__get_cpu_var() is used for multiple purposes in the kernel source. One of
them is address calculation via the form &__get_cpu_var(x). This calculates
the address for the instance of the percpu variable of the current processor
based on an offset.
Other use cases are for storing and retrieving data from the current
processors percpu area. __get_cpu_var() can be used as an lvalue when
writing data or on the right side of an assignment.
__get_cpu_var() is defined as :
#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))
__get_cpu_var() always only does an address determination. However, store
and retrieve operations could use a segment prefix (or global register on
other platforms) to avoid the address calculation.
this_cpu_write() and this_cpu_read() can directly take an offset into a
percpu area and use optimized assembly code to read and write per cpu
variables.
This patch converts __get_cpu_var into either an explicit address
calculation using this_cpu_ptr() or into a use of this_cpu operations that
use the offset. Thereby address calculations are avoided and less registers
are used when code is generated.
Transformations done to __get_cpu_var()
1. Determine the address of the percpu instance of the current processor.
DEFINE_PER_CPU(int, y);
int *x = &__get_cpu_var(y);
Converts to
int *x = this_cpu_ptr(&y);
2. Same as #1 but this time an array structure is involved.
DEFINE_PER_CPU(int, y[20]);
int *x = __get_cpu_var(y);
Converts to
int *x = this_cpu_ptr(y);
3. Retrieve the content of the current processors instance of a per cpu
variable.
DEFINE_PER_CPU(int, y);
int x = __get_cpu_var(y)
Converts to
int x = __this_cpu_read(y);
4. Retrieve the content of a percpu struct
DEFINE_PER_CPU(struct mystruct, y);
struct mystruct x = __get_cpu_var(y);
Converts to
memcpy(&x, this_cpu_ptr(&y), sizeof(x));
5. Assignment to a per cpu variable
DEFINE_PER_CPU(int, y)
__get_cpu_var(y) = x;
Converts to
__this_cpu_write(y, x);
6. Increment/Decrement etc of a per cpu variable
DEFINE_PER_CPU(int, y);
__get_cpu_var(y)++
Converts to
__this_cpu_inc(y)
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: x86@kernel.org
Acked-by: H. Peter Anvin <hpa@linux.intel.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Subsystems that want to register CPU hotplug callbacks, as well as perform
initialization for the CPUs that are already online, often do it as shown
below:
get_online_cpus();
for_each_online_cpu(cpu)
init_cpu(cpu);
register_cpu_notifier(&foobar_cpu_notifier);
put_online_cpus();
This is wrong, since it is prone to ABBA deadlocks involving the
cpu_add_remove_lock and the cpu_hotplug.lock (when running concurrently
with CPU hotplug operations).
Instead, the correct and race-free way of performing the callback
registration is:
cpu_notifier_register_begin();
for_each_online_cpu(cpu)
init_cpu(cpu);
/* Note the use of the double underscored version of the API */
__register_cpu_notifier(&foobar_cpu_notifier);
cpu_notifier_register_done();
Fix the oprofile code in x86 by using this latter form of callback
registration. But retain the calls to get/put_online_cpus(), since they are
used in other places as well, to protect the variables 'nmi_enabled' and
'ctr_running'. Strictly speaking, this is not necessary since
cpu_notifier_register_begin/done() provide a stronger synchronization
with CPU hotplug than get/put_online_cpus(). However, let's retain the
calls to get/put_online_cpus() to be consistent with the other call-sites.
By nesting get/put_online_cpus() *inside* cpu_notifier_register_begin/done(),
we avoid the ABBA deadlock possibility mentioned above.
Cc: Robert Richter <rric@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
it's always equal to ->d_sb of the second argument (parent dentry),
due to either being literally that, or ->d_sb of parent's parent.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The "event" variable is a u16 so the shift will always wrap to zero
making the line a no-op.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: <stable@vger.kernel.org> v2.6.32..
Signed-off-by: Robert Richter <robert.richter@amd.com>
We need this to better test x86 NMI timer mode. Otherwise it is very
hard to setup systems with NMI timer enabled, but we have this e.g. in
virtual machine environments.
Signed-off-by: Robert Richter <robert.richter@amd.com>
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (121 commits)
perf symbols: Increase symbol KSYM_NAME_LEN size
perf hists browser: Refuse 'a' hotkey on non symbolic views
perf ui browser: Use libslang to read keys
perf tools: Fix tracing info recording
perf hists browser: Elide DSO column when it is set to just one DSO, ditto for threads
perf hists: Don't consider filtered entries when calculating column widths
perf hists: Don't decay total_period for filtered entries
perf hists browser: Honour symbol_conf.show_{nr_samples,total_period}
perf hists browser: Do not exit on tab key with single event
perf annotate browser: Don't change selection line when returning from callq
perf tools: handle endianness of feature bitmap
perf tools: Add prelink suggestion to dso update message
perf script: Fix unknown feature comment
perf hists browser: Apply the dso and thread filters when merging new batches
perf hists: Move the dso and thread filters from hist_browser
perf ui browser: Honour the xterm colors
perf top tui: Give color hints just on the percentage, like on --stdio
perf ui browser: Make the colors configurable and change the defaults
perf tui: Remove unneeded call to newtCls on startup
perf hists: Don't format the percentage on hist_entry__snprintf
...
