03-28 01:33:09 +08:00
|
|
|
if (sysfs_create_link(&dev->kobj, &policy->kobj, "cpufreq"))
|
|
|
|
dev_err(dev, "cpufreq symlink creation failed\n");
|
2015-06-10 08:13:21 +08:00
|
|
|
}
|
|
|
|
|
2022-05-11 17:06:24 +08:00
|
|
|
static void remove_cpu_dev_symlink(struct cpufreq_policy *policy, int cpu,
|
cpufreq: create link to policy only for registered CPUs
If a cpufreq driver is registered very early in the boot stage (e.g.
registered from postcore_initcall()), then cpufreq core may generate
kernel warnings for it.
In this case, the CPUs are brought online, then the cpufreq driver is
registered, and then the CPU topology devices are registered. However,
by the time cpufreq_add_dev() gets called, the cpu device isn't stored
in the per-cpu variable (cpu_sys_devices,) which is read by
get_cpu_device().
So the cpufreq core fails to get device for the CPU, for which
cpufreq_add_dev() was called in the first place and we will hit a
WARN_ON(!cpu_dev).
Even if we reuse the 'dev' parameter passed to cpufreq_add_dev() to
avoid that warning, there might be other CPUs online that share the
policy with the cpu for which cpufreq_add_dev() is called. Eventually
get_cpu_device() will return NULL for them as well, and we will hit the
same WARN_ON() again.
In order to fix these issues, change cpufreq core to create links to the
policy for a cpu only when cpufreq_add_dev() is called for that CPU.
Reuse the 'real_cpus' mask to track that as well.
Note that cpufreq_remove_dev() already handles removal of the links for
individual CPUs and cpufreq_add_dev() has aligned with that now.
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2016-09-12 14:37:05 +08:00
|
|
|
struct device *dev)
|
2015-06-10 08:13:21 +08:00
|
|
|
{
|
cpufreq: create link to policy only for registered CPUs
If a cpufreq driver is registered very early in the boot stage (e.g.
registered from postcore_initcall()), then cpufreq core may generate
kernel warnings for it.
In this case, the CPUs are brought online, then the cpufreq driver is
registered, and then the CPU topology devices are registered. However,
by the time cpufreq_add_dev() gets called, the cpu device isn't stored
in the per-cpu variable (cpu_sys_devices,) which is read by
get_cpu_device().
So the cpufreq core fails to get device for the CPU, for which
cpufreq_add_dev() was called in the first place and we will hit a
WARN_ON(!cpu_dev).
Even if we reuse the 'dev' parameter passed to cpufreq_add_dev() to
avoid that warning, there might be other CPUs online that share the
policy with the cpu for which cpufreq_add_dev() is called. Eventually
get_cpu_device() will return NULL for them as well, and we will hit the
same WARN_ON() again.
In order to fix these issues, change cpufreq core to create links to the
policy for a cpu only when cpufreq_add_dev() is called for that CPU.
Reuse the 'real_cpus' mask to track that as well.
Note that cpufreq_remove_dev() already handles removal of the links for
individual CPUs and cpufreq_add_dev() has aligned with that now.
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2016-09-12 14:37:05 +08:00
|
|
|
dev_dbg(dev, "%s: Removing symlink\n", __func__);
|
|
|
|
sysfs_remove_link(&dev->kobj, "cpufreq");
|
2022-05-11 17:06:24 +08:00
|
|
|
cpumask_clear_cpu(cpu, policy->real_cpus);
|
2015-06-10 08:13:21 +08:00
|
|
|
}
|
|
|
|
|
2015-07-28 05:11:37 +08:00
|
|
|
static int cpufreq_add_dev_interface(struct cpufreq_policy *policy)
|
2009-07-09 06:05:42 +08:00
|
|
|
{
|
|
|
|
struct freq_attr **drv_attr;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
/* set up files for this cpu device */
|
2013-04-29 06:08:16 +08:00
|
|
|
drv_attr = cpufreq_driver->attr;
|
2015-01-02 15:04:23 +08:00
|
|
|
while (drv_attr && *drv_attr) {
|
2009-07-09 06:05:42 +08:00
|
|
|
ret = sysfs_create_file(&policy->kobj, &((*drv_attr)->attr));
|
|
|
|
if (ret)
|
2014-11-24 17:08:03 +08:00
|
|
|
return ret;
|
2009-07-09 06:05:42 +08:00
|
|
|
drv_attr++;
|
|
|
|
}
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->get) {
|
2009-07-09 06:05:42 +08:00
|
|
|
ret = sysfs_create_file(&policy->kobj, &cpuinfo_cur_freq.attr);
|
|
|
|
if (ret)
|
2014-11-24 17:08:03 +08:00
|
|
|
return ret;
|
2009-07-09 06:05:42 +08:00
|
|
|
}
|
2014-10-13 23:37:40 +08:00
|
|
|
|
|
|
|
ret = sysfs_create_file(&policy->kobj, &scaling_cur_freq.attr);
|
|
|
|
if (ret)
|
2014-11-24 17:08:03 +08:00
|
|
|
return ret;
|
2014-10-13 23:37:40 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->bios_limit) {
|
2009-11-19 19:31:01 +08:00
|
|
|
ret = sysfs_create_file(&policy->kobj, &bios_limit.attr);
|
|
|
|
if (ret)
|
2014-11-24 17:08:03 +08:00
|
|
|
return ret;
|
2009-11-19 19:31:01 +08:00
|
|
|
}
|
2009-07-09 06:05:42 +08:00
|
|
|
|
cpufreq: create link to policy only for registered CPUs
If a cpufreq driver is registered very early in the boot stage (e.g.
registered from postcore_initcall()), then cpufreq core may generate
kernel warnings for it.
In this case, the CPUs are brought online, then the cpufreq driver is
registered, and then the CPU topology devices are registered. However,
by the time cpufreq_add_dev() gets called, the cpu device isn't stored
in the per-cpu variable (cpu_sys_devices,) which is read by
get_cpu_device().
So the cpufreq core fails to get device for the CPU, for which
cpufreq_add_dev() was called in the first place and we will hit a
WARN_ON(!cpu_dev).
Even if we reuse the 'dev' parameter passed to cpufreq_add_dev() to
avoid that warning, there might be other CPUs online that share the
policy with the cpu for which cpufreq_add_dev() is called. Eventually
get_cpu_device() will return NULL for them as well, and we will hit the
same WARN_ON() again.
In order to fix these issues, change cpufreq core to create links to the
policy for a cpu only when cpufreq_add_dev() is called for that CPU.
Reuse the 'real_cpus' mask to track that as well.
Note that cpufreq_remove_dev() already handles removal of the links for
individual CPUs and cpufreq_add_dev() has aligned with that now.
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2016-09-12 14:37:05 +08:00
|
|
|
return 0;
|
2013-07-30 06:54:23 +08:00
|
|
|
}
|
|
|
|
|
2015-07-08 17:42:16 +08:00
|
|
|
static int cpufreq_init_policy(struct cpufreq_policy *policy)
|
2013-07-30 06:54:23 +08:00
|
|
|
{
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
struct cpufreq_governor *gov = NULL;
|
|
|
|
unsigned int pol = CPUFREQ_POLICY_UNKNOWN;
|
2020-06-29 16:24:58 +08:00
|
|
|
int ret;
|
2019-04-29 15:24:18 +08:00
|
|
|
|
|
|
|
if (has_target()) {
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
/* Update policy governor to the one used before hotplug. */
|
2020-06-29 16:24:58 +08:00
|
|
|
gov = get_governor(policy->last_governor);
|
2019-04-29 15:24:18 +08:00
|
|
|
if (gov) {
|
|
|
|
pr_debug("Restoring governor %s for cpu %d\n",
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
gov->name, policy->cpu);
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
} else {
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
gov = get_governor(default_governor);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!gov) {
|
|
|
|
gov = cpufreq_default_governor();
|
|
|
|
__module_get(gov->owner);
|
2019-04-29 15:24:18 +08:00
|
|
|
}
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
|
2016-02-05 09:37:42 +08:00
|
|
|
} else {
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
|
2019-04-29 15:24:18 +08:00
|
|
|
/* Use the default policy if there is no last_policy. */
|
|
|
|
if (policy->last_policy) {
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
pol = policy->last_policy;
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
} else {
|
|
|
|
pol = cpufreq_parse_policy(default_governor);
|
2020-02-27 05:39:27 +08:00
|
|
|
/*
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
* In case the default governor is neither "performance"
|
2020-02-27 05:39:27 +08:00
|
|
|
* nor "powersave", fall back to the initial policy
|
|
|
|
* value set by the driver.
|
|
|
|
*/
|
|
|
|
if (pol == CPUFREQ_POLICY_UNKNOWN)
|
|
|
|
pol = policy->policy;
|
2019-04-29 15:24:18 +08:00
|
|
|
}
|
2020-02-27 05:39:27 +08:00
|
|
|
if (pol != CPUFREQ_POLICY_PERFORMANCE &&
|
|
|
|
pol != CPUFREQ_POLICY_POWERSAVE)
|
|
|
|
return -ENODATA;
|
2015-12-02 08:52:14 +08:00
|
|
|
}
|
2019-04-29 15:24:18 +08:00
|
|
|
|
2020-06-29 16:24:58 +08:00
|
|
|
ret = cpufreq_set_policy(policy, gov, pol);
|
|
|
|
if (gov)
|
|
|
|
module_put(gov->owner);
|
|
|
|
|
|
|
|
return ret;
|
2009-07-09 06:05:42 +08:00
|
|
|
}
|
|
|
|
|
2015-07-28 05:11:37 +08:00
|
|
|
static int cpufreq_add_policy_cpu(struct cpufreq_policy *policy, unsigned int cpu)
|
2013-01-29 22:39:08 +08:00
|
|
|
{
|
2013-10-25 22:15:48 +08:00
|
|
|
int ret = 0;
|
2013-01-29 22:39:08 +08:00
|
|
|
|
2015-02-19 19:32:06 +08:00
|
|
|
/* Has this CPU been taken care of already? */
|
|
|
|
if (cpumask_test_cpu(cpu, policy->cpus))
|
|
|
|
return 0;
|
|
|
|
|
2016-02-11 20:01:12 +08:00
|
|
|
down_write(&policy->rwsem);
|
2016-05-12 21:14:12 +08:00
|
|
|
if (has_target())
|
|
|
|
cpufreq_stop_governor(policy);
|
2013-01-29 22:39:08 +08:00
|
|
|
|
|
|
|
cpumask_set_cpu(cpu, policy->cpus);
|
2013-02-07 13:25:00 +08:00
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
if (has_target()) {
|
2016-03-21 22:45:24 +08:00
|
|
|
ret = cpufreq_start_governor(policy);
|
2016-02-11 20:01:12 +08:00
|
|
|
if (ret)
|
2013-08-07 01:23:13 +08:00
|
|
|
pr_err("%s: Failed to start governor\n", __func__);
|
2013-04-22 06:48:03 +08:00
|
|
|
}
|
2016-02-11 20:01:12 +08:00
|
|
|
up_write(&policy->rwsem);
|
|
|
|
return ret;
|
2013-01-29 22:39:08 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2019-07-04 15:36:22 +08:00
|
|
|
void refresh_frequency_limits(struct cpufreq_policy *policy)
|
2019-06-20 11:05:50 +08:00
|
|
|
{
|
2019-07-08 18:57:52 +08:00
|
|
|
if (!policy_is_inactive(policy)) {
|
|
|
|
pr_debug("updating policy for CPU %u\n", policy->cpu);
|
2019-06-20 11:05:50 +08:00
|
|
|
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
cpufreq_set_policy(policy, policy->governor, policy->policy);
|
2019-07-08 18:57:52 +08:00
|
|
|
}
|
2019-06-20 11:05:50 +08:00
|
|
|
}
|
2019-07-04 15:36:22 +08:00
|
|
|
EXPORT_SYMBOL(refresh_frequency_limits);
|
2019-06-20 11:05:50 +08:00
|
|
|
|
2016-02-22 19:06:42 +08:00
|
|
|
static void handle_update(struct work_struct *work)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy =
|
|
|
|
container_of(work, struct cpufreq_policy, update);
|
2019-06-20 11:05:50 +08:00
|
|
|
|
|
|
|
pr_debug("handle_update for cpu %u called\n", policy->cpu);
|
2019-07-08 18:57:52 +08:00
|
|
|
down_write(&policy->rwsem);
|
2019-06-20 11:05:50 +08:00
|
|
|
refresh_frequency_limits(policy);
|
2019-07-08 18:57:52 +08:00
|
|
|
up_write(&policy->rwsem);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpufreq_notifier_min(struct notifier_block *nb, unsigned long freq,
|
|
|
|
void *data)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_min);
|
|
|
|
|
|
|
|
schedule_work(&policy->update);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpufreq_notifier_max(struct notifier_block *nb, unsigned long freq,
|
|
|
|
void *data)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = container_of(nb, struct cpufreq_policy, nb_max);
|
|
|
|
|
|
|
|
schedule_work(&policy->update);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void cpufreq_policy_put_kobj(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
struct kobject *kobj;
|
|
|
|
struct completion *cmp;
|
|
|
|
|
|
|
|
down_write(&policy->rwsem);
|
|
|
|
cpufreq_stats_free_table(policy);
|
|
|
|
kobj = &policy->kobj;
|
|
|
|
cmp = &policy->kobj_unregister;
|
|
|
|
up_write(&policy->rwsem);
|
|
|
|
kobject_put(kobj);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We need to make sure that the underlying kobj is
|
|
|
|
* actually not referenced anymore by anybody before we
|
|
|
|
* proceed with unloading.
|
|
|
|
*/
|
|
|
|
pr_debug("waiting for dropping of refcount\n");
|
|
|
|
wait_for_completion(cmp);
|
|
|
|
pr_debug("wait complete\n");
|
2013-01-29 22:39:08 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-07-28 05:11:50 +08:00
|
|
|
static struct cpufreq_policy *cpufreq_policy_alloc(unsigned int cpu)
|
2013-07-30 06:54:11 +08:00
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy;
|
2019-07-08 18:57:52 +08:00
|
|
|
struct device *dev = get_cpu_device(cpu);
|
2016-03-03 17:21:33 +08:00
|
|
|
int ret;
|
2013-07-30 06:54:11 +08:00
|
|
|
|
2019-07-08 18:57:52 +08:00
|
|
|
if (!dev)
|
|
|
|
return NULL;
|
|
|
|
|
2013-07-30 06:54:11 +08:00
|
|
|
policy = kzalloc(sizeof(*policy), GFP_KERNEL);
|
|
|
|
if (!policy)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (!alloc_cpumask_var(&policy->cpus, GFP_KERNEL))
|
|
|
|
goto err_free_policy;
|
|
|
|
|
|
|
|
if (!zalloc_cpumask_var(&policy->related_cpus, GFP_KERNEL))
|
|
|
|
goto err_free_cpumask;
|
|
|
|
|
cpufreq: Avoid attempts to create duplicate symbolic links
After commit 87549141d516 (cpufreq: Stop migrating sysfs files on
hotplug) there is a problem with CPUs that share cpufreq policy
objects with other CPUs and are initially offline.
Say CPU1 shares a policy with CPU0 which is online and is registered
first. As part of the registration process, cpufreq_add_dev() is
called for it. It creates the policy object and a symbolic link
to it from the CPU1's sysfs directory. If CPU1 is registered
subsequently and it is offline at that time, cpufreq_add_dev() will
attempt to create a symbolic link to the policy object for it, but
that link is present already, so a warning about that will be
triggered.
To avoid that warning, make cpufreq use an additional CPU mask
containing related CPUs that are actually present for each policy
object. That mask is initialized when the policy object is populated
after its creation (for the first online CPU using it) and it includes
CPUs from the "policy CPUs" mask returned by the cpufreq driver's
->init() callback that are physically present at that time. Symbolic
links to the policy are created only for the CPUs in that mask.
If cpufreq_add_dev() is invoked for an offline CPU, it checks the
new mask and only creates the symlink if the CPU was not in it (the
CPU is added to the mask at the same time).
In turn, cpufreq_remove_dev() drops the given CPU from the new mask,
removes its symlink to the policy object and returns, unless it is
the CPU owning the policy object. In that case, the policy object
is moved to a new CPU's sysfs directory or deleted if the CPU being
removed was the last user of the policy.
While at it, notice that cpufreq_remove_dev() can't fail, because
its return value is ignored, so make it ignore return values from
__cpufreq_remove_dev_prepare() and __cpufreq_remove_dev_finish()
and prevent these functions from aborting on errors returned by
__cpufreq_governor(). Also drop the now unused sif argument from
them.
Fixes: 87549141d516 (cpufreq: Stop migrating sysfs files on hotplug)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reported-and-tested-by: Russell King <linux@arm.linux.org.uk>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2015-07-26 08:07:47 +08:00
|
|
|
if (!zalloc_cpumask_var(&policy->real_cpus, GFP_KERNEL))
|
|
|
|
goto err_free_rcpumask;
|
|
|
|
|
2016-03-03 17:21:33 +08:00
|
|
|
ret = kobject_init_and_add(&policy->kobj, &ktype_cpufreq,
|
|
|
|
cpufreq_global_kobject, "policy%u", cpu);
|
|
|
|
if (ret) {
|
2019-07-08 18:57:52 +08:00
|
|
|
dev_err(dev, "%s: failed to init policy->kobj: %d\n", __func__, ret);
|
2019-05-10 18:35:53 +08:00
|
|
|
/*
|
|
|
|
* The entire policy object will be freed below, but the extra
|
|
|
|
* memory allocated for the kobject name needs to be freed by
|
|
|
|
* releasing the kobject.
|
|
|
|
*/
|
2019-04-30 14:05:52 +08:00
|
|
|
kobject_put(&policy->kobj);
|
2016-03-03 17:21:33 +08:00
|
|
|
goto err_free_real_cpus;
|
|
|
|
}
|
|
|
|
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
freq_constraints_init(&policy->constraints);
|
|
|
|
|
2019-07-08 18:57:52 +08:00
|
|
|
policy->nb_min.notifier_call = cpufreq_notifier_min;
|
|
|
|
policy->nb_max.notifier_call = cpufreq_notifier_max;
|
|
|
|
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MIN,
|
|
|
|
&policy->nb_min);
|
2019-07-08 18:57:52 +08:00
|
|
|
if (ret) {
|
|
|
|
dev_err(dev, "Failed to register MIN QoS notifier: %d (%*pbl)\n",
|
|
|
|
ret, cpumask_pr_args(policy->cpus));
|
|
|
|
goto err_kobj_remove;
|
|
|
|
}
|
|
|
|
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
ret = freq_qos_add_notifier(&policy->constraints, FREQ_QOS_MAX,
|
|
|
|
&policy->nb_max);
|
2019-07-08 18:57:52 +08:00
|
|
|
if (ret) {
|
|
|
|
dev_err(dev, "Failed to register MAX QoS notifier: %d (%*pbl)\n",
|
|
|
|
ret, cpumask_pr_args(policy->cpus));
|
|
|
|
goto err_min_qos_notifier;
|
|
|
|
}
|
|
|
|
|
2013-08-07 01:23:08 +08:00
|
|
|
INIT_LIST_HEAD(&policy->policy_list);
|
2013-10-18 21:40:15 +08:00
|
|
|
init_rwsem(&policy->rwsem);
|
cpufreq: Make sure frequency transitions are serialized
Whenever we change the frequency of a CPU, we call the PRECHANGE and POSTCHANGE
notifiers. They must be serialized, i.e. PRECHANGE and POSTCHANGE notifiers
should strictly alternate, thereby preventing two different sets of PRECHANGE or
POSTCHANGE notifiers from interleaving arbitrarily.
The following examples illustrate why this is important:
Scenario 1:
-----------
A thread reading the value of cpuinfo_cur_freq, will call
__cpufreq_cpu_get()->cpufreq_out_of_sync()->cpufreq_notify_transition()
The ondemand governor can decide to change the frequency of the CPU at the same
time and hence it can end up sending the notifications via ->target().
If the notifiers are not serialized, the following sequence can occur:
- PRECHANGE Notification for freq A (from cpuinfo_cur_freq)
- PRECHANGE Notification for freq B (from target())
- Freq changed by target() to B
- POSTCHANGE Notification for freq B
- POSTCHANGE Notification for freq A
We can see from the above that the last POSTCHANGE Notification happens for freq
A but the hardware is set to run at freq B.
Where would we break then?: adjust_jiffies() in cpufreq.c & cpufreq_callback()
in arch/arm/kernel/smp.c (which also adjusts the jiffies). All the
loops_per_jiffy calculations will get messed up.
Scenario 2:
-----------
The governor calls __cpufreq_driver_target() to change the frequency. At the
same time, if we change scaling_{min|max}_freq from sysfs, it will end up
calling the governor's CPUFREQ_GOV_LIMITS notification, which will also call
__cpufreq_driver_target(). And hence we end up issuing concurrent calls to
->target().
