If lockdep complains something for other subsystem, lockdep_is_held()
can be false negative, so we need to also test debug_locks before
triggering WARN.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
rcu_read_lock_sched() is better than preempt_disable() if the code is
protected by RCU_SCHED.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
If pwq_adjust_max_active() changes max_active from 0 to
saved_max_active, it needs to wakeup worker. This is already done by
thaw_workqueues().
If pwq_adjust_max_active() increases max_active for an unbound wq,
while not strictly necessary for correctness, it's still desirable to
wake up a worker so that the requested concurrency level is reached
sooner.
Move wake_up_worker() call from thaw_workqueues() to
pwq_adjust_max_active() so that it can handle both of the above two
cases. This also makes thaw_workqueues() simpler.
tj: Updated comments and description.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
We can test worker->recue_wq instead of reaching into
current_pwq->wq->rescuer and then comparing it to self.
tj: Commit message.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Process connector can now also detect coredumping events.
Main aim of patch is get notified at start of coredumping, instead of
having to wait for it to finish and then being notified through EXIT
event.
Could be used for instance by process-managers that want to get
notified as soon as possible about process failures, and not
necessarily beeing notified after coredump, which could be in the
order of minutes depending on size of coredump, piping and so on.
Signed-off-by: Jesper Derehag <jderehag@hotmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
get_unbound_pool() forgot to set POOL_FREEZING if workqueue_freezing
is set and a new pool could go out of sync with the global freezing
state.
Fix it by adding POOL_FREEZING if workqueue_freezing. wq_mutex is
already held so no further locking is necessary. This also removes
the unused static variable warning when !CONFIG_FREEZER.
tj: Updated commit message.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
The 3rd parameter of flex_array_prealloc() is the number of elements,
not the index of the last element.
The effect of the bug is, when opening cgroup.procs, a flex array will
be allocated and all elements of the array is allocated with
GFP_KERNEL flag, but the last one is GFP_ATOMIC, and if we fail to
allocate memory for it, it'll trigger a BUG_ON().
Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: stable@vger.kernel.org
Fix symbol versioning on architectures with symbol prefixes. Although
the build was free from warnings the actual modules still wouldn't load
as the ____versions table contained unprefixed symbol names, which were
being compared against the prefixed symbol names when checking the
symbol versions.
This is fixed by modifying modpost to add the symbol prefix to the
____versions table it outputs (Modules.symvers still contains unprefixed
symbol names). The check_modstruct_version() function is also fixed as
it checks the version of the unprefixed "module_layout" symbol which
would no longer work.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Michal Marek <mmarek@suse.cz>
Cc: Sam Ravnborg <sam@ravnborg.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jonathan Kliegman <kliegs@chromium.org>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> (use VMLINUX_SYMBOL_STR)
With the recent addition of the custom attributes support, unbound
pools may have allowed cpumask which isn't full. As long as some of
CPUs in the cpumask are online, its workers will maintain cpus_allowed
as set on worker creation; however, once no online CPU is left in
cpus_allowed, the scheduler will reset cpus_allowed of any workers
which get scheduled so that they can execute.
To remain compliant to the user-specified configuration, CPU affinity
needs to be restored when a CPU becomes online for an unbound pool
which doesn't currently have any online CPUs before.
This patch implement restore_unbound_workers_cpumask(), which is
called from CPU_ONLINE for all unbound pools, checks whether the
coming up CPU is the first allowed online one, and, if so, invokes
set_cpus_allowed_ptr() with the configured cpumask on all workers.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Rebinding workers of a per-cpu pool after a CPU comes online involves
a lot of back-and-forth mostly because only the task itself could
adjust CPU affinity if PF_THREAD_BOUND was set.
As CPU_ONLINE itself couldn't adjust affinity, it had to somehow
coerce the workers themselves to perform set_cpus_allowed_ptr(). Due
to the various states a worker can be in, this led to three different
paths a worker may be rebound. worker->rebind_work is queued to busy
workers. Idle ones are signaled by unlinking worker->entry and call
idle_worker_rebind(). The manager isn't covered by either and
implements its own mechanism.
