~shefty/rdma-dev.git
9 years agotracing: Remove unnecessary WARN_ONCE's from tracing_buffers_splice_read
Dave Jones [Wed, 25 Jul 2012 15:39:08 +0000 (11:39 -0400)]
tracing: Remove unnecessary WARN_ONCE's from tracing_buffers_splice_read

WARN shouldn't be used as a means of communicating failure to a userspace programmer.

Link: http://lkml.kernel.org/r/20120725153908.GA25203@redhat.com
Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Remove unneeded checks from the stack tracer
Anton Vorontsov [Wed, 18 Jul 2012 18:56:01 +0000 (11:56 -0700)]
tracing: Remove unneeded checks from the stack tracer

It seems that 'ftrace_enabled' flag should not be used inside the tracer
functions. The ftrace core is using this flag for internal purposes, and
the flag wasn't meant to be used in tracers' runtime checks.

stack tracer is the only tracer that abusing the flag. So stop it from
serving as a bad example.

Also, there is a local 'stack_trace_disabled' flag in the stack tracer,
which is never updated; so it can be removed as well.

Link: http://lkml.kernel.org/r/1342637761-9655-1-git-send-email-anton.vorontsov@linaro.org
Signed-off-by: Anton Vorontsov <anton.vorontsov@linaro.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Add a resize function to make one buffer equivalent to another buffer
Hiraku Toyooka [Wed, 17 Oct 2012 02:56:16 +0000 (11:56 +0900)]
tracing: Add a resize function to make one buffer equivalent to another buffer

Trace buffer size is now per-cpu, so that there are the following two
patterns in resizing of buffers.

  (1) resize per-cpu buffers to same given size
  (2) resize per-cpu buffers to another trace_array's buffer size
      for each CPU (such as preparing the max_tr which is equivalent
      to the global_trace's size)

__tracing_resize_ring_buffer() can be used for (1), and had
implemented (2) inside it for resetting the global_trace to the
original size.

(2) was also implemented in another place. So this patch assembles
them in a new function - resize_buffer_duplicate_size().

Link: http://lkml.kernel.org/r/20121017025616.2627.91226.stgit@falsita
Signed-off-by: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Kill unused and puzzled sample code in ftrace.h
Shan Wei [Sat, 3 Nov 2012 04:38:33 +0000 (12:38 +0800)]
tracing: Kill unused and puzzled sample code in ftrace.h

When doing per-cpu helper optimizing work, find that this code is so puzzled.
1. It's mark as comment text, maybe a sample function for guidelines
   or a todo work.
2. But, this sample code is odd where struct perf_trace_buf is nonexistent.
   commit ce71b9 delete struct perf_trace_buf definition.

   Author: Frederic Weisbecker <fweisbec@gmail.com>
   Date:   Sun Nov 22 05:26:55 2009 +0100

   tracing: Use the perf recursion protection from trace event

Is it necessary to keep there?
just compile test.

Link: http://lkml.kernel.org/r/50949FC9.6050202@gmail.com
Signed-off-by: Shan Wei <davidshan@tencent.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Show raw time stamp on stats per cpu using counter or tsc mode for trace_clock
Yoshihiro YUNOMAE [Tue, 13 Nov 2012 20:18:23 +0000 (12:18 -0800)]
tracing: Show raw time stamp on stats per cpu using counter or tsc mode for trace_clock

Show raw time stamp values for stats per cpu if you choose counter or tsc mode
for trace_clock. Although a unit of tracing time stamp is nsec in local or global mode,
the units in counter and TSC mode are tracing counter and cycles respectively.
Link: http://lkml.kernel.org/r/1352837903-32191-3-git-send-email-dhsharp@google.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Signed-off-by: Yoshihiro YUNOMAE <yoshihiro.yunomae.ez@hitachi.com>
Signed-off-by: David Sharp <dhsharp@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Format non-nanosec times from tsc clock without a decimal point.
David Sharp [Tue, 13 Nov 2012 20:18:22 +0000 (12:18 -0800)]
tracing: Format non-nanosec times from tsc clock without a decimal point.

With the addition of the "tsc" clock, formatting timestamps to look like
fractional seconds is misleading. Mark clocks as either in nanoseconds or
not, and format non-nanosecond timestamps as decimal integers.

Tested:
$ cd /sys/kernel/debug/tracing/
$ cat trace_clock
[local] global tsc
$ echo sched_switch > set_event
$ echo 1 > tracing_on ; sleep 0.0005 ; echo 0 > tracing_on
$ cat trace
          <idle>-0     [000]  6330.555552: sched_switch: prev_comm=swapper prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=bash next_pid=29964 next_prio=120
           sleep-29964 [000]  6330.555628: sched_switch: prev_comm=bash prev_pid=29964 prev_prio=120 prev_state=S ==> next_comm=swapper next_pid=0 next_prio=120
  ...
$ echo 1 > options/latency-format
$ cat trace
  <idle>-0       0 4104553247us+: sched_switch: prev_comm=swapper prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=bash next_pid=29964 next_prio=120
   sleep-29964   0 4104553322us+: sched_switch: prev_comm=bash prev_pid=29964 prev_prio=120 prev_state=S ==> next_comm=swapper next_pid=0 next_prio=120
  ...
$ echo tsc > trace_clock
$ cat trace
$ echo 1 > tracing_on ; sleep 0.0005 ; echo 0 > tracing_on
$ echo 0 > options/latency-format
$ cat trace
          <idle>-0     [000] 16490053398357: sched_switch: prev_comm=swapper prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=bash next_pid=31128 next_prio=120
           sleep-31128 [000] 16490053588518: sched_switch: prev_comm=bash prev_pid=31128 prev_prio=120 prev_state=S ==> next_comm=swapper next_pid=0 next_prio=120
  ...
echo 1 > options/latency-format
$ cat trace
  <idle>-0       0 91557653238+: sched_switch: prev_comm=swapper prev_pid=0 prev_prio=120 prev_state=R ==> next_comm=bash next_pid=31128 next_prio=120
   sleep-31128   0 91557843399+: sched_switch: prev_comm=bash prev_pid=31128 prev_prio=120 prev_state=S ==> next_comm=swapper next_pid=0 next_prio=120
  ...