Typically, platforms have the following logic in their ->target() routines:
(Eg: cpufreq-cpu0, omap, exynos, etc)
A. If new freq is more than old: Increase voltage
B. Change freq
C. If new freq is less than old: decrease voltage
Now, if the two concurrent calls to ->target() are X and Y, where X is trying to
increase the freq and Y is trying to decrease it, we get the following race
condition:
X.A: voltage gets increased for larger freq
Y.A: nothing happens
Y.B: freq gets decreased
Y.C: voltage gets decreased
X.B: freq gets increased
X.C: nothing happens
Thus we can end up setting a freq which is not supported by the voltage we have
set. That will probably make the clock to the CPU unstable and the system might
not work properly anymore.
This patch introduces a set of synchronization primitives to serialize frequency
transitions, which are to be used as shown below:
cpufreq_freq_transition_begin();
//Perform the frequency change
cpufreq_freq_transition_end();
The _begin() call sends the PRECHANGE notification whereas the _end() call sends
the POSTCHANGE notification. Also, all the necessary synchronization is handled
within these calls. In particular, even drivers which set the ASYNC_NOTIFICATION
flag can also use these APIs for performing frequency transitions (ie., you can
call _begin() from one task, and call the corresponding _end() from a different
task).
The actual synchronization underneath is not that complicated:
The key challenge is to allow drivers to begin the transition from one thread
and end it in a completely different thread (this is to enable drivers that do
asynchronous POSTCHANGE notification from bottom-halves, to also use the same
interface).
To achieve this, a 'transition_ongoing' flag, a 'transition_lock' spinlock and a
wait-queue are added per-policy. The flag and the wait-queue are used in
conjunction to create an "uninterrupted flow" from _begin() to _end(). The
spinlock is used to ensure that only one such "flow" is in flight at any given
time. Put together, this provides us all the necessary synchronization.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2014-03-24 16:05:44 +08:00
|
|
|
spin_lock_init(&policy->transition_lock);
|
|
|
|
init_waitqueue_head(&policy->transition_wait);
|
2015-01-02 15:04:38 +08:00
|
|
|
init_completion(&policy->kobj_unregister);
|
|
|
|
INIT_WORK(&policy->update, handle_update);
|
2013-10-18 21:40:15 +08:00
|
|
|
|
2015-07-28 05:11:50 +08:00
|
|
|
policy->cpu = cpu;
|
2013-07-30 06:54:11 +08:00
|
|
|
return policy;
|
|
|
|
|
2019-07-08 18:57:52 +08:00
|
|
|
err_min_qos_notifier:
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
|
|
|
|
&policy->nb_min);
|
2019-07-08 18:57:52 +08:00
|
|
|
err_kobj_remove:
|
|
|
|
cpufreq_policy_put_kobj(policy);
|
2016-03-03 17:21:33 +08:00
|
|
|
err_free_real_cpus:
|
|
|
|
free_cpumask_var(policy->real_cpus);
|
2015-06-08 20:55:29 +08:00
|
|
|
err_free_rcpumask:
|
|
|
|
free_cpumask_var(policy->related_cpus);
|
2013-07-30 06:54:11 +08:00
|
|
|
err_free_cpumask:
|
|
|
|
free_cpumask_var(policy->cpus);
|
|
|
|
err_free_policy:
|
|
|
|
kfree(policy);
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2017-01-05 12:47:27 +08:00
|
|
|
static void cpufreq_policy_free(struct cpufreq_policy *policy)
|
2013-07-30 06:54:11 +08:00
|
|
|
{
|
2015-05-08 14:23:45 +08:00
|
|
|
unsigned long flags;
|
|
|
|
int cpu;
|
|
|
|
|
2022-07-07 17:58:31 +08:00
|
|
|
/*
|
|
|
|
* The callers must ensure the policy is inactive by now, to avoid any
|
|
|
|
* races with show()/store() callbacks.
|
|
|
|
*/
|
|
|
|
if (unlikely(!policy_is_inactive(policy)))
|
|
|
|
pr_warn("%s: Freeing active policy\n", __func__);
|
|
|
|
|
2015-05-08 14:23:45 +08:00
|
|
|
/* Remove policy from list */
|
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
list_del(&policy->policy_list);
|
|
|
|
|
|
|
|
for_each_cpu(cpu, policy->related_cpus)
|
|
|
|
per_cpu(cpufreq_cpu_data, cpu) = NULL;
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MAX,
|
|
|
|
&policy->nb_max);
|
|
|
|
freq_qos_remove_notifier(&policy->constraints, FREQ_QOS_MIN,
|
|
|
|
&policy->nb_min);
|
2019-07-23 14:14:01 +08:00
|
|
|
|
2019-10-18 18:58:15 +08:00
|
|
|
/* Cancel any pending policy->update work before freeing the policy. */
|
|
|
|
cancel_work_sync(&policy->update);
|
2019-07-23 14:14:01 +08:00
|
|
|
|
|
|
|
if (policy->max_freq_req) {
|
|
|
|
/*
|
2021-12-01 15:40:20 +08:00
|
|
|
* Remove max_freq_req after sending CPUFREQ_REMOVE_POLICY
|
|
|
|
* notification, since CPUFREQ_CREATE_POLICY notification was
|
|
|
|
* sent after adding max_freq_req earlier.
|
2019-07-23 14:14:01 +08:00
|
|
|
*/
|
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_REMOVE_POLICY, policy);
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
freq_qos_remove_request(policy->max_freq_req);
|
2019-07-23 14:14:01 +08:00
|
|
|
}
|
|
|
|
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
freq_qos_remove_request(policy->min_freq_req);
|
2019-07-05 18:51:24 +08:00
|
|
|
kfree(policy->min_freq_req);
|
2019-07-08 18:57:52 +08:00
|
|
|
|
2017-01-05 12:47:27 +08:00
|
|
|
cpufreq_policy_put_kobj(policy);
|
cpufreq: Avoid attempts to create duplicate symbolic links
After commit 87549141d516 (cpufreq: Stop migrating sysfs files on
hotplug) there is a problem with CPUs that share cpufreq policy
objects with other CPUs and are initially offline.
Say CPU1 shares a policy with CPU0 which is online and is registered
first. As part of the registration process, cpufreq_add_dev() is
called for it. It creates the policy object and a symbolic link
to it from the CPU1's sysfs directory. If CPU1 is registered
subsequently and it is offline at that time, cpufreq_add_dev() will
attempt to create a symbolic link to the policy object for it, but
that link is present already, so a warning about that will be
triggered.
To avoid that warning, make cpufreq use an additional CPU mask
containing related CPUs that are actually present for each policy
object. That mask is initialized when the policy object is populated
after its creation (for the first online CPU using it) and it includes
CPUs from the "policy CPUs" mask returned by the cpufreq driver's
->init() callback that are physically present at that time. Symbolic
links to the policy are created only for the CPUs in that mask.
If cpufreq_add_dev() is invoked for an offline CPU, it checks the
new mask and only creates the symlink if the CPU was not in it (the
CPU is added to the mask at the same time).
In turn, cpufreq_remove_dev() drops the given CPU from the new mask,
removes its symlink to the policy object and returns, unless it is
the CPU owning the policy object. In that case, the policy object
is moved to a new CPU's sysfs directory or deleted if the CPU being
removed was the last user of the policy.
While at it, notice that cpufreq_remove_dev() can't fail, because
its return value is ignored, so make it ignore return values from
__cpufreq_remove_dev_prepare() and __cpufreq_remove_dev_finish()
and prevent these functions from aborting on errors returned by
__cpufreq_governor(). Also drop the now unused sif argument from
them.
Fixes: 87549141d516 (cpufreq: Stop migrating sysfs files on hotplug)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reported-and-tested-by: Russell King <linux@arm.linux.org.uk>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2015-07-26 08:07:47 +08:00
|
|
|
free_cpumask_var(policy->real_cpus);
|
2013-07-30 06:54:11 +08:00
|
|
|
free_cpumask_var(policy->related_cpus);
|
|
|
|
free_cpumask_var(policy->cpus);
|
|
|
|
kfree(policy);
|
|
|
|
}
|
|
|
|
|
2015-07-29 09:03:44 +08:00
|
|
|
static int cpufreq_online(unsigned int cpu)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2015-01-02 15:04:32 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2015-07-29 09:08:57 +08:00
|
|
|
bool new_policy;
|
2005-04-17 06:20:36 +08:00
|
|
|
unsigned long flags;
|
2015-07-29 09:03:44 +08:00
|
|
|
unsigned int j;
|
|
|
|
int ret;
|
2015-06-10 08:13:21 +08:00
|
|
|
|
2015-07-29 09:03:44 +08:00
|
|
|
pr_debug("%s: bringing CPU%u online\n", __func__, cpu);
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2015-02-19 19:32:06 +08:00
|
|
|
/* Check if this CPU already has a policy to manage it */
|
2015-05-12 14:52:12 +08:00
|
|
|
policy = per_cpu(cpufreq_cpu_data, cpu);
|
2015-07-28 05:11:21 +08:00
|
|
|
if (policy) {
|
2015-05-12 14:52:12 +08:00
|
|
|
WARN_ON(!cpumask_test_cpu(cpu, policy->related_cpus));
|
2015-07-28 05:11:21 +08:00
|
|
|
if (!policy_is_inactive(policy))
|
2015-07-28 05:11:37 +08:00
|
|
|
return cpufreq_add_policy_cpu(policy, cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-07-28 05:11:21 +08:00
|
|
|
/* This is the only online CPU for the policy. Start over. */
|
2015-07-29 09:08:57 +08:00
|
|
|
new_policy = false;
|
2015-07-28 05:11:21 +08:00
|
|
|
down_write(&policy->rwsem);
|
|
|
|
policy->cpu = cpu;
|
|
|
|
policy->governor = NULL;
|
|
|
|
} else {
|
2015-07-29 09:08:57 +08:00
|
|
|
new_policy = true;
|
2015-07-28 05:11:50 +08:00
|
|
|
policy = cpufreq_policy_alloc(cpu);
|
2013-12-27 08:07:11 +08:00
|
|
|
if (!policy)
|
2015-07-28 05:11:30 +08:00
|
|
|
return -ENOMEM;
|
2022-05-16 11:02:51 +08:00
|
|
|
down_write(&policy->rwsem);
|
2013-12-27 08:07:11 +08:00
|
|
|
}
|
cpufreq: Fix crash in cpufreq-stats during suspend/resume
Stephen Warren reported that the cpufreq-stats code hits a NULL pointer
dereference during the second attempt to suspend a system. He also
pin-pointed the problem to commit 5302c3f "cpufreq: Perform light-weight
init/teardown during suspend/resume".
That commit actually ensured that the cpufreq-stats table and the
cpufreq-stats sysfs entries are *not* torn down (ie., not freed) during
suspend/resume, which makes it all the more surprising. However, it turns
out that the root-cause is not that we access an already freed memory, but
that the reference to the allocated memory gets moved around and we lose
track of that during resume, leading to the reported crash in a subsequent
suspend attempt.
In the suspend path, during CPU offline, the value of policy->cpu is updated
by choosing one of the surviving CPUs in that policy, as long as there is
atleast one CPU in that policy. And cpufreq_stats_update_policy_cpu() is
invoked to update the reference to the stats structure by assigning it to
the new CPU. However, in the resume path, during CPU online, we end up
assigning a fresh CPU as the policy->cpu, without letting cpufreq-stats
know about this. Thus the reference to the stats structure remains
(incorrectly) associated with the old CPU. So, in a subsequent suspend attempt,
during CPU offline, we end up accessing an incorrect location to get the
stats structure, which eventually leads to the NULL pointer dereference.
Fix this by letting cpufreq-stats know about the update of the policy->cpu
during CPU online in the resume path. (Also, move the update_policy_cpu()
function higher up in the file, so that __cpufreq_add_dev() can invoke
it).
Reported-and-tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-09-12 04:12:59 +08:00
|
|
|
|
2019-02-12 19:06:04 +08:00
|
|
|
if (!new_policy && cpufreq_driver->online) {
|
2022-07-04 19:27:39 +08:00
|
|
|
/* Recover policy->cpus using related_cpus */
|
|
|
|
cpumask_copy(policy->cpus, policy->related_cpus);
|
|
|
|
|
2019-02-12 19:06:04 +08:00
|
|
|
ret = cpufreq_driver->online(policy);
|
|
|
|
if (ret) {
|
|
|
|
pr_debug("%s: %d: initialization failed\n", __func__,
|
|
|
|
__LINE__);
|
|
|
|
goto out_exit_policy;
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
cpumask_copy(policy->cpus, cpumask_of(cpu));
|
2013-01-12 13:14:38 +08:00
|
|
|
|
2019-02-12 19:06:04 +08:00
|
|
|
/*
|
|
|
|
* Call driver. From then on the cpufreq must be able
|
|
|
|
* to accept all calls to ->verify and ->setpolicy for this CPU.
|
|
|
|
*/
|
|
|
|
ret = cpufreq_driver->init(policy);
|
|
|
|
if (ret) {
|
|
|
|
pr_debug("%s: %d: initialization failed\n", __func__,
|
|
|
|
__LINE__);
|
|
|
|
goto out_free_policy;
|
|
|
|
}
|
2018-02-22 13:59:44 +08:00
|
|
|
|
2021-06-23 03:11:39 +08:00
|
|
|
/*
|
|
|
|
* The initialization has succeeded and the policy is online.
|
|
|
|
* If there is a problem with its frequency table, take it
|
|
|
|
* offline and drop it.
|
|
|
|
*/
|
2019-02-12 19:06:04 +08:00
|
|
|
ret = cpufreq_table_validate_and_sort(policy);
|
|
|
|
if (ret)
|
2021-06-23 03:11:39 +08:00
|
|
|
goto out_offline_policy;
|
2014-11-24 17:08:03 +08:00
|
|
|
|
2015-07-28 05:11:44 +08:00
|
|
|
/* related_cpus should at least include policy->cpus. */
|
2015-10-16 00:05:21 +08:00
|
|
|
cpumask_copy(policy->related_cpus, policy->cpus);
|
2015-07-28 05:11:44 +08:00
|
|
|
}
|
cpufreq: Avoid attempts to create duplicate symbolic links
After commit 87549141d516 (cpufreq: Stop migrating sysfs files on
hotplug) there is a problem with CPUs that share cpufreq policy
objects with other CPUs and are initially offline.
Say CPU1 shares a policy with CPU0 which is online and is registered
first. As part of the registration process, cpufreq_add_dev() is
called for it. It creates the policy object and a symbolic link
to it from the CPU1's sysfs directory. If CPU1 is registered
subsequently and it is offline at that time, cpufreq_add_dev() will
attempt to create a symbolic link to the policy object for it, but
that link is present already, so a warning about that will be
triggered.
To avoid that warning, make cpufreq use an additional CPU mask
containing related CPUs that are actually present for each policy
object. That mask is initialized when the policy object is populated
after its creation (for the first online CPU using it) and it includes
CPUs from the "policy CPUs" mask returned by the cpufreq driver's
->init() callback that are physically present at that time. Symbolic
links to the policy are created only for the CPUs in that mask.
If cpufreq_add_dev() is invoked for an offline CPU, it checks the
new mask and only creates the symlink if the CPU was not in it (the
CPU is added to the mask at the same time).
In turn, cpufreq_remove_dev() drops the given CPU from the new mask,
removes its symlink to the policy object and returns, unless it is
the CPU owning the policy object. In that case, the policy object
is moved to a new CPU's sysfs directory or deleted if the CPU being
removed was the last user of the policy.
While at it, notice that cpufreq_remove_dev() can't fail, because
its return value is ignored, so make it ignore return values from
__cpufreq_remove_dev_prepare() and __cpufreq_remove_dev_finish()
and prevent these functions from aborting on errors returned by
__cpufreq_governor(). Also drop the now unused sif argument from
them.
Fixes: 87549141d516 (cpufreq: Stop migrating sysfs files on hotplug)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reported-and-tested-by: Russell King <linux@arm.linux.org.uk>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2015-07-26 08:07:47 +08:00
|
|
|
|
2014-03-04 11:44:00 +08:00
|
|
|
/*
|
|
|
|
* affected cpus must always be the one, which are online. We aren't
|
|
|
|
* managing offline cpus here.
|
|
|
|
*/
|
|
|
|
cpumask_and(policy->cpus, policy->cpus, cpu_online_mask);
|
|
|
|
|
2015-07-29 09:08:57 +08:00
|
|
|
if (new_policy) {
|
2017-03-28 01:33:09 +08:00
|
|
|
for_each_cpu(j, policy->related_cpus) {
|
2015-05-08 14:23:45 +08:00
|
|
|
per_cpu(cpufreq_cpu_data, j) = policy;
|
2021-11-29 16:02:48 +08:00
|
|
|
add_cpu_dev_symlink(policy, j, get_cpu_device(j));
|
2017-03-28 01:33:09 +08:00
|
|
|
}
|
2019-07-05 18:51:24 +08:00
|
|
|
|
|
|
|
policy->min_freq_req = kzalloc(2 * sizeof(*policy->min_freq_req),
|
|
|
|
GFP_KERNEL);
|
2020-11-26 09:12:39 +08:00
|
|
|
if (!policy->min_freq_req) {
|
|
|
|
ret = -ENOMEM;
|
2019-07-05 18:51:24 +08:00
|
|
|
goto out_destroy_policy;
|
2020-11-26 09:12:39 +08:00
|
|
|
}
|
2019-07-05 18:51:24 +08:00
|
|
|
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
ret = freq_qos_add_request(&policy->constraints,
|
|
|
|
policy->min_freq_req, FREQ_QOS_MIN,
|
2021-12-17 03:32:15 +08:00
|
|
|
FREQ_QOS_MIN_DEFAULT_VALUE);
|
2019-07-05 18:51:24 +08:00
|
|
|
if (ret < 0) {
|
|
|
|
/*
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
* So we don't call freq_qos_remove_request() for an
|
2019-07-05 18:51:24 +08:00
|
|
|
* uninitialized request.
|
|
|
|
*/
|
|
|
|
kfree(policy->min_freq_req);
|
|
|
|
policy->min_freq_req = NULL;
|
|
|
|
goto out_destroy_policy;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This must be initialized right here to avoid calling
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
* freq_qos_remove_request() on uninitialized request in case
|
2019-07-05 18:51:24 +08:00
|
|
|
* of errors.
|
|
|
|
*/
|
|
|
|
policy->max_freq_req = policy->min_freq_req + 1;
|
|
|
|
|
cpufreq: Use per-policy frequency QoS
Replace the CPU device PM QoS used for the management of min and max
frequency constraints in cpufreq (and its users) with per-policy
frequency QoS to avoid problems with cpufreq policies covering
more then one CPU.
Namely, a cpufreq driver is registered with the subsys interface
which calls cpufreq_add_dev() for each CPU, starting from CPU0, so
currently the PM QoS notifiers are added to the first CPU in the
policy (i.e. CPU0 in the majority of cases).
In turn, when the cpufreq driver is unregistered, the subsys interface
doing that calls cpufreq_remove_dev() for each CPU, starting from CPU0,
and the PM QoS notifiers are only removed when cpufreq_remove_dev() is
called for the last CPU in the policy, say CPUx, which as a rule is
not CPU0 if the policy covers more than one CPU. Then, the PM QoS
notifiers cannot be removed, because CPUx does not have them, and
they are still there in the device PM QoS notifiers list of CPU0,
which prevents new PM QoS notifiers from being registered for CPU0
on the next attempt to register the cpufreq driver.
The same issue occurs when the first CPU in the policy goes offline
before unregistering the driver.
After this change it does not matter which CPU is the policy CPU at
the driver registration time and whether or not it is online all the
time, because the frequency QoS is per policy and not per CPU.