PF_THREAD_BOUND has been relaced with PF_NO_SETAFFINITY and CPU_ONLINE
itself now can manipulate CPU affinity of workers. This patch
replaces the existing rebind mechanism with direct one where
CPU_ONLINE iterates over all workers using for_each_pool_worker(),
restores CPU affinity, and clears WORKER_UNBOUND.
There are a couple subtleties. All bound idle workers should have
their runqueues set to that of the bound CPU; however, if the target
task isn't running, set_cpus_allowed_ptr() just updates the
cpus_allowed mask deferring the actual migration to when the task
wakes up. This is worked around by waking up idle workers after
restoring CPU affinity before any workers can become bound.
Another subtlety is stems from matching @pool->nr_running with the
number of running unbound workers. While DISASSOCIATED, all workers
are unbound and nr_running is zero. As workers become bound again,
nr_running needs to be adjusted accordingly; however, there is no good
way to tell whether a given worker is running without poking into
scheduler internals. Instead of clearing UNBOUND directly,
rebind_workers() replaces UNBOUND with another new NOT_RUNNING flag -
REBOUND, which will later be cleared by the workers themselves while
preparing for the next round of work item execution. The only change
needed for the workers is clearing REBOUND along with PREP.
* This patch leaves for_each_busy_worker() without any user. Removed.
* idle_worker_rebind(), busy_worker_rebind_fn(), worker->rebind_work
and rebind logic in manager_workers() removed.
* worker_thread() now looks at WORKER_DIE instead of testing whether
@worker->entry is empty to determine whether it needs to do
something special as dying is the only special thing now.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
rebind_workers() will be reimplemented in a way which makes it mostly
decoupled from the rest of worker management. Move rebind_workers()
so that it's located with other CPU hotplug related functions.
This patch is pure function relocation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Make worker_ida an idr - worker_idr and use it to implement
for_each_pool_worker() which will be used to simplify worker rebinding
on CPU_ONLINE.
pool->worker_idr is protected by both pool->manager_mutex and
pool->lock so that it can be iterated while holding either lock.
* create_worker() allocates ID without installing worker pointer and
installs the pointer later using idr_replace(). This is because
worker ID is needed when creating the actual task to name it and the
new worker shouldn't be visible to iterations before fully
initialized.
* In destroy_worker(), ID removal is moved before kthread_stop().
This is again to guarantee that only fully working workers are
visible to for_each_pool_worker().
Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
PF_THREAD_BOUND was originally used to mark kernel threads which were
bound to a specific CPU using kthread_bind() and a task with the flag
set allows cpus_allowed modifications only to itself. Workqueue is
currently abusing it to prevent userland from meddling with
cpus_allowed of workqueue workers.
What we need is a flag to prevent userland from messing with
cpus_allowed of certain kernel tasks. In kernel, anyone can
(incorrectly) squash the flag, and, for worker-type usages,
restricting cpus_allowed modification to the task itself doesn't
provide meaningful extra proection as other tasks can inject work
items to the task anyway.
This patch replaces PF_THREAD_BOUND with PF_NO_SETAFFINITY.
sched_setaffinity() checks the flag and return -EINVAL if set.
set_cpus_allowed_ptr() is no longer affected by the flag.
This will allow simplifying workqueue worker CPU affinity management.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.19 (GNU/Linux)
iQEcBAABAgAGBQJRRkrbAAoJEHm+PkMAQRiGy3oH/jrbHinYs0auurANgx4TdtWT
/WNajstKBqLOJJ6cnTR7sOqwOVlptt65EbbTs+qGyZ2Z2W/Lg0BMenHvNHo4ER8C
e7UbMdBCSLKBjAMKh1XCoZscGv4Exm8WRH3Vc5yP0Hafj3EzSAVLY1dta9WKKoQi
bh7D1ErUlbU1zczA1w5YbPF0LqFKRvyZOwebMCCAKAxv5wWAxmbcPNxVR4sufkjg
k6TkQ2ysgWivZAfy3tJYOcxiEu7ahpZVEuYdlZEJQXHRQUfoNljQlOp4BqKsYUai
5A0kaf2VpKay/7pkhvTfBBcF/jFJ68pYP6gQ2ThNdr0b5kOiAfMWj030Xyngnhg=
=iO9t
-----END PGP SIGNATURE-----
Merge tag 'v3.9-rc3' into drm-intel-next-queued
Backmerge so that I can merge Imre Deak's coalesced sg entries fixes,
which depend upon the new for_each_sg_page introduce in
commit a321e91b6d
Author: Imre Deak <imre.deak@intel.com>
Date: Wed Feb 27 17:02:56 2013 -0800
lib/scatterlist: add simple page iterator
The merge itself is just two trivial conflicts:
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Pull workqueue fix from Tejun Heo:
"Lai's patch to fix highly unlikely but still possible workqueue stall
during CPU hotunplug."