Fix up conflicts in arch/x86/kernel/kprobes.c manually.
Ingo's tree did the insane "add volatile to const array", which just
doesn't make sense ("volatile const"?). But we could remove the const
*and* make the array volatile to make doubly sure that gcc doesn't
optimize it away..
Also fix up kernel/trace/ring_buffer.c non-data-conflicts manually: the
reader_lock has been turned into a raw lock by the core locking merge,
and there was a new user of it introduced in this perf core merge. Make
sure that new use also uses the raw accessor functions.
This patch implements IBS feature detection and initialzation. The
code is shared between perf and oprofile. If IBS is available on the
system for perf, a pmu is setup.
Signed-off-by: Robert Richter <robert.richter@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1316597423-25723-3-git-send-email-robert.richter@amd.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Just convert all the files that have an nmi handler to the new routines.
Most of it is straight forward conversion. A couple of places needed some
tweaking like kgdb which separates the debug notifier from the nmi handler
and mce removes a call to notify_die.
[Thanks to Ying for finding out the history behind that mce call
https://lkml.org/lkml/2010/5/27/114
And Boris responding that he would like to remove that call because of it
https://lkml.org/lkml/2011/9/21/163]
The things that get converted are the registeration/unregistration routines
and the nmi handler itself has its args changed along with code removal
to check which list it is on (most are on one NMI list except for kgdb
which has both an NMI routine and an NMI Unknown routine).
Signed-off-by: Don Zickus <dzickus@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Corey Minyard <minyard@acm.org>
Cc: Jason Wessel <jason.wessel@windriver.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Corey Minyard <minyard@acm.org>
Cc: Jack Steiner <steiner@sgi.com>
Link: http://lkml.kernel.org/r/1317409584-23662-4-git-send-email-dzickus@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The oprofilefs_lock can be taken in atomic context (in profiling
interrupts) and therefore cannot cannot be preempted on -rt -
annotate it.
In mainline this change documents the low level nature of
the lock - otherwise there's no functional difference. Lockdep
and Sparse checking will work as usual.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In some rare cases, nmis are generated immediately after the nmi
handler of the cpu was started. This causes the counter not to be
enabled. Before enabling the nmi handlers we need to set variable
ctr_running first and make sure its value is written to memory.
Also, the patch makes all existing barriers a memory barrier instead
of a compiler barrier only.
Reported-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: <stable@kernel.org> # .35+
Signed-off-by: Robert Richter <robert.richter@amd.com>
For some performance events it's useful to set the EDGE and INV
bits and the CMASK mask in the counter control register. The list
of predefined events Intel releases for each CPU has some events which
require these settings to get more "natural" to use higher level events.
oprofile currently doesn't allow this.
This patch adds new extra configuration fields for them, so that
they can be specified in oprofilefs.
An updated oprofile daemon can then make use of this to set them.
v2: Write back masked extra value to variable.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>
Some subsystems in the x86 tree need to carry out suspend/resume and
shutdown operations with one CPU on-line and interrupts disabled and
they define sysdev classes and sysdevs or sysdev drivers for this
purpose. This leads to unnecessarily complicated code and excessive
memory usage, so switch them to using struct syscore_ops objects for
this purpose instead.