Fixes: 67d874c3b2c6 ("cpufreq: Register notifiers with the PM QoS framework")
Reported-by: Dmitry Osipenko <digetx@gmail.com>
Tested-by: Dmitry Osipenko <digetx@gmail.com>
Reported-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Sudeep Holla <sudeep.holla@arm.com>
Diagnosed-by: Viresh Kumar <viresh.kumar@linaro.org>
Link: https://lore.kernel.org/linux-pm/5ad2624194baa2f53acc1f1e627eb7684c577a19.1562210705.git.viresh.kumar@linaro.org/T/#md2d89e95906b8c91c15f582146173dce2e86e99f
Link: https://lore.kernel.org/linux-pm/20191017094612.6tbkwoq4harsjcqv@vireshk-i7/T/#m30d48cc23b9a80467fbaa16e30f90b3828a5a29b
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2019-10-16 18:47:06 +08:00
|
|
|
ret = freq_qos_add_request(&policy->constraints,
|
|
|
|
policy->max_freq_req, FREQ_QOS_MAX,
|
2021-12-17 03:32:15 +08:00
|
|
|
FREQ_QOS_MAX_DEFAULT_VALUE);
|
2019-07-05 18:51:24 +08:00
|
|
|
if (ret < 0) {
|
|
|
|
policy->max_freq_req = NULL;
|
|
|
|
goto out_destroy_policy;
|
|
|
|
}
|
2019-07-23 14:14:01 +08:00
|
|
|
|
|
|
|
blocking_notifier_call_chain(&cpufreq_policy_notifier_list,
|
|
|
|
CPUFREQ_CREATE_POLICY, policy);
|
2015-05-08 14:23:45 +08:00
|
|
|
}
|
2014-01-09 23:08:43 +08:00
|
|
|
|
2019-06-20 11:05:48 +08:00
|
|
|
if (cpufreq_driver->get && has_target()) {
|
2013-10-03 22:58:30 +08:00
|
|
|
policy->cur = cpufreq_driver->get(policy->cpu);
|
|
|
|
if (!policy->cur) {
|
2020-11-26 09:12:39 +08:00
|
|
|
ret = -EIO;
|
2013-10-03 22:58:30 +08:00
|
|
|
pr_err("%s: ->get() failed\n", __func__);
|
2018-02-22 13:59:44 +08:00
|
|
|
goto out_destroy_policy;
|
2013-10-03 22:58:30 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2013-12-03 13:50:46 +08:00
|
|
|
/*
|
|
|
|
* Sometimes boot loaders set CPU frequency to a value outside of
|
|
|
|
* frequency table present with cpufreq core. In such cases CPU might be
|
|
|
|
* unstable if it has to run on that frequency for long duration of time
|
|
|
|
* and so its better to set it to a frequency which is specified in
|
|
|
|
* freq-table. This also makes cpufreq stats inconsistent as
|
|
|
|
* cpufreq-stats would fail to register because current frequency of CPU
|
|
|
|
* isn't found in freq-table.
|
|
|
|
*
|
|
|
|
* Because we don't want this change to effect boot process badly, we go
|
|
|
|
* for the next freq which is >= policy->cur ('cur' must be set by now,
|
|
|
|
* otherwise we will end up setting freq to lowest of the table as 'cur'
|
|
|
|
* is initialized to zero).
|
|
|
|
*
|
|
|
|
* We are passing target-freq as "policy->cur - 1" otherwise
|
|
|
|
* __cpufreq_driver_target() would simply fail, as policy->cur will be
|
|
|
|
* equal to target-freq.
|
|
|
|
*/
|
|
|
|
if ((cpufreq_driver->flags & CPUFREQ_NEED_INITIAL_FREQ_CHECK)
|
|
|
|
&& has_target()) {
|
2020-10-13 13:12:47 +08:00
|
|
|
unsigned int old_freq = policy->cur;
|
|
|
|
|
2013-12-03 13:50:46 +08:00
|
|
|
/* Are we running at unknown frequency ? */
|
2020-10-13 13:12:47 +08:00
|
|
|
ret = cpufreq_frequency_table_get_index(policy, old_freq);
|
2013-12-03 13:50:46 +08:00
|
|
|
if (ret == -EINVAL) {
|
2020-10-13 13:12:47 +08:00
|
|
|
ret = __cpufreq_driver_target(policy, old_freq - 1,
|
|
|
|
CPUFREQ_RELATION_L);
|
2013-12-03 13:50:46 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Reaching here after boot in a few seconds may not
|
|
|
|
* mean that system will remain stable at "unknown"
|
|
|
|
* frequency for longer duration. Hence, a BUG_ON().
|
|
|
|
*/
|
|
|
|
BUG_ON(ret);
|
2020-10-13 13:12:47 +08:00
|
|
|
pr_info("%s: CPU%d: Running at unlisted initial frequency: %u KHz, changing to: %u KHz\n",
|
|
|
|
__func__, policy->cpu, old_freq, policy->cur);
|
2013-12-03 13:50:46 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-07-29 09:08:57 +08:00
|
|
|
if (new_policy) {
|
2015-07-28 05:11:37 +08:00
|
|
|
ret = cpufreq_add_dev_interface(policy);
|
2013-07-30 06:54:49 +08:00
|
|
|
if (ret)
|
2018-02-22 13:59:44 +08:00
|
|
|
goto out_destroy_policy;
|
2016-06-01 04:14:44 +08:00
|
|
|
|
|
|
|
cpufreq_stats_create_table(policy);
|
2006-03-05 16:37:23 +08:00
|
|
|
|
2015-05-08 14:23:45 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
list_add(&policy->policy_list, &cpufreq_policy_list);
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2021-08-10 14:34:33 +08:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Register with the energy model before
|
|
|
|
* sched_cpufreq_governor_change() is called, which will result
|
|
|
|
* in rebuilding of the sched domains, which should only be done
|
|
|
|
* once the energy model is properly initialized for the policy
|
|
|
|
* first.
|
|
|
|
*
|
|
|
|
* Also, this should be called before the policy is registered
|
|
|
|
* with cooling framework.
|
|
|
|
*/
|
|
|
|
if (cpufreq_driver->register_em)
|
|
|
|
cpufreq_driver->register_em(policy);
|
2015-05-08 14:23:45 +08:00
|
|
|
}
|
2013-08-20 14:38:23 +08:00
|
|
|
|
2015-07-08 17:42:16 +08:00
|
|
|
ret = cpufreq_init_policy(policy);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: Failed to initialize policy for cpu: %d (%d)\n",
|
|
|
|
__func__, cpu, ret);
|
2018-02-22 13:59:44 +08:00
|
|
|
goto out_destroy_policy;
|
2013-12-24 09:41:01 +08:00
|
|
|
}
|
2013-07-30 06:54:23 +08:00
|
|
|
|
2014-03-04 11:44:01 +08:00
|
|
|
up_write(&policy->rwsem);
|
2013-12-24 09:41:01 +08:00
|
|
|
|
2007-12-18 03:54:39 +08:00
|
|
|
kobject_uevent(&policy->kobj, KOBJ_ADD);
|
2014-11-27 08:37:51 +08:00
|
|
|
|
2022-01-28 11:25:53 +08:00
|
|
|
/* Callback for handling stuff after policy is ready */
|
|
|
|
if (cpufreq_driver->ready)
|
|
|
|
cpufreq_driver->ready(policy);
|
|
|
|
|
2019-06-25 19:32:41 +08:00
|
|
|
if (cpufreq_thermal_control_enabled(cpufreq_driver))
|
2019-01-30 13:22:01 +08:00
|
|
|
policy->cdev = of_cpufreq_cooling_register(policy);
|
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("initialization complete\n");
|
2006-03-29 14:48:37 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
return 0;
|
|
|
|
|
2018-02-22 13:59:44 +08:00
|
|
|
out_destroy_policy:
|
2018-02-22 13:59:43 +08:00
|
|
|
for_each_cpu(j, policy->real_cpus)
|
2022-05-11 17:06:24 +08:00
|
|
|
remove_cpu_dev_symlink(policy, j, get_cpu_device(j));
|
2018-02-22 13:59:43 +08:00
|
|
|
|
2021-06-23 03:11:39 +08:00
|
|
|
out_offline_policy:
|
|
|
|
if (cpufreq_driver->offline)
|
|
|
|
cpufreq_driver->offline(policy);
|
|
|
|
|
2018-02-22 13:59:44 +08:00
|
|
|
out_exit_policy:
|
2013-10-03 22:58:30 +08:00
|
|
|
if (cpufreq_driver->exit)
|
|
|
|
cpufreq_driver->exit(policy);
|
2017-03-28 01:33:09 +08:00
|
|
|
|
2015-07-08 17:42:15 +08:00
|
|
|
out_free_policy:
|
2022-07-07 17:58:31 +08:00
|
|
|
cpumask_clear(policy->cpus);
|
2022-05-16 11:02:51 +08:00
|
|
|
up_write(&policy->rwsem);
|
|
|
|
|
2017-01-05 12:47:27 +08:00
|
|
|
cpufreq_policy_free(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2015-07-29 09:03:44 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_add_dev - the cpufreq interface for a CPU device.
|
|
|
|
* @dev: CPU device.
|
|
|
|
* @sif: Subsystem interface structure pointer (not used)
|
|
|
|
*/
|
|
|
|
static int cpufreq_add_dev(struct device *dev, struct subsys_interface *sif)
|
|
|
|
{
|
2016-04-07 09:31:57 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2015-07-29 09:03:44 +08:00
|
|
|
unsigned cpu = dev->id;
|
cpufreq: create link to policy only for registered CPUs
If a cpufreq driver is registered very early in the boot stage (e.g.
registered from postcore_initcall()), then cpufreq core may generate
kernel warnings for it.
In this case, the CPUs are brought online, then the cpufreq driver is
registered, and then the CPU topology devices are registered. However,
by the time cpufreq_add_dev() gets called, the cpu device isn't stored
in the per-cpu variable (cpu_sys_devices,) which is read by
get_cpu_device().
So the cpufreq core fails to get device for the CPU, for which
cpufreq_add_dev() was called in the first place and we will hit a
WARN_ON(!cpu_dev).
Even if we reuse the 'dev' parameter passed to cpufreq_add_dev() to
avoid that warning, there might be other CPUs online that share the
policy with the cpu for which cpufreq_add_dev() is called. Eventually
get_cpu_device() will return NULL for them as well, and we will hit the
same WARN_ON() again.
In order to fix these issues, change cpufreq core to create links to the
policy for a cpu only when cpufreq_add_dev() is called for that CPU.
Reuse the 'real_cpus' mask to track that as well.
Note that cpufreq_remove_dev() already handles removal of the links for
individual CPUs and cpufreq_add_dev() has aligned with that now.
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2016-09-12 14:37:05 +08:00
|
|
|
int ret;
|
2015-07-29 09:03:44 +08:00
|
|
|
|
|
|
|
dev_dbg(dev, "%s: adding CPU%u\n", __func__, cpu);
|
|
|
|
|
cpufreq: create link to policy only for registered CPUs
If a cpufreq driver is registered very early in the boot stage (e.g.
registered from postcore_initcall()), then cpufreq core may generate
kernel warnings for it.
In this case, the CPUs are brought online, then the cpufreq driver is
registered, and then the CPU topology devices are registered. However,
by the time cpufreq_add_dev() gets called, the cpu device isn't stored
in the per-cpu variable (cpu_sys_devices,) which is read by
get_cpu_device().
So the cpufreq core fails to get device for the CPU, for which
cpufreq_add_dev() was called in the first place and we will hit a
WARN_ON(!cpu_dev).
Even if we reuse the 'dev' parameter passed to cpufreq_add_dev() to
avoid that warning, there might be other CPUs online that share the
policy with the cpu for which cpufreq_add_dev() is called. Eventually
get_cpu_device() will return NULL for them as well, and we will hit the
same WARN_ON() again.
In order to fix these issues, change cpufreq core to create links to the
policy for a cpu only when cpufreq_add_dev() is called for that CPU.
Reuse the 'real_cpus' mask to track that as well.
Note that cpufreq_remove_dev() already handles removal of the links for
individual CPUs and cpufreq_add_dev() has aligned with that now.
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2016-09-12 14:37:05 +08:00
|
|
|
if (cpu_online(cpu)) {
|
|
|
|
ret = cpufreq_online(cpu);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
2015-07-29 09:03:44 +08:00
|
|
|
|
cpufreq: create link to policy only for registered CPUs
If a cpufreq driver is registered very early in the boot stage (e.g.
registered from postcore_initcall()), then cpufreq core may generate
kernel warnings for it.
In this case, the CPUs are brought online, then the cpufreq driver is
registered, and then the CPU topology devices are registered. However,
by the time cpufreq_add_dev() gets called, the cpu device isn't stored
in the per-cpu variable (cpu_sys_devices,) which is read by
get_cpu_device().
So the cpufreq core fails to get device for the CPU, for which
cpufreq_add_dev() was called in the first place and we will hit a
WARN_ON(!cpu_dev).
Even if we reuse the 'dev' parameter passed to cpufreq_add_dev() to
avoid that warning, there might be other CPUs online that share the
policy with the cpu for which cpufreq_add_dev() is called. Eventually
get_cpu_device() will return NULL for them as well, and we will hit the
same WARN_ON() again.
In order to fix these issues, change cpufreq core to create links to the
policy for a cpu only when cpufreq_add_dev() is called for that CPU.
Reuse the 'real_cpus' mask to track that as well.
Note that cpufreq_remove_dev() already handles removal of the links for
individual CPUs and cpufreq_add_dev() has aligned with that now.
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2016-09-12 14:37:05 +08:00
|
|
|
/* Create sysfs link on CPU registration */
|
2016-04-07 09:31:57 +08:00
|
|
|
policy = per_cpu(cpufreq_cpu_data, cpu);
|
2017-03-28 01:33:09 +08:00
|
|
|
if (policy)
|
2021-11-29 16:02:48 +08:00
|
|
|
add_cpu_dev_symlink(policy, cpu, dev);
|
cpufreq: create link to policy only for registered CPUs
If a cpufreq driver is registered very early in the boot stage (e.g.
registered from postcore_initcall()), then cpufreq core may generate
kernel warnings for it.
In this case, the CPUs are brought online, then the cpufreq driver is
registered, and then the CPU topology devices are registered. However,
by the time cpufreq_add_dev() gets called, the cpu device isn't stored
in the per-cpu variable (cpu_sys_devices,) which is read by
get_cpu_device().
So the cpufreq core fails to get device for the CPU, for which
cpufreq_add_dev() was called in the first place and we will hit a
WARN_ON(!cpu_dev).
Even if we reuse the 'dev' parameter passed to cpufreq_add_dev() to
avoid that warning, there might be other CPUs online that share the
policy with the cpu for which cpufreq_add_dev() is called. Eventually
get_cpu_device() will return NULL for them as well, and we will hit the
same WARN_ON() again.
In order to fix these issues, change cpufreq core to create links to the
policy for a cpu only when cpufreq_add_dev() is called for that CPU.
Reuse the 'real_cpus' mask to track that as well.
Note that cpufreq_remove_dev() already handles removal of the links for
individual CPUs and cpufreq_add_dev() has aligned with that now.
Reported-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2016-09-12 14:37:05 +08:00
|
|
|
|
2017-03-28 01:33:09 +08:00
|
|
|
return 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2022-05-11 23:50:09 +08:00
|
|
|
static void __cpufreq_offline(unsigned int cpu, struct cpufreq_policy *policy)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2016-02-11 20:01:11 +08:00
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2016-05-12 21:14:12 +08:00
|
|
|
if (has_target())
|
|
|
|
cpufreq_stop_governor(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-06-10 08:20:23 +08:00
|
|
|
cpumask_clear_cpu(cpu, policy->cpus);
|
2015-05-12 14:52:34 +08:00
|
|
|
|
2015-06-10 08:20:23 +08:00
|
|
|
if (!policy_is_inactive(policy)) {
|
2022-05-11 23:48:41 +08:00
|
|
|
/* Nominate a new CPU if necessary. */
|
|
|
|
if (cpu == policy->cpu)
|
|
|
|
policy->cpu = cpumask_any(policy->cpus);
|
|
|
|
|
|
|
|
/* Start the governor again for the active policy. */
|
2015-06-10 08:20:23 +08:00
|
|
|
if (has_target()) {
|
2016-03-21 22:45:24 +08:00
|
|
|
ret = cpufreq_start_governor(policy);
|
2015-06-10 08:20:23 +08:00
|
|
|
if (ret)
|
|
|
|
pr_err("%s: Failed to start governor\n", __func__);
|
|
|
|
}
|
2013-09-07 03:53:09 +08:00
|
|
|
|
2022-05-11 23:50:09 +08:00
|
|
|
return;
|
2013-09-07 03:53:09 +08:00
|
|
|
}
|
|
|
|
|
2022-05-11 23:48:41 +08:00
|
|
|
if (has_target())
|
|
|
|
strncpy(policy->last_governor, policy->governor->name,
|
|
|
|
CPUFREQ_NAME_LEN);
|
|
|
|
else
|
|
|
|
policy->last_policy = policy->policy;
|
|
|
|
|
2019-06-25 19:32:41 +08:00
|
|
|
if (cpufreq_thermal_control_enabled(cpufreq_driver)) {
|
2019-01-30 13:22:01 +08:00
|
|
|
cpufreq_cooling_unregister(policy->cdev);
|
|
|
|
policy->cdev = NULL;
|
|
|
|
}
|
|
|
|
|
2016-05-12 21:13:35 +08:00
|
|
|
if (has_target())
|
|
|
|
cpufreq_exit_governor(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-06-10 08:13:21 +08:00
|
|
|
/*
|
2019-02-12 19:06:04 +08:00
|
|
|
* Perform the ->offline() during light-weight tear-down, as
|
|
|
|
* that allows fast recovery when the CPU comes back.
|
2015-06-10 08:13:21 +08:00
|
|
|
*/
|
2019-02-12 19:06:04 +08:00
|
|
|
if (cpufreq_driver->offline) {
|
|
|
|
cpufreq_driver->offline(policy);
|
|
|
|
} else if (cpufreq_driver->exit) {
|
2015-06-10 08:13:21 +08:00
|
|
|
cpufreq_driver->exit(policy);
|
2015-10-08 04:50:44 +08:00
|
|
|
policy->freq_table = NULL;
|
|
|
|
}
|
2022-05-11 23:50:09 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int cpufreq_offline(unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
|
|
|
|
pr_debug("%s: unregistering CPU %u\n", __func__, cpu);
|
|
|
|
|
|
|
|
policy = cpufreq_cpu_get_raw(cpu);
|
|
|
|
if (!policy) {
|
|
|
|
pr_debug("%s: No cpu_data found\n", __func__);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
down_write(&policy->rwsem);
|
|
|
|
|
|
|
|
__cpufreq_offline(cpu, policy);
|
2016-02-11 20:01:12 +08:00
|
|
|
|
|
|
|
up_write(&policy->rwsem);
|
2016-09-07 01:04:48 +08:00
|
|
|
return 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2013-10-02 16:43:14 +08:00
|
|
|
* cpufreq_remove_dev - remove a CPU device
|
2013-09-07 03:53:09 +08:00
|
|
|
*
|
|
|
|
* Removes the cpufreq interface for a CPU device.
|
|
|
|
*/
|
2015-07-30 17:34:01 +08:00
|
|
|
static void cpufreq_remove_dev(struct device *dev, struct subsys_interface *sif)
|
2007-02-06 08:12:44 +08:00
|
|
|
{
|
2011-12-22 06:29:42 +08:00
|
|
|
unsigned int cpu = dev->id;
|
cpufreq: Avoid attempts to create duplicate symbolic links
After commit 87549141d516 (cpufreq: Stop migrating sysfs files on
hotplug) there is a problem with CPUs that share cpufreq policy
objects with other CPUs and are initially offline.
Say CPU1 shares a policy with CPU0 which is online and is registered
first. As part of the registration process, cpufreq_add_dev() is
called for it. It creates the policy object and a symbolic link
to it from the CPU1's sysfs directory. If CPU1 is registered
subsequently and it is offline at that time, cpufreq_add_dev() will
attempt to create a symbolic link to the policy object for it, but
that link is present already, so a warning about that will be
triggered.
To avoid that warning, make cpufreq use an additional CPU mask
containing related CPUs that are actually present for each policy
object. That mask is initialized when the policy object is populated
after its creation (for the first online CPU using it) and it includes
CPUs from the "policy CPUs" mask returned by the cpufreq driver's
->init() callback that are physically present at that time. Symbolic
links to the policy are created only for the CPUs in that mask.
If cpufreq_add_dev() is invoked for an offline CPU, it checks the
new mask and only creates the symlink if the CPU was not in it (the
CPU is added to the mask at the same time).
In turn, cpufreq_remove_dev() drops the given CPU from the new mask,
removes its symlink to the policy object and returns, unless it is
the CPU owning the policy object. In that case, the policy object
is moved to a new CPU's sysfs directory or deleted if the CPU being
removed was the last user of the policy.
While at it, notice that cpufreq_remove_dev() can't fail, because
its return value is ignored, so make it ignore return values from
__cpufreq_remove_dev_prepare() and __cpufreq_remove_dev_finish()
and prevent these functions from aborting on errors returned by
__cpufreq_governor(). Also drop the now unused sif argument from
them.
Fixes: 87549141d516 (cpufreq: Stop migrating sysfs files on hotplug)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reported-and-tested-by: Russell King <linux@arm.linux.org.uk>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2015-07-26 08:07:47 +08:00
|
|
|
struct cpufreq_policy *policy = per_cpu(cpufreq_cpu_data, cpu);
|
2015-06-10 08:13:21 +08:00
|
|
|
|
cpufreq: Avoid attempts to create duplicate symbolic links
After commit 87549141d516 (cpufreq: Stop migrating sysfs files on
hotplug) there is a problem with CPUs that share cpufreq policy
objects with other CPUs and are initially offline.
Say CPU1 shares a policy with CPU0 which is online and is registered
first. As part of the registration process, cpufreq_add_dev() is
called for it. It creates the policy object and a symbolic link
to it from the CPU1's sysfs directory. If CPU1 is registered
subsequently and it is offline at that time, cpufreq_add_dev() will
attempt to create a symbolic link to the policy object for it, but
that link is present already, so a warning about that will be
triggered.
To avoid that warning, make cpufreq use an additional CPU mask
containing related CPUs that are actually present for each policy
object. That mask is initialized when the policy object is populated
after its creation (for the first online CPU using it) and it includes
CPUs from the "policy CPUs" mask returned by the cpufreq driver's
->init() callback that are physically present at that time. Symbolic
links to the policy are created only for the CPUs in that mask.
If cpufreq_add_dev() is invoked for an offline CPU, it checks the
new mask and only creates the symlink if the CPU was not in it (the
CPU is added to the mask at the same time).