* 'for-3.9-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq:
workqueue: fix possible pool stall bug in wq_unbind_fn()
On allocation failure, it would fail to free the old attrs array which
was no longer referenced by anything (since it would free the old
module_param_attrs struct on the way out).
Comment the suspicious-looking krealloc() usage to explain why it *isn't*
actually buggy, despite looking like a classic realloc() usage bug.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
It's a per-cpu data structure but missed the __percpu annotation.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Link: http://lkml.kernel.org/r/1363600594-11453-1-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Thomas noted that we do the wakeup preemption check after the
wakeup trace point, this means the tracepoint cannot test/report
this decision; which is rather important for latency sensitive
workloads. Therefore move the tracepoint after doing the
preemption check.
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Paul Turner <pjt@google.com>
Cc: Mike Galbraith <efault@gmx.de>
Link: http://lkml.kernel.org/r/1363254519.26965.9.camel@laptop
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull CPU runtime stats/accounting fixes from Frederic Weisbecker:
" Some users are complaining that their threadgroup's runtime accounting
freezes after a week or so of intense cpu-bound workload. This set tries
to fix the issue by reducing the risk of multiplication overflow in the
cputime scaling code. "
Stanislaw Gruszka further explained the historic context and impact of the
bug:
" Commit 0cf55e1ec0 start to use scalling
for whole thread group, so increase chances of hitting multiplication
overflow, depending on how many CPUs are on the system.
We have multiplication utime * rtime for one thread since commit
b27f03d4bd.
Overflow will happen after:
rtime * utime > 0xffffffffffffffff jiffies
if thread utilize 100% of CPU time, that gives:
rtime > sqrt(0xffffffffffffffff) jiffies
ritme > sqrt(0xffffffffffffffff) / (24 * 60 * 60 * HZ) days
For HZ 100 it will be 497 days for HZ 1000 it will be 49 days.
Bug affect only users, who run CPU intensive application for that
long period. Also they have to be interested on utime,stime values,
as bug has no other visible effect as making those values incorrect. "
Signed-off-by: Ingo Molnar <mingo@kernel.org>
perf_event_task_event() iterates pmu list and generate events
for each eligible pmu context. But if task_event has task_ctx
like in EXIT it'll generate events even though the pmu doesn't
have an eligible one. Fix it by moving the code to proper
places.
Before this patch:
$ perf record -n true
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.006 MB perf.data (~248 samples) ]
$ perf report -D | tail
Aggregated stats:
TOTAL events: 73
MMAP events: 67
COMM events: 2
EXIT events: 4
cycles stats:
TOTAL events: 73
MMAP events: 67
COMM events: 2
EXIT events: 4
After this patch:
$ perf report -D | tail
Aggregated stats:
TOTAL events: 70
MMAP events: 67
COMM events: 2
EXIT events: 1
cycles stats:
TOTAL events: 70
MMAP events: 67
COMM events: 2
EXIT events: 1
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1363332433-7637-1-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When cpu/task clock events are initialized, their sampling
frequencies are converted to have a fixed value. However it
missed to update the hwc->last_period which was set to 1 for
initial sampling frequency calibration.
Because this hwc->last_period value is used as a period in
perf_swevent_ hrtime(), every recorded sample will have an
incorrected period of 1.