Generally, there are three categories of subsystems that use
sysdevs for implementing PM operations: (1) subsystems whose
suspend/resume callbacks ignore their arguments entirely (the
majority), (2) subsystems whose suspend/resume callbacks use their
struct sys_device argument, but don't really need to do that,
because they can be implemented differently in an arguably simpler
way (io_apic.c), and (3) subsystems whose suspend/resume callbacks
use their struct sys_device argument, but the value of that argument
is always the same and could be ignored (microcode_core.c). In all
of these cases the subsystems in question may be readily converted to
using struct syscore_ops objects for power management and shutdown.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
* 'for-2.6.38' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/percpu: (30 commits)
gameport: use this_cpu_read instead of lookup
x86: udelay: Use this_cpu_read to avoid address calculation
x86: Use this_cpu_inc_return for nmi counter
x86: Replace uses of current_cpu_data with this_cpu ops
x86: Use this_cpu_ops to optimize code
vmstat: User per cpu atomics to avoid interrupt disable / enable
irq_work: Use per cpu atomics instead of regular atomics
cpuops: Use cmpxchg for xchg to avoid lock semantics
x86: this_cpu_cmpxchg and this_cpu_xchg operations
percpu: Generic this_cpu_cmpxchg() and this_cpu_xchg support
percpu,x86: relocate this_cpu_add_return() and friends
connector: Use this_cpu operations
xen: Use this_cpu_inc_return
taskstats: Use this_cpu_ops
random: Use this_cpu_inc_return
fs: Use this_cpu_inc_return in buffer.c
highmem: Use this_cpu_xx_return() operations
vmstat: Use this_cpu_inc_return for vm statistics
x86: Support for this_cpu_add, sub, dec, inc_return
percpu: Generic support for this_cpu_add, sub, dec, inc_return
...
Fixed up conflicts: in arch/x86/kernel/{apic/nmi.c, apic/x2apic_uv_x.c, process.c}
as per Tejun.
With priorities in place and no one really understanding the difference between
DIE_NMI and DIE_NMI_IPI, just remove DIE_NMI_IPI and convert everyone to DIE_NMI.
This also simplifies default_do_nmi() a little bit. Instead of calling the
die_notifier in both the if and else part, just pull it out and call it before
the if-statement. This has the side benefit of avoiding a call to the ioport
to see if there is an external NMI sitting around until after the (more frequent)
internal NMIs are dealt with.
Patch-Inspired-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Don Zickus <dzickus@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1294348732-15030-5-git-send-email-dzickus@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In order to consolidate the NMI die_chain events, we need to setup the priorities
for the die notifiers.
I started by defining a bunch of common priorities that can be used by the
notifier blocks. Then I modified the notifier blocks to use the newly created
priorities.
Now that the priorities are straightened out, it should be easier to remove the
event DIE_NMI_IPI.
Signed-off-by: Don Zickus <dzickus@redhat.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1294348732-15030-4-git-send-email-dzickus@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Go through x86 code and replace __get_cpu_var and get_cpu_var
instances that refer to a scalar and are not used for address
determinations.
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Now, that we only call the exit function if init succeeds with commit:
979048e oprofile: don't call arch exit code from init code on failure
we can simplify the x86 init/exit functions too. Variable using_nmi
becomes obsolete.
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch adds CPU type detection for dunnington processor (Family 6
/ Model 29) to be identified as core 2 family cpu type (wikipedia
source).
I tested oprofile on Intel(R) Xeon(R) CPU E7440 reporting itself as
model 29, and it runs without an issue.
Spec:
http://www.intel.com/Assets/en_US/PDF/specupdate/320336.pdf
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Cc: stable@kernel.org
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch adds CPU type detection for the Intel Celeron 540, which is
part of the Core 2 family according to Wikipedia; the family and ID pair
is absent from the Volume 3B table referenced in the source code
comments. I have tested this patch on an Intel Celeron 540 machine
reporting itself as Family 6 Model 22, and OProfile runs on the machine
without issue.
Spec:
http://download.intel.com/design/mobile/SPECUPDT/317667.pdf
Signed-off-by: Patrick Simmons <linuxrocks123@netscape.net>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Cc: stable@kernel.org
Signed-off-by: Robert Richter <robert.richter@amd.com>
The use of the return value of init_sysfs() with commit
10f0412 oprofile, x86: fix init_sysfs error handling
discovered the following build error for !CONFIG_PM:
.../linux/arch/x86/oprofile/nmi_int.c: In function ‘op_nmi_init’:
.../linux/arch/x86/oprofile/nmi_int.c:784: error: expected expression before ‘do’
make[2]: *** [arch/x86/oprofile/nmi_int.o] Error 1
make[1]: *** [arch/x86/oprofile] Error 2
This patch fixes this.
Reported-by: Ingo Molnar <mingo@elte.hu>
Cc: stable@kernel.org
Signed-off-by: Robert Richter <robert.richter@amd.com>
On failure init_sysfs() might not properly free resources. The error
code of the function is not checked. And, when reinitializing the exit
function might be called twice. This patch fixes all this.
Cc: stable@kernel.org
Signed-off-by: Robert Richter <robert.richter@amd.com>
Newer Intel processors identifying themselves as model 30 are not recognized by
oprofile.