In turn, cpufreq_remove_dev() drops the given CPU from the new mask,
removes its symlink to the policy object and returns, unless it is
the CPU owning the policy object. In that case, the policy object
is moved to a new CPU's sysfs directory or deleted if the CPU being
removed was the last user of the policy.
While at it, notice that cpufreq_remove_dev() can't fail, because
its return value is ignored, so make it ignore return values from
__cpufreq_remove_dev_prepare() and __cpufreq_remove_dev_finish()
and prevent these functions from aborting on errors returned by
__cpufreq_governor(). Also drop the now unused sif argument from
them.
Fixes: 87549141d516 (cpufreq: Stop migrating sysfs files on hotplug)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reported-and-tested-by: Russell King <linux@arm.linux.org.uk>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2015-07-26 08:07:47 +08:00
|
|
|
if (!policy)
|
2015-08-31 23:47:40 +08:00
|
|
|
return;
|
2015-06-10 08:13:21 +08:00
|
|
|
|
2022-05-11 23:51:39 +08:00
|
|
|
down_write(&policy->rwsem);
|
|
|
|
|
2016-02-11 20:01:11 +08:00
|
|
|
if (cpu_online(cpu))
|
2022-05-11 23:51:39 +08:00
|
|
|
__cpufreq_offline(cpu, policy);
|
2015-06-10 08:13:21 +08:00
|
|
|
|
2022-05-11 17:06:24 +08:00
|
|
|
remove_cpu_dev_symlink(policy, cpu, dev);
|
2015-06-10 08:13:21 +08:00
|
|
|
|
2022-05-11 23:51:39 +08:00
|
|
|
if (!cpumask_empty(policy->real_cpus)) {
|
|
|
|
up_write(&policy->rwsem);
|
|
|
|
return;
|
2019-02-12 19:06:04 +08:00
|
|
|
}
|
2022-05-11 23:51:39 +08:00
|
|
|
|
|
|
|
/* We did light-weight exit earlier, do full tear down now */
|
|
|
|
if (cpufreq_driver->offline)
|
|
|
|
cpufreq_driver->exit(policy);
|
|
|
|
|
|
|
|
up_write(&policy->rwsem);
|
|
|
|
|
|
|
|
cpufreq_policy_free(policy);
|
2007-02-06 08:12:44 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2020-11-19 03:02:42 +08:00
|
|
|
* cpufreq_out_of_sync - Fix up actual and saved CPU frequency difference.
|
|
|
|
* @policy: Policy managing CPUs.
|
|
|
|
* @new_freq: New CPU frequency.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2020-11-19 03:02:42 +08:00
|
|
|
* Adjust to the current frequency first and clean up later by either calling
|
|
|
|
* cpufreq_update_policy(), or scheduling handle_update().
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2015-01-02 15:04:28 +08:00
|
|
|
static void cpufreq_out_of_sync(struct cpufreq_policy *policy,
|
2006-10-26 18:50:58 +08:00
|
|
|
unsigned int new_freq)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
struct cpufreq_freqs freqs;
|
2013-03-24 14:26:43 +08:00
|
|
|
|
2014-03-12 01:03:00 +08:00
|
|
|
pr_debug("Warning: CPU frequency out of sync: cpufreq and timing core thinks of %u, is %u kHz\n",
|
2015-01-02 15:04:28 +08:00
|
|
|
policy->cur, new_freq);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-01-02 15:04:28 +08:00
|
|
|
freqs.old = policy->cur;
|
2005-04-17 06:20:36 +08:00
|
|
|
freqs.new = new_freq;
|
2013-03-24 14:26:43 +08:00
|
|
|
|
2014-03-24 16:05:45 +08:00
|
|
|
cpufreq_freq_transition_begin(policy, &freqs);
|
|
|
|
cpufreq_freq_transition_end(policy, &freqs, 0);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2019-06-20 11:05:49 +08:00
|
|
|
static unsigned int cpufreq_verify_current_freq(struct cpufreq_policy *policy, bool update)
|
|
|
|
{
|
|
|
|
unsigned int new_freq;
|
|
|
|
|
|
|
|
new_freq = cpufreq_driver->get(policy->cpu);
|
|
|
|
if (!new_freq)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If fast frequency switching is used with the given policy, the check
|
|
|
|
* against policy->cur is pointless, so skip it in that case.
|
|
|
|
*/
|
|
|
|
if (policy->fast_switch_enabled || !has_target())
|
|
|
|
return new_freq;
|
|
|
|
|
|
|
|
if (policy->cur != new_freq) {
|
2022-05-04 16:21:35 +08:00
|
|
|
/*
|
|
|
|
* For some platforms, the frequency returned by hardware may be
|
|
|
|
* slightly different from what is provided in the frequency
|
|
|
|
* table, for example hardware may return 499 MHz instead of 500
|
|
|
|
* MHz. In such cases it is better to avoid getting into
|
|
|
|
* unnecessary frequency updates.
|
|
|
|
*/
|
|
|
|
if (abs(policy->cur - new_freq) < HZ_PER_MHZ)
|
|
|
|
return policy->cur;
|
|
|
|
|
2019-06-20 11:05:49 +08:00
|
|
|
cpufreq_out_of_sync(policy, new_freq);
|
|
|
|
if (update)
|
|
|
|
schedule_work(&policy->update);
|
|
|
|
}
|
|
|
|
|
|
|
|
return new_freq;
|
|
|
|
}
|
|
|
|
|
2006-02-28 13:43:23 +08:00
|
|
|
/**
|
2006-12-13 17:19:15 +08:00
|
|
|
* cpufreq_quick_get - get the CPU frequency (in kHz) from policy->cur
|
2005-12-03 02:43:20 +08:00
|
|
|
* @cpu: CPU number
|
|
|
|
*
|
|
|
|
* This is the last known freq, without actually getting it from the driver.
|
|
|
|
* Return value will be same as what is shown in scaling_cur_freq in sysfs.
|
|
|
|
*/
|
|
|
|
unsigned int cpufreq_quick_get(unsigned int cpu)
|
|
|
|
{
|
2013-02-07 01:02:08 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2006-10-26 18:50:58 +08:00
|
|
|
unsigned int ret_freq = 0;
|
2016-03-11 16:43:07 +08:00
|
|
|
unsigned long flags;
|
2005-12-03 02:43:20 +08:00
|
|
|
|
2016-03-11 16:43:07 +08:00
|
|
|
read_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
if (cpufreq_driver && cpufreq_driver->setpolicy && cpufreq_driver->get) {
|
|
|
|
ret_freq = cpufreq_driver->get(cpu);
|
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
return ret_freq;
|
|
|
|
}
|
|
|
|
|
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2013-02-07 01:02:08 +08:00
|
|
|
|
|
|
|
policy = cpufreq_cpu_get(cpu);
|
2005-12-03 02:43:20 +08:00
|
|
|
if (policy) {
|
2006-10-26 18:50:58 +08:00
|
|
|
ret_freq = policy->cur;
|
2005-12-03 02:43:20 +08:00
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
|
|
|
|
2008-02-08 05:33:49 +08:00
|
|
|
return ret_freq;
|
2005-12-03 02:43:20 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_quick_get);
|
|
|
|
|
2011-06-29 01:59:12 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_quick_get_max - get the max reported CPU frequency for this CPU
|
|
|
|
* @cpu: CPU number
|
|
|
|
*
|
|
|
|
* Just return the max possible frequency for a given CPU.
|
|
|
|
*/
|
|
|
|
unsigned int cpufreq_quick_get_max(unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
|
|
|
unsigned int ret_freq = 0;
|
|
|
|
|
|
|
|
if (policy) {
|
|
|
|
ret_freq = policy->max;
|
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret_freq;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_quick_get_max);
|
|
|
|
|
2020-03-05 17:06:25 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_get_hw_max_freq - get the max hardware frequency of the CPU
|
|
|
|
* @cpu: CPU number
|
|
|
|
*
|
|
|
|
* The default return value is the max_freq field of cpuinfo.
|
|
|
|
*/
|
|
|
|
__weak unsigned int cpufreq_get_hw_max_freq(unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
|
|
|
unsigned int ret_freq = 0;
|
|
|
|
|
|
|
|
if (policy) {
|
|
|
|
ret_freq = policy->cpuinfo.max_freq;
|
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret_freq;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_get_hw_max_freq);
|
|
|
|
|
2015-01-02 15:04:29 +08:00
|
|
|
static unsigned int __cpufreq_get(struct cpufreq_policy *policy)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2019-04-19 14:27:58 +08:00
|
|
|
if (unlikely(policy_is_inactive(policy)))
|
2019-06-20 11:05:49 +08:00
|
|
|
return 0;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2019-06-20 11:05:49 +08:00
|
|
|
return cpufreq_verify_current_freq(policy, true);
|
2007-02-06 08:12:44 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2007-02-06 08:12:44 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_get - get the current CPU frequency (in kHz)
|
|
|
|
* @cpu: CPU number
|
|
|
|
*
|
|
|
|
* Get the CPU current (static) CPU frequency
|
|
|
|
*/
|
|
|
|
unsigned int cpufreq_get(unsigned int cpu)
|
|
|
|
{
|
2014-03-05 04:42:15 +08:00
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
|
2007-02-06 08:12:44 +08:00
|
|
|
unsigned int ret_freq = 0;
|
|
|
|
|
2014-03-05 04:42:15 +08:00
|
|
|
if (policy) {
|
|
|
|
down_read(&policy->rwsem);
|
2019-04-19 14:27:58 +08:00
|
|
|
if (cpufreq_driver->get)
|
|
|
|
ret_freq = __cpufreq_get(policy);
|
2014-03-05 04:42:15 +08:00
|
|
|
up_read(&policy->rwsem);
|
2007-02-06 08:12:44 +08:00
|
|
|
|
2014-03-05 04:42:15 +08:00
|
|
|
cpufreq_cpu_put(policy);
|
|
|
|
}
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2008-02-08 05:33:49 +08:00
|
|
|
return ret_freq;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_get);
|
|
|
|
|
2011-12-22 06:29:42 +08:00
|
|
|
static struct subsys_interface cpufreq_interface = {
|
|
|
|
.name = "cpufreq",
|
|
|
|
.subsys = &cpu_subsys,
|
|
|
|
.add_dev = cpufreq_add_dev,
|
|
|
|
.remove_dev = cpufreq_remove_dev,
|
cpufreq: Use syscore_ops for boot CPU suspend/resume (v2)
The cpufreq subsystem uses sysdev suspend and resume for
executing cpufreq_suspend() and cpufreq_resume(), respectively,
during system suspend, after interrupts have been switched off on the
boot CPU, and during system resume, while interrupts are still off on
the boot CPU. In both cases the other CPUs are off-line at the
relevant point (either they have been switched off via CPU hotplug
during suspend, or they haven't been switched on yet during resume).
For this reason, although it may seem that cpufreq_suspend() and
cpufreq_resume() are executed for all CPUs in the system, they are
only called for the boot CPU in fact, which is quite confusing.
To remove the confusion and to prepare for elimiating sysdev
suspend and resume operations from the kernel enirely, convernt
cpufreq to using a struct syscore_ops object for the boot CPU
suspend and resume and rename the callbacks so that their names
reflect their purpose. In addition, put some explanatory remarks
into their kerneldoc comments.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
2011-03-24 05:16:32 +08:00
|
|
|
};
|
|
|
|
|
2014-03-04 11:00:27 +08:00
|
|
|
/*
|
|
|
|
* In case platform wants some specific frequency to be configured
|
|
|
|
* during suspend..
|
|
|
|
*/
|
|
|
|
int cpufreq_generic_suspend(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
if (!policy->suspend_freq) {
|
2015-09-09 00:41:02 +08:00
|
|
|
pr_debug("%s: suspend_freq not defined\n", __func__);
|
|
|
|
return 0;
|
2014-03-04 11:00:27 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
pr_debug("%s: Setting suspend-freq: %u\n", __func__,
|
|
|
|
policy->suspend_freq);
|
|
|
|
|
|
|
|
ret = __cpufreq_driver_target(policy, policy->suspend_freq,
|
|
|
|
CPUFREQ_RELATION_H);
|
|
|
|
if (ret)
|
|
|
|
pr_err("%s: unable to set suspend-freq: %u. err: %d\n",
|
|
|
|
__func__, policy->suspend_freq, ret);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_generic_suspend);
|
|
|
|
|
2005-04-29 22:40:12 +08:00
|
|
|
/**
|
2020-11-19 03:02:42 +08:00
|
|
|
* cpufreq_suspend() - Suspend CPUFreq governors.
|
cpufreq: Use syscore_ops for boot CPU suspend/resume (v2)
The cpufreq subsystem uses sysdev suspend and resume for
executing cpufreq_suspend() and cpufreq_resume(), respectively,
during system suspend, after interrupts have been switched off on the
boot CPU, and during system resume, while interrupts are still off on
the boot CPU. In both cases the other CPUs are off-line at the
relevant point (either they have been switched off via CPU hotplug
during suspend, or they haven't been switched on yet during resume).
For this reason, although it may seem that cpufreq_suspend() and
cpufreq_resume() are executed for all CPUs in the system, they are
only called for the boot CPU in fact, which is quite confusing.
To remove the confusion and to prepare for elimiating sysdev
suspend and resume operations from the kernel enirely, convernt
cpufreq to using a struct syscore_ops object for the boot CPU
suspend and resume and rename the callbacks so that their names
reflect their purpose. In addition, put some explanatory remarks
into their kerneldoc comments.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
2011-03-24 05:16:32 +08:00
|
|
|
*
|
2014-03-04 11:00:26 +08:00
|
|
|
* Called during system wide Suspend/Hibernate cycles for suspending governors
|
|
|
|
* as some platforms can't change frequency after this point in suspend cycle.
|
|
|
|
* Because some of the devices (like: i2c, regulators, etc) they use for
|
|
|
|
* changing frequency are suspended quickly after this point.
|
2005-04-29 22:40:12 +08:00
|
|
|
*/
|
2014-03-04 11:00:26 +08:00
|
|
|
void cpufreq_suspend(void)
|
2005-04-29 22:40:12 +08:00
|
|
|
{
|
2013-08-07 01:23:05 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2005-04-29 22:40:12 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
if (!cpufreq_driver)
|
|
|
|
return;
|
2005-04-29 22:40:12 +08:00
|
|
|
|
2016-05-02 08:27:19 +08:00
|
|
|
if (!has_target() && !cpufreq_driver->suspend)
|
2014-09-30 12:03:17 +08:00
|
|
|
goto suspend;
|
2005-04-29 22:40:12 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
pr_debug("%s: Suspending Governors\n", __func__);
|
|
|
|
|
2015-05-12 14:50:11 +08:00
|
|
|
for_each_active_policy(policy) {
|
2016-05-02 08:27:19 +08:00
|
|
|
if (has_target()) {
|
|
|
|
down_write(&policy->rwsem);
|
2016-05-12 21:14:12 +08:00
|
|
|
cpufreq_stop_governor(policy);
|
2016-05-02 08:27:19 +08:00
|
|
|
up_write(&policy->rwsem);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cpufreq_driver->suspend && cpufreq_driver->suspend(policy))
|
2019-08-22 07:16:32 +08:00
|
|
|
pr_err("%s: Failed to suspend driver: %s\n", __func__,
|
|
|
|
cpufreq_driver->name);
|
2005-04-29 22:40:12 +08:00
|
|
|
}
|
2014-09-30 12:03:17 +08:00
|
|
|
|
|
|
|
suspend:
|
|
|
|
cpufreq_suspended = true;
|
2005-04-29 22:40:12 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
2020-11-19 03:02:42 +08:00
|
|
|
* cpufreq_resume() - Resume CPUFreq governors.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2014-03-04 11:00:26 +08:00
|
|
|
* Called during system wide Suspend/Hibernate cycle for resuming governors that
|
|
|
|
* are suspended with cpufreq_suspend().
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2014-03-04 11:00:26 +08:00
|
|
|
void cpufreq_resume(void)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2013-08-07 01:23:05 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2016-02-11 20:01:12 +08:00
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
if (!cpufreq_driver)
|
cpufreq: Skip cpufreq resume if it's not suspended
cpufreq_resume can be called even without preceding cpufreq_suspend.
This can happen in following scenario:
suspend_devices_and_enter
--> dpm_suspend_start
--> dpm_prepare
--> device_prepare : this function errors out
--> dpm_suspend: this is skipped due to dpm_prepare failure
this means cpufreq_suspend is skipped over
--> goto Recover_platform, due to previous error
--> goto Resume_devices
--> dpm_resume_end
--> dpm_resume
--> cpufreq_resume
In case schedutil is used as frequency governor, cpufreq_resume will
eventually call sugov_start, which does following:
memset(sg_cpu, 0, sizeof(*sg_cpu));
....
This effectively erases function pointer for frequency update, causing
crash later on. The function pointer would have been set correctly if
subsequent cpufreq_add_update_util_hook runs successfully, but that
function returns earlier because cpufreq_suspend was not called:
if (WARN_ON(per_cpu(cpufreq_update_util_data, cpu)))
return;
The fix is to check cpufreq_suspended first, if it's false, that means
cpufreq_suspend was not called in the first place, so do not resume
cpufreq.
Signed-off-by: Bo Yan <byan@nvidia.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Dropped printing a message ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2018-01-24 05:57:55 +08:00
|
|
|
return;
|
|
|
|
|
|
|
|
if (unlikely(!cpufreq_suspended))
|
2014-03-04 11:00:26 +08:00
|
|
|
return;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-09-18 15:03:07 +08:00
|
|
|
cpufreq_suspended = false;
|
|
|
|
|
2016-05-02 08:27:19 +08:00
|
|
|
if (!has_target() && !cpufreq_driver->resume)
|
cpufreq: Use syscore_ops for boot CPU suspend/resume (v2)
The cpufreq subsystem uses sysdev suspend and resume for
executing cpufreq_suspend() and cpufreq_resume(), respectively,
during system suspend, after interrupts have been switched off on the
boot CPU, and during system resume, while interrupts are still off on
the boot CPU. In both cases the other CPUs are off-line at the
relevant point (either they have been switched off via CPU hotplug
during suspend, or they haven't been switched on yet during resume).
For this reason, although it may seem that cpufreq_suspend() and
cpufreq_resume() are executed for all CPUs in the system, they are
only called for the boot CPU in fact, which is quite confusing.
To remove the confusion and to prepare for elimiating sysdev
suspend and resume operations from the kernel enirely, convernt
cpufreq to using a struct syscore_ops object for the boot CPU
suspend and resume and rename the callbacks so that their names
reflect their purpose. In addition, put some explanatory remarks
into their kerneldoc comments.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
2011-03-24 05:16:32 +08:00
|
|
|
return;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
pr_debug("%s: Resuming Governors\n", __func__);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-05-12 14:50:11 +08:00
|
|
|
for_each_active_policy(policy) {
|
2016-02-11 20:01:12 +08:00
|
|
|
if (cpufreq_driver->resume && cpufreq_driver->resume(policy)) {
|
2014-03-24 15:00:29 +08:00
|
|
|
pr_err("%s: Failed to resume driver: %p\n", __func__,
|
|
|
|
policy);
|
2016-05-02 08:27:19 +08:00
|
|
|
} else if (has_target()) {
|
2016-02-11 20:01:12 +08:00
|
|
|
down_write(&policy->rwsem);
|
2016-03-21 22:45:24 +08:00
|
|
|
ret = cpufreq_start_governor(policy);
|
2016-02-11 20:01:12 +08:00
|
|
|
up_write(&policy->rwsem);
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
pr_err("%s: Failed to start governor for policy: %p\n",
|
|
|
|
__func__, policy);
|
|
|
|
}
|
2014-03-04 11:00:26 +08:00
|
|
|
}
|
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2020-10-23 23:35:46 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_driver_test_flags - Test cpufreq driver's flags against given ones.
|
|
|
|
* @flags: Flags to test against the current cpufreq driver's flags.
|
|
|
|
*
|
|
|
|
* Assumes that the driver is there, so callers must ensure that this is the
|
|
|
|
* case.
|
|
|
|
*/
|
|
|
|
bool cpufreq_driver_test_flags(u16 flags)
|
|
|
|
{
|
|
|
|
return !!(cpufreq_driver->flags & flags);
|
|
|
|
}
|
|
|
|
|
2013-01-20 18:24:28 +08:00
|
|
|
/**
|
2020-11-19 03:02:42 +08:00
|
|
|
* cpufreq_get_current_driver - Return the current driver's name.
|
2013-01-20 18:24:28 +08:00
|
|
|
*
|
2020-11-19 03:02:42 +08:00
|
|
|
* Return the name string of the currently registered cpufreq driver or NULL if
|
|
|
|
* none.
|
2013-01-20 18:24:28 +08:00
|
|
|
*/
|
|
|
|
const char *cpufreq_get_current_driver(void)
|
|
|
|
{
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver)
|
|
|
|
return cpufreq_driver->name;
|
|
|
|
|
|
|
|
return NULL;
|
2013-01-20 18:24:28 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_get_current_driver);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-10-19 17:30:27 +08:00
|
|
|
/**
|
2020-11-19 03:02:42 +08:00
|
|
|
* cpufreq_get_driver_data - Return current driver data.
|
2014-10-19 17:30:27 +08:00
|
|
|
*
|
2020-11-19 03:02:42 +08:00
|
|
|
* Return the private data of the currently registered cpufreq driver, or NULL
|
|
|
|
* if no cpufreq driver has been registered.
|
2014-10-19 17:30:27 +08:00
|
|
|
*/
|
|
|
|
void *cpufreq_get_driver_data(void)
|
|
|
|
{
|
|
|
|
if (cpufreq_driver)
|
|
|
|
return cpufreq_driver->driver_data;
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_get_driver_data);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*********************************************************************
|
|
|
|
* NOTIFIER LISTS INTERFACE *
|
|
|
|
*********************************************************************/
|
|
|
|
|
|
|
|
/**
|
2020-11-19 03:02:42 +08:00
|
|
|
* cpufreq_register_notifier - Register a notifier with cpufreq.
|
|
|
|
* @nb: notifier function to register.
|
|
|
|
* @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2020-11-19 03:02:42 +08:00
|
|
|
* Add a notifier to one of two lists: either a list of notifiers that run on
|
|
|
|
* clock rate changes (once before and once after every transition), or a list
|
|
|
|
* of notifiers that ron on cpufreq policy changes.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2020-11-19 03:02:42 +08:00
|
|
|
* This function may sleep and it has the same return values as
|
|
|
|
* blocking_notifier_chain_register().