$ perf record -e task-clock noploop 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.158 MB perf.data (~6919 samples) ]
$ perf report -n --show-total-period --stdio
# Samples: 4K of event 'task-clock'
# Event count (approx.): 4000
#
# Overhead Samples Period Command Shared Object Symbol
# ........ ............ ............ ....... ............. ..................
#
99.95% 3998 3998 noploop noploop [.] main
0.03% 1 1 noploop libc-2.15.so [.] init_cacheinfo
0.03% 1 1 noploop ld-2.15.so [.] open_verify
Note that it doesn't affect the non-sampling event so that the
perf stat still gets correct value with or without this patch.
$ perf stat -e task-clock noploop 1
Performance counter stats for 'noploop 1':
1000.272525 task-clock # 1.000 CPUs utilized
1.000560605 seconds time elapsed
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1363574507-18808-1-git-send-email-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
There are some new processors whose TSC clocksource won't stop during
suspend. Currently, after system resumes, kernel will use persistent
clock or RTC to compensate the sleep time, but with these nonstop
clocksources, we could skip the special compensation from external
sources, and just use current clocksource for time recounting.
This can solve some time drift bugs caused by some not-so-accurate or
error-prone RTC devices.
The current way to count suspended time is first try to use the persistent
clock, and then try the RTC if persistent clock can't be used. This
patch will change the trying order to:
suspend-nonstop clocksource -> persistent clock -> RTC
When counting the sleep time with nonstop clocksource, use an accurate way
suggested by Jason Gunthorpe to cover very large delta cycles.
Signed-off-by: Feng Tang <feng.tang@intel.com>
[jstultz: Small optimization, avoiding re-reading the clocksource]
Signed-off-by: John Stultz <john.stultz@linaro.org>
When warping the clock (from a local time RTC), use
timekeeping_inject_offset() to atomically add the offset.
This avoids any minor time error caused by the delay between
reading the time, and then setting the adjusted time.
Signed-off-by: John Stultz <john.stultz@linaro.org>
If the Hardware Clock kept in local time,kernel will adjust the time
to be UTC time.But if Hardware Clock kept in UTC time,system will make
a dummy settimeofday call first (sys_tz.tz_minuteswest = 0) to make sure
the time is not shifted,so at this point I think maybe it is not necessary
to set the kernel time once the sys_tz.tz_minuteswest is zero.
Signed-off-by: Dong Zhu <bluezhudong@gmail.com>
[jstultz: Updated to merge with conflicting changes ]
Signed-off-by: John Stultz <john.stultz@linaro.org>
ftrace_dump() had a lot of issues. What ftrace_dump() does, is when
ftrace_dump_on_oops is set (via a kernel parameter or sysctl), it
will dump out the ftrace buffers to the console when either a oops,
panic, or a sysrq-z occurs.
This was written a long time ago when ftrace was fragile to recursion.
But it wasn't written well even for that.
There's a possible deadlock that can occur if a ftrace_dump() is happening
and an NMI triggers another dump. This is because it grabs a lock
before checking if the dump ran.
It also totally disables ftrace, and tracing for no good reasons.
As the ring_buffer now checks if it is read via a oops or NMI, where
there's a chance that the buffer gets corrupted, it will disable
itself. No need to have ftrace_dump() do the same.
ftrace_dump() is now cleaned up where it uses an atomic counter to
make sure only one dump happens at a time. A simple atomic_inc_return()
is enough that is needed for both other CPUs and NMIs. No need for
a spinlock, as if one CPU is running the dump, no other CPU needs
to do it too.
The tracing_on variable is turned off and not turned on. The original
code did this, but it wasn't pretty. By just disabling this variable
we get the result of not seeing traces that happen between crashes.
For sysrq-z, it doesn't get turned on, but the user can always write
a '1' to the tracing_on file. If they are using sysrq-z, then they should
know about tracing_on.
The new code is much easier to read and less error prone. No more
deadlock possibility when an NMI triggers here.
Reported-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
Cc: stable@vger.kernel.org
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
trace_event_mutex is an rw semaphore now, not a mutex, change the name.