<cpuinfo snippet>
model : 30
model name : Intel(R) Xeon(R) CPU X3470 @ 2.93GHz
</cpuinfo snippet>
Running oprofile on these machines gives the following:
+ opcontrol --init
+ opcontrol --list-events
oprofile: available events for CPU type "Intel Architectural Perfmon"
See Intel 64 and IA-32 Architectures Software Developer's Manual
Volume 3B (Document 253669) Chapter 18 for architectural perfmon events
This is a limited set of fallback events because oprofile doesn't know your CPU
CPU_CLK_UNHALTED: (counter: all)
Clock cycles when not halted (min count: 6000)
INST_RETIRED: (counter: all)
number of instructions retired (min count: 6000)
LLC_MISSES: (counter: all)
Last level cache demand requests from this core that missed the LLC
(min count: 6000)
Unit masks (default 0x41)
----------
0x41: No unit mask
LLC_REFS: (counter: all)
Last level cache demand requests from this core (min count: 6000)
Unit masks (default 0x4f)
----------
0x4f: No unit mask
BR_MISS_PRED_RETIRED: (counter: all)
number of mispredicted branches retired (precise) (min count: 500)
+ opcontrol --shutdown
Tested using oprofile 0.9.6.
Signed-off-by: Josh Hunt <johunt@akamai.com>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>
Back when the patch was submitted for "Add Xeon 7500 series support to
oprofile", Robert Richter had asked for a followon patch that
converted all the CPU ID values to hex.
I have done that here for the "i386/core_i7" and "i386/atom" class
processors in the ppro_init() function and also added some comments on
where to find documentation on the Intel processors.
Signed-off-by: John L. Villalovos <john.l.villalovos@intel.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>
Current IBS code is not hotplug capable. An offline cpu might not be
initialized or deinitialized properly. This patch fixes this by
removing on_each_cpu() functions. The IBS init/deinit code is executed
in the per-cpu functions model->setup_ctrs() and model->cpu_down()
which are also called by hotplug notifiers. model->cpu_down() replaces
model->exit() that became obsolete.
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch moves the cpu notifier registration from nmi_init() to
nmi_setup(). The corresponding unregistration function is now in
nmi_shutdown(). Thus, the hotplug code is only active, if the oprofile
daemon is running.
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
Reordering some functions. Necessary for the next patch. No functional
changes.
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch adds checks to the nmi handler. Now samples are only
generated and counters reenabled, if the counters are running.
Otherwise the counters are stopped, if oprofile is using the nmi. In
other cases it will ignore the nmi notification.
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch reworks oprofile cpu hotplug code as follows:
Introduce ctr_running variable to check, if counters are running or
not. The state must be known for taking a cpu on or offline and when
switching counters during counter multiplexing.
Protect on_each_cpu() sections with get_online_cpus()/put_online_cpu()
functions. This is necessary if notifiers or states are
modified. Within these sections the cpu mask may not change.
Switch only between counters in nmi_cpu_switch(), if counters are
running. Otherwise the switch may restart a counter though they are
disabled.
Add nmi_cpu_setup() and nmi_cpu_shutdown() to cpu hotplug code. The
function must also be called to avoid uninitialzed counter usage.
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
CPU notifier register functions also exist if CONFIG_SMP is
disabled. This change is part of hotplug code rework and also
necessary for later patches.
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
In case a counter is already reserved by the watchdog or perf_event
subsystem, oprofile ignored this counters silently. This case is
handled now and oprofile_setup() now reports an error.
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch improves the error handler in nmi_setup(). Most parts of
the code are moved to allocate_msrs(). In case of an error
allocate_msrs() also frees already allocated memory. nmi_setup()
becomes easier and better extendable.
Signed-off-by: Robert Richter <robert.richter@amd.com>
Multiple virtual counters share one physical counter. The reservation
of virtual counters fails due to duplicate allocation of the same
counter. The counters are already reserved. Thus, virtual counter
reservation may removed at all. This also makes the code easier.
Cc: stable@kernel.org
Signed-off-by: Robert Richter <robert.richter@amd.com>
Add Xeon 7500 series support to oprofile.
Straight forward: it's the same as Core i7, so just detect
the model number. No user space changes needed.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>
With multiplexing enabled oprofile crashs when profiling more than 28
events. This patch fixes this.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch adds a check for the availability of a counter. A virtual
counter is used only if its physical counter is not reserved.
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch moves the multiplexing switch counter from x86 code to
common oprofile statistic variables. Now the value will be available
and usable for all architectures. The initialization and
incrementation also moved to common code.
Signed-off-by: Robert Richter <robert.richter@amd.com>
To setup a counter for all cpus, its structure is cloned from cpu
0. This patch implements mux_clone() to do this part for multiplexing
data.
Signed-off-by: Robert Richter <robert.richter@amd.com>
This patch checks if the model supports multiplexing. Only then
multiplexing will be enabled. The code is added to the common x86
initialization.
Signed-off-by: Robert Richter <robert.richter@amd.com>