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
int cpufreq_register_notifier(struct notifier_block *nb, unsigned int list)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2013-01-18 00:22:21 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -EINVAL;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (list) {
|
|
|
|
case CPUFREQ_TRANSITION_NOTIFIER:
|
2016-03-30 09:47:49 +08:00
|
|
|
mutex_lock(&cpufreq_fast_switch_lock);
|
|
|
|
|
|
|
|
if (cpufreq_fast_switch_count > 0) {
|
|
|
|
mutex_unlock(&cpufreq_fast_switch_lock);
|
|
|
|
return -EBUSY;
|
|
|
|
}
|
2006-10-04 17:17:06 +08:00
|
|
|
ret = srcu_notifier_chain_register(
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
&cpufreq_transition_notifier_list, nb);
|
2016-03-30 09:47:49 +08:00
|
|
|
if (!ret)
|
|
|
|
cpufreq_fast_switch_count--;
|
|
|
|
|
|
|
|
mutex_unlock(&cpufreq_fast_switch_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case CPUFREQ_POLICY_NOTIFIER:
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
ret = blocking_notifier_chain_register(
|
|
|
|
&cpufreq_policy_notifier_list, nb);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
ret = -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_register_notifier);
|
|
|
|
|
|
|
|
/**
|
2020-11-19 03:02:42 +08:00
|
|
|
* cpufreq_unregister_notifier - Unregister a notifier from cpufreq.
|
|
|
|
* @nb: notifier block to be unregistered.
|
|
|
|
* @list: CPUFREQ_TRANSITION_NOTIFIER or CPUFREQ_POLICY_NOTIFIER.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2020-11-19 03:02:42 +08:00
|
|
|
* Remove a notifier from one of the cpufreq notifier lists.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2020-11-19 03:02:42 +08:00
|
|
|
* This function may sleep and it has the same return values as
|
|
|
|
* blocking_notifier_chain_unregister().
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
|
|
|
int cpufreq_unregister_notifier(struct notifier_block *nb, unsigned int list)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2013-01-18 00:22:21 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -EINVAL;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
switch (list) {
|
|
|
|
case CPUFREQ_TRANSITION_NOTIFIER:
|
2016-03-30 09:47:49 +08:00
|
|
|
mutex_lock(&cpufreq_fast_switch_lock);
|
|
|
|
|
2006-10-04 17:17:06 +08:00
|
|
|
ret = srcu_notifier_chain_unregister(
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
&cpufreq_transition_notifier_list, nb);
|
2016-03-30 09:47:49 +08:00
|
|
|
if (!ret && !WARN_ON(cpufreq_fast_switch_count >= 0))
|
|
|
|
cpufreq_fast_switch_count++;
|
|
|
|
|
|
|
|
mutex_unlock(&cpufreq_fast_switch_lock);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
case CPUFREQ_POLICY_NOTIFIER:
|
[PATCH] Notifier chain update: API changes
The kernel's implementation of notifier chains is unsafe. There is no
protection against entries being added to or removed from a chain while the
chain is in use. The issues were discussed in this thread:
http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2
We noticed that notifier chains in the kernel fall into two basic usage
classes:
"Blocking" chains are always called from a process context
and the callout routines are allowed to sleep;
"Atomic" chains can be called from an atomic context and
the callout routines are not allowed to sleep.
We decided to codify this distinction and make it part of the API. Therefore
this set of patches introduces three new, parallel APIs: one for blocking
notifiers, one for atomic notifiers, and one for "raw" notifiers (which is
really just the old API under a new name). New kinds of data structures are
used for the heads of the chains, and new routines are defined for
registration, unregistration, and calling a chain. The three APIs are
explained in include/linux/notifier.h and their implementation is in
kernel/sys.c.
With atomic and blocking chains, the implementation guarantees that the chain
links will not be corrupted and that chain callers will not get messed up by
entries being added or removed. For raw chains the implementation provides no
guarantees at all; users of this API must provide their own protections. (The
idea was that situations may come up where the assumptions of the atomic and
blocking APIs are not appropriate, so it should be possible for users to
handle these things in their own way.)
There are some limitations, which should not be too hard to live with. For
atomic/blocking chains, registration and unregistration must always be done in
a process context since the chain is protected by a mutex/rwsem. Also, a
callout routine for a non-raw chain must not try to register or unregister
entries on its own chain. (This did happen in a couple of places and the code
had to be changed to avoid it.)
Since atomic chains may be called from within an NMI handler, they cannot use
spinlocks for synchronization. Instead we use RCU. The overhead falls almost
entirely in the unregister routine, which is okay since unregistration is much
less frequent that calling a chain.
Here is the list of chains that we adjusted and their classifications. None
of them use the raw API, so for the moment it is only a placeholder.
ATOMIC CHAINS
-------------
arch/i386/kernel/traps.c: i386die_chain
arch/ia64/kernel/traps.c: ia64die_chain
arch/powerpc/kernel/traps.c: powerpc_die_chain
arch/sparc64/kernel/traps.c: sparc64die_chain
arch/x86_64/kernel/traps.c: die_chain
drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list
kernel/panic.c: panic_notifier_list
kernel/profile.c: task_free_notifier
net/bluetooth/hci_core.c: hci_notifier
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain
net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain
net/ipv6/addrconf.c: inet6addr_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_chain
net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain
net/netlink/af_netlink.c: netlink_chain
BLOCKING CHAINS
---------------
arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain
arch/s390/kernel/process.c: idle_chain
arch/x86_64/kernel/process.c idle_notifier
drivers/base/memory.c: memory_chain
drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list
drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list
drivers/macintosh/adb.c: adb_client_list
drivers/macintosh/via-pmu.c sleep_notifier_list
drivers/macintosh/via-pmu68k.c sleep_notifier_list
drivers/macintosh/windfarm_core.c wf_client_list
drivers/usb/core/notify.c usb_notifier_list
drivers/video/fbmem.c fb_notifier_list
kernel/cpu.c cpu_chain
kernel/module.c module_notify_list
kernel/profile.c munmap_notifier
kernel/profile.c task_exit_notifier
kernel/sys.c reboot_notifier_list
net/core/dev.c netdev_chain
net/decnet/dn_dev.c: dnaddr_chain
net/ipv4/devinet.c: inetaddr_chain
It's possible that some of these classifications are wrong. If they are,
please let us know or submit a patch to fix them. Note that any chain that
gets called very frequently should be atomic, because the rwsem read-locking
used for blocking chains is very likely to incur cache misses on SMP systems.
(However, if the chain's callout routines may sleep then the chain cannot be
atomic.)
The patch set was written by Alan Stern and Chandra Seetharaman, incorporating
material written by Keith Owens and suggestions from Paul McKenney and Andrew
Morton.
[jes@sgi.com: restructure the notifier chain initialization macros]
Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
|
|
|
ret = blocking_notifier_chain_unregister(
|
|
|
|
&cpufreq_policy_notifier_list, nb);
|
2005-04-17 06:20:36 +08:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
ret = -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_unregister_notifier);
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* GOVERNORS *
|
|
|
|
*********************************************************************/
|
|
|
|
|
2016-03-30 09:47:49 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_driver_fast_switch - Carry out a fast CPU frequency switch.
|
|
|
|
* @policy: cpufreq policy to switch the frequency for.
|
|
|
|
* @target_freq: New frequency to set (may be approximate).
|
|
|
|
*
|
|
|
|
* Carry out a fast frequency switch without sleeping.
|
|
|
|
*
|
|
|
|
* The driver's ->fast_switch() callback invoked by this function must be
|
|
|
|
* suitable for being called from within RCU-sched read-side critical sections
|
|
|
|
* and it is expected to select the minimum available frequency greater than or
|
|
|
|
* equal to @target_freq (CPUFREQ_RELATION_L).
|
|
|
|
*
|
|
|
|
* This function must not be called if policy->fast_switch_enabled is unset.
|
|
|
|
*
|
|
|
|
* Governors calling this function must guarantee that it will never be invoked
|
|
|
|
* twice in parallel for the same policy and that it will never be called in
|
|
|
|
* parallel with either ->target() or ->target_index() for the same policy.
|
|
|
|
*
|
2017-08-09 12:51:46 +08:00
|
|
|
* Returns the actual frequency set for the CPU.
|
|
|
|
*
|
|
|
|
* If 0 is returned by the driver's ->fast_switch() callback to indicate an
|
|
|
|
* error condition, the hardware configuration must be preserved.
|
2016-03-30 09:47:49 +08:00
|
|
|
*/
|
|
|
|
unsigned int cpufreq_driver_fast_switch(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq)
|
|
|
|
{
|
cpufreq: move invariance setter calls in cpufreq core
To properly scale its per-entity load-tracking signals, the task scheduler
needs to be given a frequency scale factor, i.e. some image of the current
frequency the CPU is running at. Currently, this scale can be computed
either by using counters (APERF/MPERF on x86, AMU on arm64), or by
piggy-backing on the frequency selection done by cpufreq.
For the latter, drivers have to explicitly set the scale factor
themselves, despite it being purely boiler-plate code: the required
information depends entirely on the kind of frequency switch callback
implemented by the driver, i.e. either of: target_index(), target(),
fast_switch() and setpolicy().
The fitness of those callbacks with regard to driving the Frequency
Invariance Engine (FIE) is studied below:
target_index()
==============
Documentation states that the chosen frequency "must be determined by
freq_table[index].frequency". It isn't clear if it *has* to be that
frequency, or if it can use that frequency value to do some computation
that ultimately leads to a different frequency selection. All drivers
go for the former, while the vexpress-spc-cpufreq has an atypical
implementation which is handled separately.
Therefore, the hook works on the assumption the core can use
freq_table[index].frequency.
target()
=======
This has been flagged as deprecated since:
commit 9c0ebcf78fde ("cpufreq: Implement light weight ->target_index() routine")
It also doesn't have that many users:
gx-suspmod.c:439: .target = cpufreq_gx_target,
s3c24xx-cpufreq.c:428: .target = s3c_cpufreq_target,
intel_pstate.c:2528: .target = intel_cpufreq_target,
cppc_cpufreq.c:401: .target = cppc_cpufreq_set_target,
cpufreq-nforce2.c:371: .target = nforce2_target,
sh-cpufreq.c:163: .target = sh_cpufreq_target,
pcc-cpufreq.c:573: .target = pcc_cpufreq_target,
Similarly to the path taken for target_index() calls in the cpufreq core
during a frequency change, all of the drivers above will mark the end of a
frequency change by a call to cpufreq_freq_transition_end().
Therefore, cpufreq_freq_transition_end() can be used as the location for
the arch_set_freq_scale() call to potentially inform the scheduler of the
frequency change.
This change maintains the previous functionality for the drivers that
implement the target_index() callback, while also adding support for the
few drivers that implement the deprecated target() callback.
fast_switch()
=============
This callback *has* to return the frequency that was selected.
setpolicy()
===========
This callback does not have any designated way of informing what was the
end choice. But there are only two drivers using setpolicy(), and none
of them have current FIE support:
drivers/cpufreq/longrun.c:281: .setpolicy = longrun_set_policy,
drivers/cpufreq/intel_pstate.c:2215: .setpolicy = intel_pstate_set_policy,
The intel_pstate is known to use counter-driven frequency invariance.
Conclusion
==========
Given that the significant majority of current FIE enabled drivers use
callbacks that lend themselves to triggering the setting of the FIE scale
factor in a generic way, move the invariance setter calls to cpufreq core.
As a result of setting the frequency scale factor in cpufreq core, after
callbacks that lend themselves to trigger it, remove this functionality
from the driver side.
To be noted that despite marking a successful frequency change, many
cpufreq drivers will consider the new frequency as the requested
frequency, although this is might not be the one granted by the hardware.
Therefore, the call to arch_set_freq_scale() is a "best effort" one, and
it is up to the architecture if the new frequency is used in the new
frequency scale factor setting (determined by the implementation of
arch_set_freq_scale()) or eventually used by the scheduler (determined
by the implementation of arch_scale_freq_capacity()). The architecture
is in a better position to decide if it has better methods to obtain
more accurate information regarding the current frequency and use that
information instead (for example, the use of counters).
Also, the implementation to arch_set_freq_scale() will now have to handle
error conditions (current frequency == 0) in order to prevent the
overhead in cpufreq core when the default arch_set_freq_scale()
implementation is used.
Signed-off-by: Ionela Voinescu <ionela.voinescu@arm.com>
Suggested-by: Valentin Schneider <valentin.schneider@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-09-02 04:55:46 +08:00
|
|
|
unsigned int freq;
|
2020-10-05 15:56:05 +08:00
|
|
|
int cpu;
|
cpufreq: move invariance setter calls in cpufreq core
To properly scale its per-entity load-tracking signals, the task scheduler
needs to be given a frequency scale factor, i.e. some image of the current
frequency the CPU is running at. Currently, this scale can be computed
either by using counters (APERF/MPERF on x86, AMU on arm64), or by
piggy-backing on the frequency selection done by cpufreq.
For the latter, drivers have to explicitly set the scale factor
themselves, despite it being purely boiler-plate code: the required
information depends entirely on the kind of frequency switch callback
implemented by the driver, i.e. either of: target_index(), target(),
fast_switch() and setpolicy().
The fitness of those callbacks with regard to driving the Frequency
Invariance Engine (FIE) is studied below:
target_index()
==============
Documentation states that the chosen frequency "must be determined by
freq_table[index].frequency". It isn't clear if it *has* to be that
frequency, or if it can use that frequency value to do some computation
that ultimately leads to a different frequency selection. All drivers
go for the former, while the vexpress-spc-cpufreq has an atypical
implementation which is handled separately.
Therefore, the hook works on the assumption the core can use
freq_table[index].frequency.
target()
=======
This has been flagged as deprecated since:
commit 9c0ebcf78fde ("cpufreq: Implement light weight ->target_index() routine")
It also doesn't have that many users:
gx-suspmod.c:439: .target = cpufreq_gx_target,
s3c24xx-cpufreq.c:428: .target = s3c_cpufreq_target,
intel_pstate.c:2528: .target = intel_cpufreq_target,
cppc_cpufreq.c:401: .target = cppc_cpufreq_set_target,
cpufreq-nforce2.c:371: .target = nforce2_target,
sh-cpufreq.c:163: .target = sh_cpufreq_target,
pcc-cpufreq.c:573: .target = pcc_cpufreq_target,
Similarly to the path taken for target_index() calls in the cpufreq core
during a frequency change, all of the drivers above will mark the end of a
frequency change by a call to cpufreq_freq_transition_end().
Therefore, cpufreq_freq_transition_end() can be used as the location for
the arch_set_freq_scale() call to potentially inform the scheduler of the
frequency change.
This change maintains the previous functionality for the drivers that
implement the target_index() callback, while also adding support for the
few drivers that implement the deprecated target() callback.
fast_switch()
=============
This callback *has* to return the frequency that was selected.
setpolicy()
===========
This callback does not have any designated way of informing what was the
end choice. But there are only two drivers using setpolicy(), and none
of them have current FIE support:
drivers/cpufreq/longrun.c:281: .setpolicy = longrun_set_policy,
drivers/cpufreq/intel_pstate.c:2215: .setpolicy = intel_pstate_set_policy,
The intel_pstate is known to use counter-driven frequency invariance.
Conclusion
==========
Given that the significant majority of current FIE enabled drivers use
callbacks that lend themselves to triggering the setting of the FIE scale
factor in a generic way, move the invariance setter calls to cpufreq core.
As a result of setting the frequency scale factor in cpufreq core, after
callbacks that lend themselves to trigger it, remove this functionality
from the driver side.
To be noted that despite marking a successful frequency change, many
cpufreq drivers will consider the new frequency as the requested
frequency, although this is might not be the one granted by the hardware.
Therefore, the call to arch_set_freq_scale() is a "best effort" one, and
it is up to the architecture if the new frequency is used in the new
frequency scale factor setting (determined by the implementation of
arch_set_freq_scale()) or eventually used by the scheduler (determined
by the implementation of arch_scale_freq_capacity()). The architecture
is in a better position to decide if it has better methods to obtain
more accurate information regarding the current frequency and use that
information instead (for example, the use of counters).
Also, the implementation to arch_set_freq_scale() will now have to handle
error conditions (current frequency == 0) in order to prevent the
overhead in cpufreq core when the default arch_set_freq_scale()
implementation is used.
Signed-off-by: Ionela Voinescu <ionela.voinescu@arm.com>
Suggested-by: Valentin Schneider <valentin.schneider@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-09-02 04:55:46 +08:00
|
|
|
|
2016-06-02 04:36:26 +08:00
|
|
|
target_freq = clamp_val(target_freq, policy->min, policy->max);
|
cpufreq: move invariance setter calls in cpufreq core
To properly scale its per-entity load-tracking signals, the task scheduler
needs to be given a frequency scale factor, i.e. some image of the current
frequency the CPU is running at. Currently, this scale can be computed
either by using counters (APERF/MPERF on x86, AMU on arm64), or by
piggy-backing on the frequency selection done by cpufreq.
For the latter, drivers have to explicitly set the scale factor
themselves, despite it being purely boiler-plate code: the required
information depends entirely on the kind of frequency switch callback
implemented by the driver, i.e. either of: target_index(), target(),
fast_switch() and setpolicy().
The fitness of those callbacks with regard to driving the Frequency
Invariance Engine (FIE) is studied below:
target_index()
==============
Documentation states that the chosen frequency "must be determined by
freq_table[index].frequency". It isn't clear if it *has* to be that
frequency, or if it can use that frequency value to do some computation
that ultimately leads to a different frequency selection. All drivers
go for the former, while the vexpress-spc-cpufreq has an atypical
implementation which is handled separately.
Therefore, the hook works on the assumption the core can use
freq_table[index].frequency.
target()
=======
This has been flagged as deprecated since:
commit 9c0ebcf78fde ("cpufreq: Implement light weight ->target_index() routine")
It also doesn't have that many users:
gx-suspmod.c:439: .target = cpufreq_gx_target,
s3c24xx-cpufreq.c:428: .target = s3c_cpufreq_target,
intel_pstate.c:2528: .target = intel_cpufreq_target,
cppc_cpufreq.c:401: .target = cppc_cpufreq_set_target,
cpufreq-nforce2.c:371: .target = nforce2_target,
sh-cpufreq.c:163: .target = sh_cpufreq_target,
pcc-cpufreq.c:573: .target = pcc_cpufreq_target,
Similarly to the path taken for target_index() calls in the cpufreq core
during a frequency change, all of the drivers above will mark the end of a
frequency change by a call to cpufreq_freq_transition_end().
Therefore, cpufreq_freq_transition_end() can be used as the location for
the arch_set_freq_scale() call to potentially inform the scheduler of the
frequency change.
This change maintains the previous functionality for the drivers that
implement the target_index() callback, while also adding support for the
few drivers that implement the deprecated target() callback.
fast_switch()
=============
This callback *has* to return the frequency that was selected.
setpolicy()
===========
This callback does not have any designated way of informing what was the
end choice. But there are only two drivers using setpolicy(), and none
of them have current FIE support:
drivers/cpufreq/longrun.c:281: .setpolicy = longrun_set_policy,
drivers/cpufreq/intel_pstate.c:2215: .setpolicy = intel_pstate_set_policy,
The intel_pstate is known to use counter-driven frequency invariance.
Conclusion
==========
Given that the significant majority of current FIE enabled drivers use
callbacks that lend themselves to triggering the setting of the FIE scale
factor in a generic way, move the invariance setter calls to cpufreq core.
As a result of setting the frequency scale factor in cpufreq core, after
callbacks that lend themselves to trigger it, remove this functionality
from the driver side.
To be noted that despite marking a successful frequency change, many
cpufreq drivers will consider the new frequency as the requested
frequency, although this is might not be the one granted by the hardware.
Therefore, the call to arch_set_freq_scale() is a "best effort" one, and
it is up to the architecture if the new frequency is used in the new
frequency scale factor setting (determined by the implementation of
arch_set_freq_scale()) or eventually used by the scheduler (determined
by the implementation of arch_scale_freq_capacity()). The architecture
is in a better position to decide if it has better methods to obtain
more accurate information regarding the current frequency and use that
information instead (for example, the use of counters).