Link: http://lkml.kernel.org/r/513D843B.40109@huawei.com
Signed-off-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
[ Forward ported to my new code ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
ppc64 has its own syscall prefix like ".SyS" or ".sys". Make the
comment in arch_syscall_match_sym_name() more understandable.
Link: http://lkml.kernel.org/r/513D842F.40205@huawei.com
Signed-off-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
trace_destroy_fields() is not used outside of the file. It can be
a static function.
Link: http://lkml.kernel.org/r/513D842A.2000907@huawei.com
Signed-off-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
By moving find_event_field() and trace_find_field() into trace_events.c,
the ftrace_common_fields list and trace_get_fields() can become local to
the trace_events.c file.
find_event_field() is renamed to trace_find_event_field() to conform to
the tracing global function names.
Link: http://lkml.kernel.org/r/513D8426.9070109@huawei.com
Signed-off-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
[ rostedt: Modified trace_find_field() to trace_find_event_field() ]
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
TRACE_MAX_PRINT macro is defined, but is not used.
Link: http://lkml.kernel.org/r/513D8421.4070404@huawei.com
Signed-off-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Use pr_warn_once, instead of making an open coded implementation.
Link: http://lkml.kernel.org/r/513D8419.20400@huawei.com
Signed-off-by: zhangwei(Jovi) <jovi.zhangwei@huawei.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When testing my large changes to the ftrace system, there was
a bug that looked like the ring buffer was dropping events.
I wrote up a quick integrity checker of the ring buffer to
see if it was.
Although the bug ended up being something stupid I did in ftrace,
and had nothing to do with the ring buffer, I figured if I spent
the time to write up this test, I might as well include it in the
kernel.
I cleaned it up a bit, as the original version was rather ugly.
Not saying this version is pretty, but it's a beauty queen
compared to what I original wrote.
To enable the start up test, set CONFIG_RING_BUFFER_STARTUP_TEST.
Note, it runs for 10 seconds, so it will slow your boot time
by at least 10 more seconds.
What it does is documented in both the comments and the Kconfig
help.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
We have CONFIG_SYMBOL_PREFIX, which three archs define to the string
"_". But Al Viro broke this in "consolidate cond_syscall and
SYSCALL_ALIAS declarations" (in linux-next), and he's not the first to
do so.
Using CONFIG_SYMBOL_PREFIX is awkward, since we usually just want to
prefix it so something. So various places define helpers which are
defined to nothing if CONFIG_SYMBOL_PREFIX isn't set:
1) include/asm-generic/unistd.h defines __SYMBOL_PREFIX.
2) include/asm-generic/vmlinux.lds.h defines VMLINUX_SYMBOL(sym)
3) include/linux/export.h defines MODULE_SYMBOL_PREFIX.
4) include/linux/kernel.h defines SYMBOL_PREFIX (which differs from #7)
5) kernel/modsign_certificate.S defines ASM_SYMBOL(sym)
6) scripts/modpost.c defines MODULE_SYMBOL_PREFIX
7) scripts/Makefile.lib defines SYMBOL_PREFIX on the commandline if
CONFIG_SYMBOL_PREFIX is set, so that we have a non-string version
for pasting.
(arch/h8300/include/asm/linkage.h defines SYMBOL_NAME(), too).
Let's solve this properly:
1) No more generic prefix, just CONFIG_HAVE_UNDERSCORE_SYMBOL_PREFIX.
2) Make linux/export.h usable from asm.
3) Define VMLINUX_SYMBOL() and VMLINUX_SYMBOL_STR().
4) Make everyone use them.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Reviewed-by: James Hogan <james.hogan@imgtec.com>
Tested-by: James Hogan <james.hogan@imgtec.com> (metag)
The function trace_clock() calls "local_clock()" which is exactly
the same clock that perf uses. I'm not sure why perf doesn't call
trace_clock(), as trace_clock() doesn't have any users.
But now it does. As trace_clock() calls local_clock() like perf does,
I added the trace_clock "perf" option that uses trace_clock().
Now the ftrace buffers can use the same clock as perf uses. This
will be useful when perf starts reading the ftrace buffers, and will
be able to interleave them with the same clock data.