Also, the implementation to arch_set_freq_scale() will now have to handle
error conditions (current frequency == 0) in order to prevent the
overhead in cpufreq core when the default arch_set_freq_scale()
implementation is used.
Signed-off-by: Ionela Voinescu <ionela.voinescu@arm.com>
Suggested-by: Valentin Schneider <valentin.schneider@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-09-02 04:55:46 +08:00
|
|
|
freq = cpufreq_driver->fast_switch(policy, target_freq);
|
|
|
|
|
2020-10-05 15:56:04 +08:00
|
|
|
if (!freq)
|
|
|
|
return 0;
|
|
|
|
|
2020-10-05 15:56:05 +08:00
|
|
|
policy->cur = freq;
|
cpufreq: move invariance setter calls in cpufreq core
To properly scale its per-entity load-tracking signals, the task scheduler
needs to be given a frequency scale factor, i.e. some image of the current
frequency the CPU is running at. Currently, this scale can be computed
either by using counters (APERF/MPERF on x86, AMU on arm64), or by
piggy-backing on the frequency selection done by cpufreq.
For the latter, drivers have to explicitly set the scale factor
themselves, despite it being purely boiler-plate code: the required
information depends entirely on the kind of frequency switch callback
implemented by the driver, i.e. either of: target_index(), target(),
fast_switch() and setpolicy().
The fitness of those callbacks with regard to driving the Frequency
Invariance Engine (FIE) is studied below:
target_index()
==============
Documentation states that the chosen frequency "must be determined by
freq_table[index].frequency". It isn't clear if it *has* to be that
frequency, or if it can use that frequency value to do some computation
that ultimately leads to a different frequency selection. All drivers
go for the former, while the vexpress-spc-cpufreq has an atypical
implementation which is handled separately.
Therefore, the hook works on the assumption the core can use
freq_table[index].frequency.
target()
=======
This has been flagged as deprecated since:
commit 9c0ebcf78fde ("cpufreq: Implement light weight ->target_index() routine")
It also doesn't have that many users:
gx-suspmod.c:439: .target = cpufreq_gx_target,
s3c24xx-cpufreq.c:428: .target = s3c_cpufreq_target,
intel_pstate.c:2528: .target = intel_cpufreq_target,
cppc_cpufreq.c:401: .target = cppc_cpufreq_set_target,
cpufreq-nforce2.c:371: .target = nforce2_target,
sh-cpufreq.c:163: .target = sh_cpufreq_target,
pcc-cpufreq.c:573: .target = pcc_cpufreq_target,
Similarly to the path taken for target_index() calls in the cpufreq core
during a frequency change, all of the drivers above will mark the end of a
frequency change by a call to cpufreq_freq_transition_end().
Therefore, cpufreq_freq_transition_end() can be used as the location for
the arch_set_freq_scale() call to potentially inform the scheduler of the
frequency change.
This change maintains the previous functionality for the drivers that
implement the target_index() callback, while also adding support for the
few drivers that implement the deprecated target() callback.
fast_switch()
=============
This callback *has* to return the frequency that was selected.
setpolicy()
===========
This callback does not have any designated way of informing what was the
end choice. But there are only two drivers using setpolicy(), and none
of them have current FIE support:
drivers/cpufreq/longrun.c:281: .setpolicy = longrun_set_policy,
drivers/cpufreq/intel_pstate.c:2215: .setpolicy = intel_pstate_set_policy,
The intel_pstate is known to use counter-driven frequency invariance.
Conclusion
==========
Given that the significant majority of current FIE enabled drivers use
callbacks that lend themselves to triggering the setting of the FIE scale
factor in a generic way, move the invariance setter calls to cpufreq core.
As a result of setting the frequency scale factor in cpufreq core, after
callbacks that lend themselves to trigger it, remove this functionality
from the driver side.
To be noted that despite marking a successful frequency change, many
cpufreq drivers will consider the new frequency as the requested
frequency, although this is might not be the one granted by the hardware.
Therefore, the call to arch_set_freq_scale() is a "best effort" one, and
it is up to the architecture if the new frequency is used in the new
frequency scale factor setting (determined by the implementation of
arch_set_freq_scale()) or eventually used by the scheduler (determined
by the implementation of arch_scale_freq_capacity()). The architecture
is in a better position to decide if it has better methods to obtain
more accurate information regarding the current frequency and use that
information instead (for example, the use of counters).
Also, the implementation to arch_set_freq_scale() will now have to handle
error conditions (current frequency == 0) in order to prevent the
overhead in cpufreq core when the default arch_set_freq_scale()
implementation is used.
Signed-off-by: Ionela Voinescu <ionela.voinescu@arm.com>
Suggested-by: Valentin Schneider <valentin.schneider@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-09-02 04:55:46 +08:00
|
|
|
arch_set_freq_scale(policy->related_cpus, freq,
|
|
|
|
policy->cpuinfo.max_freq);
|
2020-10-05 15:56:04 +08:00
|
|
|
cpufreq_stats_record_transition(policy, freq);
|
2016-03-30 09:47:49 +08:00
|
|
|
|
2020-10-05 15:56:05 +08:00
|
|
|
if (trace_cpu_frequency_enabled()) {
|
|
|
|
for_each_cpu(cpu, policy->cpus)
|
|
|
|
trace_cpu_frequency(freq, cpu);
|
|
|
|
}
|
|
|
|
|
cpufreq: move invariance setter calls in cpufreq core
To properly scale its per-entity load-tracking signals, the task scheduler
needs to be given a frequency scale factor, i.e. some image of the current
frequency the CPU is running at. Currently, this scale can be computed
either by using counters (APERF/MPERF on x86, AMU on arm64), or by
piggy-backing on the frequency selection done by cpufreq.
For the latter, drivers have to explicitly set the scale factor
themselves, despite it being purely boiler-plate code: the required
information depends entirely on the kind of frequency switch callback
implemented by the driver, i.e. either of: target_index(), target(),
fast_switch() and setpolicy().
The fitness of those callbacks with regard to driving the Frequency
Invariance Engine (FIE) is studied below:
target_index()
==============
Documentation states that the chosen frequency "must be determined by
freq_table[index].frequency". It isn't clear if it *has* to be that
frequency, or if it can use that frequency value to do some computation
that ultimately leads to a different frequency selection. All drivers
go for the former, while the vexpress-spc-cpufreq has an atypical
implementation which is handled separately.
Therefore, the hook works on the assumption the core can use
freq_table[index].frequency.
target()
=======
This has been flagged as deprecated since:
commit 9c0ebcf78fde ("cpufreq: Implement light weight ->target_index() routine")
It also doesn't have that many users:
gx-suspmod.c:439: .target = cpufreq_gx_target,
s3c24xx-cpufreq.c:428: .target = s3c_cpufreq_target,
intel_pstate.c:2528: .target = intel_cpufreq_target,
cppc_cpufreq.c:401: .target = cppc_cpufreq_set_target,
cpufreq-nforce2.c:371: .target = nforce2_target,
sh-cpufreq.c:163: .target = sh_cpufreq_target,
pcc-cpufreq.c:573: .target = pcc_cpufreq_target,
Similarly to the path taken for target_index() calls in the cpufreq core
during a frequency change, all of the drivers above will mark the end of a
frequency change by a call to cpufreq_freq_transition_end().
Therefore, cpufreq_freq_transition_end() can be used as the location for
the arch_set_freq_scale() call to potentially inform the scheduler of the
frequency change.
This change maintains the previous functionality for the drivers that
implement the target_index() callback, while also adding support for the
few drivers that implement the deprecated target() callback.
fast_switch()
=============
This callback *has* to return the frequency that was selected.
setpolicy()
===========
This callback does not have any designated way of informing what was the
end choice. But there are only two drivers using setpolicy(), and none
of them have current FIE support:
drivers/cpufreq/longrun.c:281: .setpolicy = longrun_set_policy,
drivers/cpufreq/intel_pstate.c:2215: .setpolicy = intel_pstate_set_policy,
The intel_pstate is known to use counter-driven frequency invariance.
Conclusion
==========
Given that the significant majority of current FIE enabled drivers use
callbacks that lend themselves to triggering the setting of the FIE scale
factor in a generic way, move the invariance setter calls to cpufreq core.
As a result of setting the frequency scale factor in cpufreq core, after
callbacks that lend themselves to trigger it, remove this functionality
from the driver side.
To be noted that despite marking a successful frequency change, many
cpufreq drivers will consider the new frequency as the requested
frequency, although this is might not be the one granted by the hardware.
Therefore, the call to arch_set_freq_scale() is a "best effort" one, and
it is up to the architecture if the new frequency is used in the new
frequency scale factor setting (determined by the implementation of
arch_set_freq_scale()) or eventually used by the scheduler (determined
by the implementation of arch_scale_freq_capacity()). The architecture
is in a better position to decide if it has better methods to obtain
more accurate information regarding the current frequency and use that
information instead (for example, the use of counters).
Also, the implementation to arch_set_freq_scale() will now have to handle
error conditions (current frequency == 0) in order to prevent the
overhead in cpufreq core when the default arch_set_freq_scale()
implementation is used.
Signed-off-by: Ionela Voinescu <ionela.voinescu@arm.com>
Suggested-by: Valentin Schneider <valentin.schneider@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Acked-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-09-02 04:55:46 +08:00
|
|
|
return freq;
|
2016-03-30 09:47:49 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_driver_fast_switch);
|
|
|
|
|
2020-12-15 04:08:00 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_driver_adjust_perf - Adjust CPU performance level in one go.
|
|
|
|
* @cpu: Target CPU.
|
|
|
|
* @min_perf: Minimum (required) performance level (units of @capacity).
|
2021-02-18 17:53:38 +08:00
|
|
|
* @target_perf: Target (desired) performance level (units of @capacity).
|
2020-12-15 04:08:00 +08:00
|
|
|
* @capacity: Capacity of the target CPU.
|
|
|
|
*
|
|
|
|
* Carry out a fast performance level switch of @cpu without sleeping.
|
|
|
|
*
|
|
|
|
* The driver's ->adjust_perf() callback invoked by this function must be
|
|
|
|
* suitable for being called from within RCU-sched read-side critical sections
|
|
|
|
* and it is expected to select a suitable performance level equal to or above
|
|
|
|
* @min_perf and preferably equal to or below @target_perf.
|
|
|
|
*
|
|
|
|
* This function must not be called if policy->fast_switch_enabled is unset.
|
|
|
|
*
|
|
|
|
* Governors calling this function must guarantee that it will never be invoked
|
|
|
|
* twice in parallel for the same CPU and that it will never be called in
|
|
|
|
* parallel with either ->target() or ->target_index() or ->fast_switch() for
|
|
|
|
* the same CPU.
|
|
|
|
*/
|
|
|
|
void cpufreq_driver_adjust_perf(unsigned int cpu,
|
|
|
|
unsigned long min_perf,
|
|
|
|
unsigned long target_perf,
|
|
|
|
unsigned long capacity)
|
|
|
|
{
|
|
|
|
cpufreq_driver->adjust_perf(cpu, min_perf, target_perf, capacity);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_driver_has_adjust_perf - Check "direct fast switch" callback.
|
|
|
|
*
|
|
|
|
* Return 'true' if the ->adjust_perf callback is present for the
|
|
|
|
* current driver or 'false' otherwise.
|
|
|
|
*/
|
|
|
|
bool cpufreq_driver_has_adjust_perf(void)
|
|
|
|
{
|
|
|
|
return !!cpufreq_driver->adjust_perf;
|
|
|
|
}
|
|
|
|
|
2014-06-03 01:19:28 +08:00
|
|
|
/* Must set freqs->new to intermediate frequency */
|
|
|
|
static int __target_intermediate(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_freqs *freqs, int index)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
freqs->new = cpufreq_driver->get_intermediate(policy, index);
|
|
|
|
|
|
|
|
/* We don't need to switch to intermediate freq */
|
|
|
|
if (!freqs->new)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
pr_debug("%s: cpu: %d, switching to intermediate freq: oldfreq: %u, intermediate freq: %u\n",
|
|
|
|
__func__, policy->cpu, freqs->old, freqs->new);
|
|
|
|
|
|
|
|
cpufreq_freq_transition_begin(policy, freqs);
|
|
|
|
ret = cpufreq_driver->target_intermediate(policy, index);
|
|
|
|
cpufreq_freq_transition_end(policy, freqs, ret);
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
pr_err("%s: Failed to change to intermediate frequency: %d\n",
|
|
|
|
__func__, ret);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2016-06-03 13:28:50 +08:00
|
|
|
static int __target_index(struct cpufreq_policy *policy, int index)
|
2014-05-21 16:59:29 +08:00
|
|
|
{
|
2014-06-03 01:19:28 +08:00
|
|
|
struct cpufreq_freqs freqs = {.old = policy->cur, .flags = 0};
|
2020-10-22 19:57:45 +08:00
|
|
|
unsigned int restore_freq, intermediate_freq = 0;
|
2016-06-03 13:28:50 +08:00
|
|
|
unsigned int newfreq = policy->freq_table[index].frequency;
|
2014-05-21 16:59:29 +08:00
|
|
|
int retval = -EINVAL;
|
|
|
|
bool notify;
|
|
|
|
|
2016-06-03 13:28:50 +08:00
|
|
|
if (newfreq == policy->cur)
|
|
|
|
return 0;
|
|
|
|
|
2020-10-22 19:57:45 +08:00
|
|
|
/* Save last value to restore later on errors */
|
|
|
|
restore_freq = policy->cur;
|
|
|
|
|
2014-05-21 16:59:29 +08:00
|
|
|
notify = !(cpufreq_driver->flags & CPUFREQ_ASYNC_NOTIFICATION);
|
|
|
|
if (notify) {
|
2014-06-03 01:19:28 +08:00
|
|
|
/* Handle switching to intermediate frequency */
|
|
|
|
if (cpufreq_driver->get_intermediate) {
|
|
|
|
retval = __target_intermediate(policy, &freqs, index);
|
|
|
|
if (retval)
|
|
|
|
return retval;
|
|
|
|
|
|
|
|
intermediate_freq = freqs.new;
|
|
|
|
/* Set old freq to intermediate */
|
|
|
|
if (intermediate_freq)
|
|
|
|
freqs.old = freqs.new;
|
|
|
|
}
|
2014-05-21 16:59:29 +08:00
|
|
|
|
2016-06-03 13:28:50 +08:00
|
|
|
freqs.new = newfreq;
|
2014-05-21 16:59:29 +08:00
|
|
|
pr_debug("%s: cpu: %d, oldfreq: %u, new freq: %u\n",
|
|
|
|
__func__, policy->cpu, freqs.old, freqs.new);
|
|
|
|
|
|
|
|
cpufreq_freq_transition_begin(policy, &freqs);
|
|
|
|
}
|
|
|
|
|
|
|
|
retval = cpufreq_driver->target_index(policy, index);
|
|
|
|
if (retval)
|
|
|
|
pr_err("%s: Failed to change cpu frequency: %d\n", __func__,
|
|
|
|
retval);
|
|
|
|
|
2014-06-03 01:19:28 +08:00
|
|
|
if (notify) {
|
2014-05-21 16:59:29 +08:00
|
|
|
cpufreq_freq_transition_end(policy, &freqs, retval);
|
|
|
|
|
2014-06-03 01:19:28 +08:00
|
|
|
/*
|
|
|
|
* Failed after setting to intermediate freq? Driver should have
|
|
|
|
* reverted back to initial frequency and so should we. Check
|
|
|
|
* here for intermediate_freq instead of get_intermediate, in
|
2015-05-23 01:18:22 +08:00
|
|
|
* case we haven't switched to intermediate freq at all.
|
2014-06-03 01:19:28 +08:00
|
|
|
*/
|
|
|
|
if (unlikely(retval && intermediate_freq)) {
|
|
|
|
freqs.old = intermediate_freq;
|
2020-10-22 19:57:45 +08:00
|
|
|
freqs.new = restore_freq;
|
2014-06-03 01:19:28 +08:00
|
|
|
cpufreq_freq_transition_begin(policy, &freqs);
|
|
|
|
cpufreq_freq_transition_end(policy, &freqs, 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2014-05-21 16:59:29 +08:00
|
|
|
return retval;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
int __cpufreq_driver_target(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq,
|
|
|
|
unsigned int relation)
|
|
|
|
{
|
2012-10-31 08:28:21 +08:00
|
|
|
unsigned int old_target_freq = target_freq;
|
2005-10-31 06:59:54 +08:00
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -ENODEV;
|
|
|
|
|
2021-06-29 14:27:07 +08:00
|
|
|
target_freq = __resolve_freq(policy, target_freq, relation);
|
2012-10-31 08:28:21 +08:00
|
|
|
|
|
|
|
pr_debug("target for CPU %u: %u kHz, relation %u, requested %u kHz\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
policy->cpu, target_freq, relation, old_target_freq);
|
2012-10-31 08:28:15 +08:00
|
|
|
|
2013-10-25 22:15:48 +08:00
|
|
|
/*
|
|
|
|
* This might look like a redundant call as we are checking it again
|
|
|
|
* after finding index. But it is left intentionally for cases where
|
|
|
|
* exactly same freq is called again and so we can save on few function
|
|
|
|
* calls.
|
|
|
|
*/
|
2020-10-23 23:35:19 +08:00
|
|
|
if (target_freq == policy->cur &&
|
|
|
|
!(cpufreq_driver->flags & CPUFREQ_NEED_UPDATE_LIMITS))
|
2012-10-31 08:28:15 +08:00
|
|
|
return 0;
|
|
|
|
|
2021-09-08 22:05:28 +08:00
|
|
|
if (cpufreq_driver->target) {
|
|
|
|
/*
|
|
|
|
* If the driver hasn't setup a single inefficient frequency,
|
|
|
|
* it's unlikely it knows how to decode CPUFREQ_RELATION_E.
|
|
|
|
*/
|
|
|
|
if (!policy->efficiencies_available)
|
|
|
|
relation &= ~CPUFREQ_RELATION_E;
|
|
|
|
|
2016-02-26 07:03:01 +08:00
|
|
|
return cpufreq_driver->target(policy, target_freq, relation);
|
2021-09-08 22:05:28 +08:00
|
|
|
}
|
2013-10-25 22:15:48 +08:00
|
|
|
|
2016-02-26 07:03:01 +08:00
|
|
|
if (!cpufreq_driver->target_index)
|
|
|
|
return -EINVAL;
|
2013-10-25 22:15:48 +08:00
|
|
|
|
2021-06-29 14:27:07 +08:00
|
|
|
return __target_index(policy, policy->cached_resolved_idx);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(__cpufreq_driver_target);
|
|
|
|
|
|
|
|
int cpufreq_driver_target(struct cpufreq_policy *policy,
|
|
|
|
unsigned int target_freq,
|
|
|
|
unsigned int relation)
|
|
|
|
{
|
2019-08-13 20:21:21 +08:00
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
down_write(&policy->rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
ret = __cpufreq_driver_target(policy, target_freq, relation);
|
|
|
|
|
2013-10-18 21:40:15 +08:00
|
|
|
up_write(&policy->rwsem);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_driver_target);
|
|
|
|
|
2016-02-05 09:37:42 +08:00
|
|
|
__weak struct cpufreq_governor *cpufreq_fallback_governor(void)
|
|
|
|
{
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
2016-05-14 07:01:46 +08:00
|
|
|
static int cpufreq_init_governor(struct cpufreq_policy *policy)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2005-07-29 00:43:56 +08:00
|
|
|
int ret;
|
2007-10-03 04:28:13 +08:00
|
|
|
|
2014-03-04 11:00:26 +08:00
|
|
|
/* Don't start any governor operations if we are entering suspend */
|
|
|
|
if (cpufreq_suspended)
|
|
|
|
return 0;
|
2014-12-18 14:28:19 +08:00
|
|
|
/*
|
|
|
|
* Governor might not be initiated here if ACPI _PPC changed
|
|
|
|
* notification happened, so check it.
|
|
|
|
*/
|
|
|
|
if (!policy->governor)
|
|
|
|
return -EINVAL;
|
2014-03-04 11:00:26 +08:00
|
|
|
|
2017-07-19 18:12:46 +08:00
|
|
|
/* Platform doesn't want dynamic frequency switching ? */
|
2020-11-11 01:25:57 +08:00
|
|
|
if (policy->governor->flags & CPUFREQ_GOV_DYNAMIC_SWITCHING &&
|
2017-07-19 18:12:49 +08:00
|
|
|
cpufreq_driver->flags & CPUFREQ_NO_AUTO_DYNAMIC_SWITCHING) {
|
2016-02-05 09:37:42 +08:00
|
|
|
struct cpufreq_governor *gov = cpufreq_fallback_governor();
|
|
|
|
|
|
|
|
if (gov) {
|
2017-07-19 18:12:48 +08:00
|
|
|
pr_warn("Can't use %s governor as dynamic switching is disallowed. Fallback to %s governor\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
policy->governor->name, gov->name);
|
2007-10-03 04:28:13 +08:00
|
|
|
policy->governor = gov;
|
2016-02-05 09:37:42 +08:00
|
|
|
} else {
|
|
|
|
return -EINVAL;
|
2007-10-03 04:28:13 +08:00
|
|
|
}
|
2007-10-03 04:28:12 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2016-05-14 07:01:46 +08:00
|
|
|
if (!try_module_get(policy->governor->owner))
|
|
|
|
return -EINVAL;
|
cpufreq: Fix governor start/stop race condition
Cpufreq governors' stop and start operations should be carried out
in sequence. Otherwise, there will be unexpected behavior, like in
the example below.
Suppose there are 4 CPUs and policy->cpu=CPU0, CPU1/2/3 are linked
to CPU0. The normal sequence is:
1) Current governor is userspace. An application tries to set the
governor to ondemand. It will call __cpufreq_set_policy() in
which it will stop the userspace governor and then start the
ondemand governor.
2) Current governor is userspace. The online of CPU3 runs on CPU0.
It will call cpufreq_add_policy_cpu() in which it will first
stop the userspace governor, and then start it again.
If the sequence of the above two cases interleaves, it becomes:
1) Application stops userspace governor
2) Hotplug stops userspace governor
which is a problem, because the governor shouldn't be stopped twice
in a row. What happens next is:
3) Application starts ondemand governor
4) Hotplug starts a governor
In step 4, the hotplug is supposed to start the userspace governor,
but now the governor has been changed by the application to ondemand,
so the ondemand governor is started once again, which is incorrect.
The solution is to prevent policy governors from being stopped
multiple times in a row. A governor should only be stopped once for
one policy. After it has been stopped, no more governor stop
operations should be executed.
Also add a mutex to serialize governor operations.
[rjw: Changelog. And you owe me a beverage of my choice.]
Signed-off-by: Xiaoguang Chen <chenxg@marvell.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2013-06-19 15:00:07 +08:00
|
|
|
|
2016-05-14 07:01:46 +08:00
|
|
|
pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2016-06-03 05:24:15 +08:00
|
|
|
if (policy->governor->init) {
|
|
|
|
ret = policy->governor->init(policy);
|
|
|
|
if (ret) {
|
2016-05-12 21:13:35 +08:00
|
|
|
module_put(policy->governor->owner);
|
2016-06-03 05:24:15 +08:00
|
|
|
return ret;
|
|
|
|
}
|
2016-05-12 21:13:35 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2020-11-11 01:26:37 +08:00
|
|
|
policy->strict_target = !!(policy->governor->flags & CPUFREQ_GOV_STRICT_TARGET);
|
|
|
|
|
2016-05-14 07:01:46 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void cpufreq_exit_governor(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
if (cpufreq_suspended || !policy->governor)
|
|
|
|
return;
|
|
|
|
|
|
|
|
pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
|
|
|
|
|
2016-06-03 05:24:15 +08:00
|
|
|
if (policy->governor->exit)
|
|
|
|
policy->governor->exit(policy);
|
2016-05-14 07:01:46 +08:00
|
|
|
|
|
|
|
module_put(policy->governor->owner);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
cpufreq: intel_pstate: Implement passive mode with HWP enabled
Allow intel_pstate to work in the passive mode with HWP enabled and
make it set the HWP minimum performance limit (HWP floor) to the
P-state value given by the target frequency supplied by the cpufreq
governor, so as to prevent the HWP algorithm and the CPU scheduler
from working against each other, at least when the schedutil governor
is in use, and update the intel_pstate documentation accordingly.
Among other things, this allows utilization clamps to be taken
into account, at least to a certain extent, when intel_pstate is
in use and makes it more likely that sufficient capacity for
deadline tasks will be provided.
After this change, the resulting behavior of an HWP system with
intel_pstate in the passive mode should be close to the behavior
of the analogous non-HWP system with intel_pstate in the passive
mode, except that the HWP algorithm is generally allowed to make the
CPU run at a frequency above the floor P-state set by intel_pstate in
the entire available range of P-states, while without HWP a CPU can
run in a P-state above the requested one if the latter falls into the
range of turbo P-states (referred to as the turbo range) or if the
P-states of all CPUs in one package are coordinated with each other
at the hardware level.
[Note that in principle the HWP floor may not be taken into account
by the processor if it falls into the turbo range, in which case the
processor has a license to choose any P-state, either below or above
the HWP floor, just like a non-HWP processor in the case when the
target P-state falls into the turbo range.]
With this change applied, intel_pstate in the passive mode assumes
complete control over the HWP request MSR and concurrent changes of
that MSR (eg. via the direct MSR access interface) are overridden by
it.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
2020-08-06 20:03:55 +08:00
|
|
|
int cpufreq_start_governor(struct cpufreq_policy *policy)
|
2016-03-21 22:45:24 +08:00
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2016-05-14 07:01:46 +08:00
|
|
|
if (cpufreq_suspended)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (!policy->governor)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
|
|
|
|
|
2019-06-20 11:05:46 +08:00
|
|
|
if (cpufreq_driver->get)
|
2019-06-20 11:05:49 +08:00
|
|
|
cpufreq_verify_current_freq(policy, false);
|
2016-03-21 22:47:48 +08:00
|
|
|
|
2016-06-03 05:24:15 +08:00
|
|
|
if (policy->governor->start) {
|
|
|
|
ret = policy->governor->start(policy);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (policy->governor->limits)
|
|
|
|
policy->governor->limits(policy);
|
2016-05-14 06:59:27 +08:00
|
|
|
|
|
|
|
return 0;
|
2016-03-21 22:45:24 +08:00
|
|
|
}
|
|
|
|
|
cpufreq: intel_pstate: Implement passive mode with HWP enabled
Allow intel_pstate to work in the passive mode with HWP enabled and
make it set the HWP minimum performance limit (HWP floor) to the
P-state value given by the target frequency supplied by the cpufreq
governor, so as to prevent the HWP algorithm and the CPU scheduler
from working against each other, at least when the schedutil governor
is in use, and update the intel_pstate documentation accordingly.
Among other things, this allows utilization clamps to be taken
into account, at least to a certain extent, when intel_pstate is
in use and makes it more likely that sufficient capacity for
deadline tasks will be provided.
After this change, the resulting behavior of an HWP system with
intel_pstate in the passive mode should be close to the behavior
of the analogous non-HWP system with intel_pstate in the passive
mode, except that the HWP algorithm is generally allowed to make the
CPU run at a frequency above the floor P-state set by intel_pstate in
the entire available range of P-states, while without HWP a CPU can
run in a P-state above the requested one if the latter falls into the
range of turbo P-states (referred to as the turbo range) or if the
P-states of all CPUs in one package are coordinated with each other
at the hardware level.
[Note that in principle the HWP floor may not be taken into account
by the processor if it falls into the turbo range, in which case the
processor has a license to choose any P-state, either below or above
the HWP floor, just like a non-HWP processor in the case when the
target P-state falls into the turbo range.]
With this change applied, intel_pstate in the passive mode assumes
complete control over the HWP request MSR and concurrent changes of
that MSR (eg. via the direct MSR access interface) are overridden by
it.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
2020-08-06 20:03:55 +08:00
|
|
|
void cpufreq_stop_governor(struct cpufreq_policy *policy)
|
2016-05-14 07:01:46 +08:00
|
|
|
{
|
|
|
|
if (cpufreq_suspended || !policy->governor)
|
|
|
|
return;
|
|
|
|
|
|
|
|
pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
|
|
|
|
|
2016-06-03 05:24:15 +08:00
|
|
|
if (policy->governor->stop)
|
|
|
|
policy->governor->stop(policy);
|
2016-05-14 07:01:46 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void cpufreq_governor_limits(struct cpufreq_policy *policy)
|
|
|
|
{
|
|
|
|
if (cpufreq_suspended || !policy->governor)
|
|
|
|
return;
|
|
|
|
|
|
|
|
pr_debug("%s: for CPU %u\n", __func__, policy->cpu);
|
|
|
|
|
2016-06-03 05:24:15 +08:00
|
|
|
if (policy->governor->limits)
|
|
|
|
policy->governor->limits(policy);
|
2016-03-21 22:45:24 +08:00
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
int cpufreq_register_governor(struct cpufreq_governor *governor)
|
|
|
|
{
|
2006-07-07 03:30:26 +08:00
|
|
|
int err;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
if (!governor)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -ENODEV;
|
|
|
|
|
2006-01-14 07:54:22 +08:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2006-02-28 13:43:23 +08:00
|
|
|
|
2006-07-07 03:30:26 +08:00
|
|
|
err = -EBUSY;
|
2015-01-02 15:04:26 +08:00
|
|
|
if (!find_governor(governor->name)) {
|
2006-07-07 03:30:26 +08:00
|
|
|
err = 0;
|
|
|
|
list_add(&governor->governor_list, &cpufreq_governor_list);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2006-02-28 13:43:23 +08:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2006-07-07 03:30:26 +08:00
|
|
|
return err;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_register_governor);
|
|
|
|
|
|
|
|
void cpufreq_unregister_governor(struct cpufreq_governor *governor)
|
|
|
|
{
|
2015-05-12 14:52:34 +08:00
|
|
|
struct cpufreq_policy *policy;
|
|
|
|
unsigned long flags;
|
2009-11-12 22:18:46 +08:00
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!governor)
|
|
|
|
return;
|
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return;
|
|
|
|
|
2015-05-12 14:52:34 +08:00
|
|
|
/* clear last_governor for all inactive policies */
|
|
|
|
read_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
for_each_inactive_policy(policy) {
|
2015-05-12 14:52:51 +08:00
|
|
|
if (!strcmp(policy->last_governor, governor->name)) {
|
|
|
|
policy->governor = NULL;
|
2015-05-12 14:52:34 +08:00
|
|
|
strcpy(policy->last_governor, "\0");
|
2015-05-12 14:52:51 +08:00
|
|
|
}
|
2009-11-12 22:18:46 +08:00
|
|
|
}
|
2015-05-12 14:52:34 +08:00
|
|
|
read_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2009-11-12 22:18:46 +08:00
|
|
|
|
2006-01-14 07:54:22 +08:00
|
|
|
mutex_lock(&cpufreq_governor_mutex);
|
2005-04-17 06:20:36 +08:00
|
|
|
list_del(&governor->governor_list);
|
2006-01-14 07:54:22 +08:00
|
|
|
mutex_unlock(&cpufreq_governor_mutex);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_unregister_governor);
|
|
|
|
|
|
|
|
|
|
|
|
/*********************************************************************
|
|
|
|
* POLICY INTERFACE *
|
|
|
|
*********************************************************************/
|
|
|
|
|
|
|
|
/**
|
|
|
|
* cpufreq_get_policy - get the current cpufreq_policy
|
2009-01-18 14:37:11 +08:00
|
|
|
* @policy: struct cpufreq_policy into which the current cpufreq_policy
|
|
|
|
* is written
|
2020-07-15 16:26:23 +08:00
|
|
|
* @cpu: CPU to find the policy for
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
* Reads the current cpufreq policy.
|
|
|
|
*/
|
|
|
|
int cpufreq_get_policy(struct cpufreq_policy *policy, unsigned int cpu)
|
|
|
|
{
|
|
|
|
struct cpufreq_policy *cpu_policy;
|
|
|
|
if (!policy)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
cpu_policy = cpufreq_cpu_get(cpu);
|
|
|
|
if (!cpu_policy)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2013-08-07 01:23:06 +08:00
|
|
|
memcpy(policy, cpu_policy, sizeof(*policy));
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
cpufreq_cpu_put(cpu_policy);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_get_policy);
|
|
|
|
|
2019-02-20 07:22:51 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_set_policy - Modify cpufreq policy parameters.
|
|
|
|
* @policy: Policy object to modify.
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
* @new_gov: Policy governor pointer.
|
|
|
|
* @new_pol: Policy value (for drivers with built-in governors).
|
2019-02-20 07:22:51 +08:00
|
|
|
*
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
* Invoke the cpufreq driver's ->verify() callback to sanity-check the frequency
|
|
|
|
* limits to be set for the policy, update @policy with the verified limits
|
|
|
|
* values and either invoke the driver's ->setpolicy() callback (if present) or
|
|
|
|
* carry out a governor update for @policy. That is, run the current governor's
|
|
|
|
* ->limits() callback (if @new_gov points to the same object as the one in
|
|
|
|
* @policy) or replace the governor for @policy with @new_gov.
|
2019-02-20 07:22:51 +08:00
|
|
|
*
|
|
|
|
* The cpuinfo part of @policy is not updated by this function.
|
2006-07-26 21:40:07 +08:00
|
|
|
*/
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
static int cpufreq_set_policy(struct cpufreq_policy *policy,
|
|
|
|
struct cpufreq_governor *new_gov,
|
|
|
|
unsigned int new_pol)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
struct cpufreq_policy_data new_data;
|
2014-02-18 05:56:35 +08:00
|
|
|
struct cpufreq_governor *old_gov;
|
|
|
|
int ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
memcpy(&new_data.cpuinfo, &policy->cpuinfo, sizeof(policy->cpuinfo));
|
|
|
|
new_data.freq_table = policy->freq_table;
|
|
|
|
new_data.cpu = policy->cpu;
|
2015-07-30 18:10:40 +08:00
|
|
|
/*
|
2019-07-08 18:57:52 +08:00
|
|
|
* PM QoS framework collects all the requests from users and provide us
|
|
|
|
* the final aggregated value here.
|
|
|
|
*/
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
new_data.min = freq_qos_read_value(&policy->constraints, FREQ_QOS_MIN);
|
|
|
|
new_data.max = freq_qos_read_value(&policy->constraints, FREQ_QOS_MAX);
|
|
|
|
|
|
|
|
pr_debug("setting new policy for CPU %u: %u - %u kHz\n",
|
|
|
|
new_data.cpu, new_data.min, new_data.max);
|
2006-07-06 05:12:20 +08:00
|
|
|
|
2019-10-22 18:17:57 +08:00
|
|
|
/*
|
|
|
|
* Verify that the CPU speed can be set within these limits and make sure
|
|
|
|
* that min <= max.
|
|
|
|
*/
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
ret = cpufreq_driver->verify(&new_data);
|
2005-04-17 06:20:36 +08:00
|
|
|
if (ret)
|
2014-02-18 05:56:35 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
cpufreq: Make policy min/max hard requirements
When applying the policy min/max limits, the requested frequency is
simply clamped to not be out of range. It means, however, if one of the
boundaries isn't an available frequency, the frequency resolution can
return a value out of those limits, depending on the relation used.
e.g. freq{0,1,2} being available frequencies.
freq0 policy->min freq1 policy->max freq2
| | | | |
17kHz 18kHz 19kHz 20kHz 21kHz
__resolve_freq(21kHz, CPUFREQ_RELATION_L) -> 21kHz (out of bounds)
__resolve_freq(17kHz, CPUFREQ_RELATION_H) -> 17kHz (out of bounds)
If, during the policy init, we resolve the requested min/max to existing
frequencies, we ensure that any CPUFREQ_RELATION_* would resolve to a
frequency which is inside the policy min/max range.
Making the policy limits rigid helps to introduce the inefficient
frequencies support. Resolving an inefficient frequency to an efficient
one should not transgress policy->max (which can be set for thermal
reason) and having a value we can trust simplify this comparison.
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2021-09-08 22:05:26 +08:00
|
|
|
/*
|
|
|
|
* Resolve policy min/max to available frequencies. It ensures
|
|
|
|
* no frequency resolution will neither overshoot the requested maximum
|
|
|
|
* nor undershoot the requested minimum.
|
|
|
|
*/
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
policy->min = new_data.min;
|
|
|
|
policy->max = new_data.max;
|
cpufreq: Make policy min/max hard requirements
When applying the policy min/max limits, the requested frequency is
simply clamped to not be out of range. It means, however, if one of the
boundaries isn't an available frequency, the frequency resolution can
return a value out of those limits, depending on the relation used.
e.g. freq{0,1,2} being available frequencies.
freq0 policy->min freq1 policy->max freq2
| | | | |
17kHz 18kHz 19kHz 20kHz 21kHz
__resolve_freq(21kHz, CPUFREQ_RELATION_L) -> 21kHz (out of bounds)
__resolve_freq(17kHz, CPUFREQ_RELATION_H) -> 17kHz (out of bounds)
If, during the policy init, we resolve the requested min/max to existing
frequencies, we ensure that any CPUFREQ_RELATION_* would resolve to a
frequency which is inside the policy min/max range.
Making the policy limits rigid helps to introduce the inefficient
frequencies support. Resolving an inefficient frequency to an efficient
one should not transgress policy->max (which can be set for thermal
reason) and having a value we can trust simplify this comparison.
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2021-09-08 22:05:26 +08:00
|
|
|
policy->min = __resolve_freq(policy, policy->min, CPUFREQ_RELATION_L);
|
|
|
|
policy->max = __resolve_freq(policy, policy->max, CPUFREQ_RELATION_H);
|
2018-07-25 01:35:44 +08:00
|
|
|
trace_cpu_frequency_limits(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2016-07-14 04:25:25 +08:00
|
|
|
policy->cached_target_freq = UINT_MAX;
|
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("new min and max freqs are %u - %u kHz\n",
|
2014-03-12 01:03:00 +08:00
|
|
|
policy->min, policy->max);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver->setpolicy) {
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
policy->policy = new_pol;
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("setting range\n");
|
2019-02-20 07:26:30 +08:00
|
|
|
return cpufreq_driver->setpolicy(policy);
|
2014-02-18 05:56:35 +08:00
|
|
|
}
|
2005-04-17 06:20:36 +08:00
|
|
|
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
if (new_gov == policy->governor) {
|
2019-02-20 07:25:18 +08:00
|
|
|
pr_debug("governor limits update\n");
|
2016-05-14 07:01:46 +08:00
|
|
|
cpufreq_governor_limits(policy);
|
2016-05-14 06:59:27 +08:00
|
|
|
return 0;
|
2016-03-21 22:45:24 +08:00
|
|
|
}
|
2013-03-27 23:58:57 +08:00
|
|
|
|
2014-02-18 05:56:35 +08:00
|
|
|
pr_debug("governor switch\n");
|
|
|
|
|
|
|
|
/* save old, working values */
|
|
|
|
old_gov = policy->governor;
|
|
|
|
/* end old governor */
|
|
|
|
if (old_gov) {
|
2016-05-12 21:14:12 +08:00
|
|
|
cpufreq_stop_governor(policy);
|
2016-05-12 21:13:35 +08:00
|
|
|
cpufreq_exit_governor(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2014-02-18 05:56:35 +08:00
|
|
|
/* start new governor */
|
cpufreq: Avoid creating excessively large stack frames
In the process of modifying a cpufreq policy, the cpufreq core makes
a copy of it including all of the internals which is stored on the
CPU stack. Because struct cpufreq_policy is relatively large, this
may cause the size of the stack frame to exceed the 2 KB limit and
so the GCC complains when -Wframe-larger-than= is used.
In fact, it is not necessary to copy the entire policy structure
in order to modify it, however.
First, because cpufreq_set_policy() obtains the min and max policy
limits from frequency QoS now, it is not necessary to pass the limits
to it from the callers. The only things that need to be passed to it
from there are the new governor pointer or (if there is a built-in
governor in the driver) the "policy" value representing the governor
choice. They both can be passed as individual arguments, though, so
make cpufreq_set_policy() take them this way and rework its callers
accordingly. This avoids making copies of cpufreq policies in the
callers of cpufreq_set_policy().
Second, cpufreq_set_policy() still needs to pass the new policy
data to the ->verify() callback of the cpufreq driver whose task
is to sanitize the min and max policy limits. It still does not
need to make a full copy of struct cpufreq_policy for this purpose,
but it needs to pass a few items from it to the driver in case they
are needed (different drivers have different needs in that respect
and all of them have to be covered). For this reason, introduce
struct cpufreq_policy_data to hold copies of the members of
struct cpufreq_policy used by the existing ->verify() driver
callbacks and pass a pointer to a temporary structure of that
type to ->verify() (instead of passing a pointer to full struct
cpufreq_policy to it).
While at it, notice that intel_pstate and longrun don't really need
to verify the "policy" value in struct cpufreq_policy, so drop those
check from them to avoid copying "policy" into struct
cpufreq_policy_data (which allows it to be slightly smaller).
Also while at it fix up white space in a couple of places and make
cpufreq_set_policy() static (as it can be so).