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Add a simple trace clock called "uptime" for those that are
interested in the uptime of the trace. It uses jiffies as that's
the safest method, as other uptime clocks grab seq locks, which could
cause a deadlock if taken from an event or function tracer.
Requested-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently, the only way to stop the latency tracers from doing function
tracing is to fully disable the function tracer from the proc file
system:
echo 0 > /proc/sys/kernel/ftrace_enabled
This is a big hammer approach as it disables function tracing for
all users. This includes kprobes, perf, stack tracer, etc.
Instead, create a function-trace option that the latency tracers can
check to determine if it should enable function tracing or not.
This option can be set or cleared even while the tracer is active
and the tracers will disable or enable function tracing depending
on how the option was set.
Instead of using the proc file, disable latency function tracing with
echo 0 > /debug/tracing/options/function-trace
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Clark Williams <williams@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently, the depth reported in the stack tracer stack_trace file
does not match the stack_max_size file. This is because the stack_max_size
includes the overhead of stack tracer itself while the depth does not.
The first time a max is triggered, a calculation is not performed that
figures out the overhead of the stack tracer and subtracts it from
the stack_max_size variable. The overhead is stored and is subtracted
from the reported stack size for comparing for a new max.
Now the stack_max_size corresponds to the reported depth:
# cat stack_max_size
4640
# cat stack_trace
Depth Size Location (48 entries)
----- ---- --------
0) 4640 32 _raw_spin_lock+0x18/0x24
1) 4608 112 ____cache_alloc+0xb7/0x22d
2) 4496 80 kmem_cache_alloc+0x63/0x12f
3) 4416 16 mempool_alloc_slab+0x15/0x17
[...]
While testing against and older gcc on x86 that uses mcount instead
of fentry, I found that pasing in ip + MCOUNT_INSN_SIZE let the
stack trace show one more function deep which was missing before.
Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When gcc 4.6 on x86 is used, the function tracer will use the new
option -mfentry which does a call to "fentry" at every function
instead of "mcount". The significance of this is that fentry is
called as the first operation of the function instead of the mcount
usage of being called after the stack.
This causes the stack tracer to show some bogus results for the size
of the last function traced, as well as showing "ftrace_call" instead
of the function. This is due to the stack frame not being set up
by the function that is about to be traced.
# cat stack_trace
Depth Size Location (48 entries)
----- ---- --------
0) 4824 216 ftrace_call+0x5/0x2f
1) 4608 112 ____cache_alloc+0xb7/0x22d
2) 4496 80 kmem_cache_alloc+0x63/0x12f
The 216 size for ftrace_call includes both the ftrace_call stack
(which includes the saving of registers it does), as well as the
stack size of the parent.
To fix this, if CC_USING_FENTRY is defined, then the stack_tracer
will reserve the first item in stack_dump_trace[] array when
calling save_stack_trace(), and it will fill it in with the parent ip.
Then the code will look for the parent pointer on the stack and
give the real size of the parent's stack pointer:
# cat stack_trace
Depth Size Location (14 entries)
----- ---- --------
0) 2640 48 update_group_power+0x26/0x187
1) 2592 224 update_sd_lb_stats+0x2a5/0x4ac
2) 2368 160 find_busiest_group+0x31/0x1f1
3) 2208 256 load_balance+0xd9/0x662
I'm Cc'ing stable, although it's not urgent, as it only shows bogus
size for item #0, the rest of the trace is legit. It should still be
corrected in previous stable releases.
Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Use the stack of stack_trace_call() instead of check_stack() as
the test pointer for max stack size. It makes it a bit cleaner
and a little more accurate.
Adding stable, as a later fix depends on this patch.
Cc: stable@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Altough the trace_dump_stack() already skips three functions in
the call to stack trace, which gets the stack trace to start
at the caller of the function, the caller may want to skip some
more too (as it may have helper functions).