Fixes: 3000ce3c52f8 ("cpufreq: Use per-policy frequency QoS")
Link: https://lore.kernel.org/linux-pm/CAMuHMdX6-jb1W8uC2_237m8ctCpsnGp=JCxqt8pCWVqNXHmkVg@mail.gmail.com
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: 5.4+ <stable@vger.kernel.org> # 5.4+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
2020-01-27 06:40:11 +08:00
|
|
|
policy->governor = new_gov;
|
2016-05-14 07:01:46 +08:00
|
|
|
ret = cpufreq_init_governor(policy);
|
2015-07-18 14:01:03 +08:00
|
|
|
if (!ret) {
|
2016-03-21 22:45:24 +08:00
|
|
|
ret = cpufreq_start_governor(policy);
|
|
|
|
if (!ret) {
|
2019-02-20 07:25:18 +08:00
|
|
|
pr_debug("governor change\n");
|
2018-12-03 17:56:21 +08:00
|
|
|
sched_cpufreq_governor_change(policy, old_gov);
|
2016-03-21 22:45:24 +08:00
|
|
|
return 0;
|
|
|
|
}
|
2016-03-30 09:47:49 +08:00
|
|
|
cpufreq_exit_governor(policy);
|
2014-02-18 05:56:35 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* new governor failed, so re-start old one */
|
|
|
|
pr_debug("starting governor %s failed\n", policy->governor->name);
|
|
|
|
if (old_gov) {
|
|
|
|
policy->governor = old_gov;
|
2016-05-14 07:01:46 +08:00
|
|
|
if (cpufreq_init_governor(policy))
|
2015-07-18 14:01:03 +08:00
|
|
|
policy->governor = NULL;
|
|
|
|
else
|
2016-03-21 22:45:24 +08:00
|
|
|
cpufreq_start_governor(policy);
|
2014-02-18 05:56:35 +08:00
|
|
|
}
|
|
|
|
|
2015-07-18 14:01:03 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
2019-02-20 07:22:51 +08:00
|
|
|
* cpufreq_update_policy - Re-evaluate an existing cpufreq policy.
|
|
|
|
* @cpu: CPU to re-evaluate the policy for.
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
2019-02-20 07:22:51 +08:00
|
|
|
* Update the current frequency for the cpufreq policy of @cpu and use
|
2019-07-05 18:51:24 +08:00
|
|
|
* cpufreq_set_policy() to re-apply the min and max limits, which triggers the
|
|
|
|
* evaluation of policy notifiers and the cpufreq driver's ->verify() callback
|
|
|
|
* for the policy in question, among other things.
|
2005-04-17 06:20:36 +08:00
|
|
|
*/
|
2016-11-18 20:59:21 +08:00
|
|
|
void cpufreq_update_policy(unsigned int cpu)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
2019-03-26 19:16:58 +08:00
|
|
|
struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpu);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-06-19 02:27:32 +08:00
|
|
|
if (!policy)
|
2016-11-18 20:59:21 +08:00
|
|
|
return;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-06-19 16:49:33 +08:00
|
|
|
/*
|
|
|
|
* BIOS might change freq behind our back
|
|
|
|
* -> ask driver for current freq and notify governors about a change
|
|
|
|
*/
|
2019-06-20 11:05:48 +08:00
|
|
|
if (cpufreq_driver->get && has_target() &&
|
2019-06-20 11:05:49 +08:00
|
|
|
(cpufreq_suspended || WARN_ON(!cpufreq_verify_current_freq(policy, false))))
|
2019-02-20 07:24:25 +08:00
|
|
|
goto unlock;
|
2016-11-18 20:59:21 +08:00
|
|
|
|
2019-06-20 11:05:50 +08:00
|
|
|
refresh_frequency_limits(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2014-06-19 02:27:32 +08:00
|
|
|
unlock:
|
2019-03-26 19:16:58 +08:00
|
|
|
cpufreq_cpu_release(policy);
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL(cpufreq_update_policy);
|
|
|
|
|
2019-03-26 19:15:13 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_update_limits - Update policy limits for a given CPU.
|
|
|
|
* @cpu: CPU to update the policy limits for.
|
|
|
|
*
|
|
|
|
* Invoke the driver's ->update_limits callback if present or call
|
|
|
|
* cpufreq_update_policy() for @cpu.
|
|
|
|
*/
|
|
|
|
void cpufreq_update_limits(unsigned int cpu)
|
|
|
|
{
|
|
|
|
if (cpufreq_driver->update_limits)
|
|
|
|
cpufreq_driver->update_limits(cpu);
|
|
|
|
else
|
|
|
|
cpufreq_update_policy(cpu);
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_update_limits);
|
|
|
|
|
2013-12-20 22:24:49 +08:00
|
|
|
/*********************************************************************
|
|
|
|
* BOOST *
|
|
|
|
*********************************************************************/
|
2020-05-30 10:08:30 +08:00
|
|
|
static int cpufreq_boost_set_sw(struct cpufreq_policy *policy, int state)
|
2013-12-20 22:24:49 +08:00
|
|
|
{
|
2020-05-30 10:08:30 +08:00
|
|
|
int ret;
|
2016-02-11 20:01:12 +08:00
|
|
|
|
2020-05-30 10:08:30 +08:00
|
|
|
if (!policy->freq_table)
|
|
|
|
return -ENXIO;
|
2016-06-03 13:28:47 +08:00
|
|
|
|
2020-05-30 10:08:30 +08:00
|
|
|
ret = cpufreq_frequency_table_cpuinfo(policy, policy->freq_table);
|
|
|
|
if (ret) {
|
|
|
|
pr_err("%s: Policy frequency update failed\n", __func__);
|
|
|
|
return ret;
|
2013-12-20 22:24:49 +08:00
|
|
|
}
|
|
|
|
|
2020-05-30 10:08:30 +08:00
|
|
|
ret = freq_qos_update_request(policy->max_freq_req, policy->max);
|
|
|
|
if (ret < 0)
|
|
|
|
return ret;
|
|
|
|
|
2020-05-18 18:49:45 +08:00
|
|
|
return 0;
|
2013-12-20 22:24:49 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
int cpufreq_boost_trigger_state(int state)
|
|
|
|
{
|
2020-05-30 10:08:30 +08:00
|
|
|
struct cpufreq_policy *policy;
|
2013-12-20 22:24:49 +08:00
|
|
|
unsigned long flags;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
if (cpufreq_driver->boost_enabled == state)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
cpufreq_driver->boost_enabled = state;
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
2021-08-03 22:16:11 +08:00
|
|
|
cpus_read_lock();
|
2020-05-30 10:08:30 +08:00
|
|
|
for_each_active_policy(policy) {
|
|
|
|
ret = cpufreq_driver->set_boost(policy, state);
|
|
|
|
if (ret)
|
|
|
|
goto err_reset_state;
|
2013-12-20 22:24:49 +08:00
|
|
|
}
|
2021-08-03 22:16:11 +08:00
|
|
|
cpus_read_unlock();
|
2020-05-30 10:08:30 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
err_reset_state:
|
2021-08-03 22:16:11 +08:00
|
|
|
cpus_read_unlock();
|
2020-05-30 10:08:30 +08:00
|
|
|
|
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
|
|
|
cpufreq_driver->boost_enabled = !state;
|
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
|
|
|
|
|
|
|
pr_err("%s: Cannot %s BOOST\n",
|
|
|
|
__func__, state ? "enable" : "disable");
|
2013-12-20 22:24:49 +08:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2015-12-27 07:23:48 +08:00
|
|
|
static bool cpufreq_boost_supported(void)
|
2013-12-20 22:24:49 +08:00
|
|
|
{
|
2019-04-09 10:25:36 +08:00
|
|
|
return cpufreq_driver->set_boost;
|
2013-12-20 22:24:49 +08:00
|
|
|
}
|
|
|
|
|
2015-07-29 18:53:09 +08:00
|
|
|
static int create_boost_sysfs_file(void)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
2015-10-16 00:05:23 +08:00
|
|
|
ret = sysfs_create_file(cpufreq_global_kobject, &boost.attr);
|
2015-07-29 18:53:09 +08:00
|
|
|
if (ret)
|
|
|
|
pr_err("%s: cannot register global BOOST sysfs file\n",
|
|
|
|
__func__);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void remove_boost_sysfs_file(void)
|
|
|
|
{
|
|
|
|
if (cpufreq_boost_supported())
|
2015-10-16 00:05:23 +08:00
|
|
|
sysfs_remove_file(cpufreq_global_kobject, &boost.attr);
|
2015-07-29 18:53:09 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
int cpufreq_enable_boost_support(void)
|
|
|
|
{
|
|
|
|
if (!cpufreq_driver)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (cpufreq_boost_supported())
|
|
|
|
return 0;
|
|
|
|
|
2015-12-27 07:27:38 +08:00
|
|
|
cpufreq_driver->set_boost = cpufreq_boost_set_sw;
|
2015-07-29 18:53:09 +08:00
|
|
|
|
|
|
|
/* This will get removed on driver unregister */
|
|
|
|
return create_boost_sysfs_file();
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_enable_boost_support);
|
|
|
|
|
2013-12-20 22:24:49 +08:00
|
|
|
int cpufreq_boost_enabled(void)
|
|
|
|
{
|
|
|
|
return cpufreq_driver->boost_enabled;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_boost_enabled);
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/*********************************************************************
|
|
|
|
* REGISTER / UNREGISTER CPUFREQ DRIVER *
|
|
|
|
*********************************************************************/
|
2016-09-07 01:04:48 +08:00
|
|
|
static enum cpuhp_state hp_online;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
cpufreq: Bring CPUs up even if cpufreq_online() failed
There is a report that after commit 27622b061eb4 ("cpufreq: Convert
to hotplug state machine"), the normal CPU offline/online cycle
fails on some platforms.
According to the ftrace result, this problem was triggered on
platforms using acpi-cpufreq as the default cpufreq driver,
and due to the lack of some ACPI freq method (eg. _PCT),
cpufreq_online() failed and returned a negative value, so the CPU
hotplug state machine rolled back the CPU online process. Actually,
from the user's perspective, the failure of cpufreq_online() should
not prevent that CPU from being brought up, although cpufreq might
not work on that CPU.
BTW, during system startup cpufreq_online() is not invoked via CPU
online but by the cpufreq device creation process, so the APs can be
brought up even though cpufreq_online() fails in that stage.
This patch ignores the return value of cpufreq_online/offline() and
lets the cpufreq framework deal with the failure. cpufreq_online()
itself will do a proper rollback in that case and if _PCT is missing,
the ACPI cpufreq driver will print a warning if the corresponding
debug options have been enabled.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=194581
Fixes: 27622b061eb4 ("cpufreq: Convert to hotplug state machine")
Reported-and-tested-by: Tomasz Maciej Nowak <tmn505@gmail.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Cc: 4.9+ <stable@vger.kernel.org> # 4.9+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2017-04-09 13:45:16 +08:00
|
|
|
static int cpuhp_cpufreq_online(unsigned int cpu)
|
|
|
|
{
|
|
|
|
cpufreq_online(cpu);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int cpuhp_cpufreq_offline(unsigned int cpu)
|
|
|
|
{
|
|
|
|
cpufreq_offline(cpu);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
/**
|
|
|
|
* cpufreq_register_driver - register a CPU Frequency driver
|
|
|
|
* @driver_data: A struct cpufreq_driver containing the values#
|
|
|
|
* submitted by the CPU Frequency driver.
|
|
|
|
*
|
2013-06-19 16:49:33 +08:00
|
|
|
* Registers a CPU Frequency driver to this core code. This code
|
2016-02-21 11:50:01 +08:00
|
|
|
* returns zero on success, -EEXIST when another driver got here first
|
2006-02-28 13:43:23 +08:00
|
|
|
* (and isn't unregistered in the meantime).
|
2005-04-17 06:20:36 +08:00
|
|
|
*
|
|
|
|
*/
|
2007-02-27 06:55:48 +08:00
|
|
|
int cpufreq_register_driver(struct cpufreq_driver *driver_data)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
int ret;
|
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -ENODEV;
|
|
|
|
|
2019-11-14 11:36:17 +08:00
|
|
|
/*
|
|
|
|
* The cpufreq core depends heavily on the availability of device
|
|
|
|
* structure, make sure they are available before proceeding further.
|
|
|
|
*/
|
|
|
|
if (!get_cpu_device(0))
|
|
|
|
return -EPROBE_DEFER;
|
|
|
|
|
2005-04-17 06:20:36 +08:00
|
|
|
if (!driver_data || !driver_data->verify || !driver_data->init ||
|
2013-10-25 22:15:48 +08:00
|
|
|
!(driver_data->setpolicy || driver_data->target_index ||
|
2014-03-19 19:48:30 +08:00
|
|
|
driver_data->target) ||
|
|
|
|
(driver_data->setpolicy && (driver_data->target_index ||
|
2014-06-03 01:19:28 +08:00
|
|
|
driver_data->target)) ||
|
2019-02-14 18:46:21 +08:00
|
|
|
(!driver_data->get_intermediate != !driver_data->target_intermediate) ||
|
2019-02-12 19:06:04 +08:00
|
|
|
(!driver_data->online != !driver_data->offline))
|
2005-04-17 06:20:36 +08:00
|
|
|
return -EINVAL;
|
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("trying to register driver %s\n", driver_data->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-07-30 07:45:07 +08:00
|
|
|
/* Protect against concurrent CPU online/offline. */
|
2017-05-24 16:15:20 +08:00
|
|
|
cpus_read_lock();
|
2015-07-30 07:45:07 +08:00
|
|
|
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-04-29 06:08:16 +08:00
|
|
|
if (cpufreq_driver) {
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2015-07-30 07:45:07 +08:00
|
|
|
ret = -EEXIST;
|
|
|
|
goto out;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
2013-04-29 06:08:16 +08:00
|
|
|
cpufreq_driver = driver_data;
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2020-09-02 04:55:47 +08:00
|
|
|
/*
|
|
|
|
* Mark support for the scheduler's frequency invariance engine for
|
|
|
|
* drivers that implement target(), target_index() or fast_switch().
|
|
|
|
*/
|
|
|
|
if (!cpufreq_driver->setpolicy) {
|
|
|
|
static_branch_enable_cpuslocked(&cpufreq_freq_invariance);
|
|
|
|
pr_debug("supports frequency invariance");
|
|
|
|
}
|
|
|
|
|
2015-01-02 15:04:30 +08:00
|
|
|
if (driver_data->setpolicy)
|
|
|
|
driver_data->flags |= CPUFREQ_CONST_LOOPS;
|
|
|
|
|
2015-12-27 07:27:38 +08:00
|
|
|
if (cpufreq_boost_supported()) {
|
|
|
|
ret = create_boost_sysfs_file();
|
|
|
|
if (ret)
|
|
|
|
goto err_null_driver;
|
|
|
|
}
|
2013-12-20 22:24:49 +08:00
|
|
|
|
2011-12-22 06:29:42 +08:00
|
|
|
ret = subsys_interface_register(&cpufreq_interface);
|
2011-03-02 00:41:10 +08:00
|
|
|
if (ret)
|
2013-12-20 22:24:49 +08:00
|
|
|
goto err_boost_unreg;
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2021-02-02 12:55:11 +08:00
|
|
|
if (unlikely(list_empty(&cpufreq_policy_list))) {
|
2005-04-17 06:20:36 +08:00
|
|
|
/* if all ->init() calls failed, unregister */
|
2017-05-26 23:37:31 +08:00
|
|
|
ret = -ENODEV;
|
2015-01-02 15:04:35 +08:00
|
|
|
pr_debug("%s: No CPU initialized for driver %s\n", __func__,
|
|
|
|
driver_data->name);
|
|
|
|
goto err_if_unreg;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
|
2017-05-24 16:15:20 +08:00
|
|
|
ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN,
|
|
|
|
"cpufreq:online",
|
|
|
|
cpuhp_cpufreq_online,
|
|
|
|
cpuhp_cpufreq_offline);
|
2016-09-07 01:04:48 +08:00
|
|
|
if (ret < 0)
|
|
|
|
goto err_if_unreg;
|
|
|
|
hp_online = ret;
|
2016-09-20 22:56:28 +08:00
|
|
|
ret = 0;
|
2016-09-07 01:04:48 +08:00
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("driver %s up and running\n", driver_data->name);
|
2016-05-16 19:07:19 +08:00
|
|
|
goto out;
|
2015-07-30 07:45:07 +08:00
|
|
|
|
2011-12-22 06:29:42 +08:00
|
|
|
err_if_unreg:
|
|
|
|
subsys_interface_unregister(&cpufreq_interface);
|
2013-12-20 22:24:49 +08:00
|
|
|
err_boost_unreg:
|
2015-07-29 18:53:09 +08:00
|
|
|
remove_boost_sysfs_file();
|
2011-03-02 00:41:10 +08:00
|
|
|
err_null_driver:
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-04-29 06:08:16 +08:00
|
|
|
cpufreq_driver = NULL;
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2016-05-16 19:07:19 +08:00
|
|
|
out:
|
2017-05-24 16:15:20 +08:00
|
|
|
cpus_read_unlock();
|
2016-05-16 19:07:19 +08:00
|
|
|
return ret;
|
2005-04-17 06:20:36 +08:00
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_register_driver);
|
|
|
|
|
2020-07-15 16:26:23 +08:00
|
|
|
/*
|
2005-04-17 06:20:36 +08:00
|
|
|
* cpufreq_unregister_driver - unregister the current CPUFreq driver
|
|
|
|
*
|
2013-06-19 16:49:33 +08:00
|
|
|
* Unregister the current CPUFreq driver. Only call this if you have
|
2005-04-17 06:20:36 +08:00
|
|
|
* the right to do so, i.e. if you have succeeded in initialising before!
|
|
|
|
* Returns zero if successful, and -EINVAL if the cpufreq_driver is
|
|
|
|
* currently not initialised.
|
|
|
|
*/
|
2007-02-27 06:55:48 +08:00
|
|
|
int cpufreq_unregister_driver(struct cpufreq_driver *driver)
|
2005-04-17 06:20:36 +08:00
|
|
|
{
|
|
|
|
unsigned long flags;
|
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
if (!cpufreq_driver || (driver != cpufreq_driver))
|
2005-04-17 06:20:36 +08:00
|
|
|
return -EINVAL;
|
|
|
|
|
2011-03-27 21:04:46 +08:00
|
|
|
pr_debug("unregistering driver %s\n", driver->name);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2015-07-22 23:59:11 +08:00
|
|
|
/* Protect against concurrent cpu hotplug */
|
2017-05-24 16:15:20 +08:00
|
|
|
cpus_read_lock();
|
2011-12-22 06:29:42 +08:00
|
|
|
subsys_interface_unregister(&cpufreq_interface);
|
2015-07-29 18:53:09 +08:00
|
|
|
remove_boost_sysfs_file();
|
2020-09-02 04:55:47 +08:00
|
|
|
static_branch_disable_cpuslocked(&cpufreq_freq_invariance);
|
2017-05-24 16:15:20 +08:00
|
|
|
cpuhp_remove_state_nocalls_cpuslocked(hp_online);
|
2005-04-17 06:20:36 +08:00
|
|
|
|
2013-02-23 00:24:34 +08:00
|
|
|
write_lock_irqsave(&cpufreq_driver_lock, flags);
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2013-04-29 06:08:16 +08:00
|
|
|
cpufreq_driver = NULL;
|
2013-08-07 01:23:11 +08:00
|
|
|
|
2013-02-23 00:24:34 +08:00
|
|
|
write_unlock_irqrestore(&cpufreq_driver_lock, flags);
|
2017-05-24 16:15:20 +08:00
|
|
|
cpus_read_unlock();
|
2005-04-17 06:20:36 +08:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
EXPORT_SYMBOL_GPL(cpufreq_unregister_driver);
|
2007-02-06 08:12:44 +08:00
|
|
|
|
|
|
|
static int __init cpufreq_core_init(void)
|
|
|
|
{
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
struct cpufreq_governor *gov = cpufreq_default_governor();
|
|
|
|
|
2012-03-14 07:18:39 +08:00
|
|
|
if (cpufreq_disabled())
|
|
|
|
return -ENODEV;
|
|
|
|
|
2015-10-16 00:05:22 +08:00
|
|
|
cpufreq_global_kobject = kobject_create_and_add("cpufreq", &cpu_subsys.dev_root->kobj);
|
2009-07-24 21:25:05 +08:00
|
|
|
BUG_ON(!cpufreq_global_kobject);
|
|
|
|
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
if (!strlen(default_governor))
|
|
|
|
strncpy(default_governor, gov->name, CPUFREQ_NAME_LEN);
|
|
|
|
|
2007-02-06 08:12:44 +08:00
|
|
|
return 0;
|
|
|
|
}
|
2017-03-01 05:44:16 +08:00
|
|
|
module_param(off, int, 0444);
|
cpufreq: Specify default governor on command line
Currently, the only way to specify the default CPUfreq governor is
via Kconfig options, which suits users who can build the kernel
themselves perfectly.
However, for those who use a distro-like kernel (such as Android,
with the Generic Kernel Image project), the only way to use a
non-default governor is to boot to userspace, and to then switch
using the sysfs interface. Being able to specify the default governor
on the command line, like is the case for cpuidle, would allow those
users to specify their governor of choice earlier on, and to simplify
the userspace boot procedure slighlty.
To support this use-case, add a kernel command line parameter
allowing the default governor for CPUfreq to be specified, which
takes precedence over the built-in default.
This implementation has one notable limitation: the default governor
must be registered before the driver. This is solved for builtin
governors and drivers using appropriate *_initcall() functions. And
in the modular case, this must be reflected as a constraint on the
module loading order.
Signed-off-by: Quentin Perret <qperret@google.com>
[ Viresh: Converted 'default_governor' to a string and parsing it only
at initcall level, and several updates to
cpufreq_init_policy(). ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
2020-06-29 16:25:00 +08:00
|
|
|
module_param_string(default_governor, default_governor, CPUFREQ_NAME_LEN, 0444);
|
2007-02-06 08:12:44 +08:00
|
|
|
core_initcall(cpufreq_core_init);
|