Add a skip argument to the trace_dump_stack() that lets the caller
skip back tracing functions that it doesn't care about.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Add triggers to function tracer that lets an event get enabled or
disabled when a function is called:
format is:
<function>:enable_event:<system>:<event>[:<count>]
<function>:disable_event:<system>:<event>[:<count>]
echo 'schedule:enable_event:sched:sched_switch' > /debug/tracing/set_ftrace_filter
Every time schedule is called, it will enable the sched_switch event.
echo 'schedule:disable_event:sched:sched_switch:2' > /debug/tracing/set_ftrace_filter
The first two times schedule is called while the sched_switch
event is enabled, it will disable it. It will not count for a time
that the event is already disabled (or enabled for enable_event).
[ fixed return without mutex_unlock() - thanks to Dan Carpenter and smatch ]
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
In order to let triggers enable or disable events, we need a 'soft'
method for doing so. For example, if a function probe is added that
lets a user enable or disable events when a function is called, that
change must be done without taking locks or a mutex, and definitely
it can't sleep. But the full enabling of a tracepoint is expensive.
By adding a 'SOFT_DISABLE' flag, and converting the flags to be updated
without the protection of a mutex (using set/clear_bit()), this soft
disable flag can be used to allow critical sections to enable or disable
events from being traced (after the event has been placed into "SOFT_MODE").
Some caveats though: The comm recorder (to map pids with a comm) can not
be soft disabled (yet). If you disable an event with with a "soft"
disable and wait a while before reading the trace, the comm cache may be
replaced and you'll get a bunch of <...> for comms in the trace.
Reading the "enable" file for an event that is disabled will now give
you "0*" where the '*' denotes that the tracepoint is still active but
the event itself is "disabled".
[ fixed _BIT used in & operation : thanks to Dan Carpenter and smatch ]
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Tom Zanussi <tom.zanussi@linux.intel.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
The entries to the probe hash must be freed after a synchronize_sched()
after the entry has been removed from the hash.
As the entries are registered with ops that may have their own callbacks,
and these callbacks may sleep, we can not use call_rcu_sched() because
the rcu callbacks registered with that are called from a softirq context.
Instead of using call_rcu_sched(), manually save the entries on a free_list
and at the end of the loop that removes the entries, do a synchronize_sched()
and then go through the free_list, freeing the entries.
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
When a function probe is created, each function that the probe is
attached to, a "callback" method is called. On release of the probe,
each function entry calls the "free" method.
First, "callback" is a confusing name and does not really match what
it does. Callback sounds like it will be called when the probe
triggers. But that's not the case. This is really an "init" function,
so lets rename it as such.
Secondly, both "init" and "free" do not pass enough information back
to the handlers. Pass back the ops, ip and data for each time the
method is called. We have the information, might as well use it.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
echo 'schedule:snapshot:1' > /debug/tracing/set_ftrace_filter
This will cause the scheduler to trigger a snapshot the next time
it's called (you can use any function that's not called by NMI).
Even though it triggers only once, you still need to remove it with:
echo '!schedule:snapshot:0' > /debug/tracing/set_ftrace_filter
The :1 can be left off for the first command:
echo 'schedule:snapshot' > /debug/tracing/set_ftrace_filter
But this will cause all calls to schedule to trigger a snapshot.
This must be removed without the ':0'
echo '!schedule:snapshot' > /debug/tracing/set_ftrace_filter
As adding a "count" is a different operation (internally).
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Add alloc_snapshot() and free_snapshot() to allocate and free the
snapshot buffer respectively, and use these to remove duplicate
code.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Currently the function probe enables all functions and runs a "hash"
against every function call to see if it should call a probe. This
is extremely wasteful.
Note, a probe is something like:
echo schedule:traceoff > /debug/tracing/set_ftrace_filter
When schedule is called, the probe will disable tracing. But currently,
it has a call back for *all* functions, and checks to see if the
called function is the probe that is needed.
The probe function has been created before ftrace was rewritten to
allow for more than one "op" to be registered by the function tracer.
When probes were created, it couldn't limit the functions without also
limiting normal function calls. But now we can, it's about time
to update the probe code.
Todo, have separate ops for different entries. That is, assign
a ftrace_ops per probe, instead of one op for all probes. But
as there's not many probes assigned, this may not be that urgent.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