v2:
Move arch-specific bits out of generic code.
v4:
Fix x86_32 build due to 64-bit division.

Google-Bug-Id: 6980623
Link: http://lkml.kernel.org/r/1352837903-32191-2-git-send-email-dhsharp@google.com
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: David Sharp <dhsharp@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing,x86: Add a TSC trace_clock
David Sharp [Tue, 13 Nov 2012 20:18:21 +0000 (12:18 -0800)]
tracing,x86: Add a TSC trace_clock

In order to promote interoperability between userspace tracers and ftrace,
add a trace_clock that reports raw TSC values which will then be recorded
in the ring buffer. Userspace tracers that also record TSCs are then on
exactly the same time base as the kernel and events can be unambiguously
interlaced.

Tested: Enabled a tracepoint and the "tsc" trace_clock and saw very large
timestamp values.

v2:
Move arch-specific bits out of generic code.
v3:
Rename "x86-tsc", cleanups
v7:
Generic arch bits in Kbuild.

Google-Bug-Id: 6980623
Link: http://lkml.kernel.org/r/1352837903-32191-1-git-send-email-dhsharp@google.com
Acked-by: Ingo Molnar <mingo@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@linux.intel.com>
Signed-off-by: David Sharp <dhsharp@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Add trace_options kernel command line parameter
Steven Rostedt [Fri, 2 Nov 2012 02:56:07 +0000 (22:56 -0400)]
tracing: Add trace_options kernel command line parameter

Add trace_options to the kernel command line parameter to be able to
set options at early boot. For example, to enable stack dumps of
events, add the following:

  trace_options=stacktrace

This along with the trace_event option, you can get not only
traces of the events but also the stack dumps with them.

Requested-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Use irq_work for wake ups and remove *_nowake_*() functions
Steven Rostedt [Fri, 2 Nov 2012 00:54:21 +0000 (20:54 -0400)]
tracing: Use irq_work for wake ups and remove *_nowake_*() functions

Have the ring buffer commit function use the irq_work infrastructure to
wake up any waiters waiting on the ring buffer for new data. The irq_work
was created for such a purpose, where doing the actual wake up at the
time of adding data is too dangerous, as an event or function trace may
be in the midst of the work queue locks and cause deadlocks. The irq_work
will either delay the action to the next timer interrupt, or trigger an IPI
to itself forcing an interrupt to do the work (in a safe location).

With irq_work, all ring buffer commits can safely do wakeups, removing
the need for the ring buffer commit "nowake" variants, which were used
by events and function tracing. All commits can now safely use the
normal commit, and the "nowake" variants can be removed.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Remove deprecated tracing_enabled file
Steven Rostedt [Thu, 1 Nov 2012 15:51:40 +0000 (11:51 -0400)]
tracing: Remove deprecated tracing_enabled file

The tracing_enabled file was used as a quick way to stop
tracers, and try to bring down overhead for things like
the latency tracers (irqsoff, wakeup, etc). But it didn't
work that well.

The tracing_on file was created as a really fast way to
stop recording into the ftrace ring buffer and can interact
with the kernel. That is a tracing_off() call in the kernel
can disable recording of events, and then from userspace one
could echo 1 into the tracing_on file to continue it. The
tracing_enabled function did too much to allow for this.

The tracing_on has taken over as a way to start and stop tracing
and the tracing_enabled file should not be used. But because of
its existance, it still confuses people. Over a year ago the
following commit was added:

 commit 6752ab4a9c30d5411b2dfdb251a3f1cb18aae487
 Author: Steven Rostedt <srostedt@redhat.com>
 Date:   Tue Feb 8 13:54:06 2011 -0500

    tracing: Deprecate tracing_enabled for tracing_on

This commit added a WARN_ON() if the tracing_enabled file's variable
was changed. After this was added, only LatencyTop complained, and
they soon fixed their tool as there was no reason that LatencyTop
should touch this file as it was using the perf ring buffers which
this file does not interact with. But since that time no one else
has complained about this WARN_ON(). Thus it is safe to assume that
this file is no longer needed. Time to get rid of it.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Make tracing_enabled be equal to tracing_on
Steven Rostedt [Fri, 11 May 2012 18:25:30 +0000 (14:25 -0400)]
tracing: Make tracing_enabled be equal to tracing_on

The tracing_enabled file has been deprecated as it never was able
to serve its purpose well. The tracing_on file has taken over.
Instead of having code to keep tracing_enabled, have the tracing_enabled
file just set tracing_on, and remove the tracing_enabled variable.

This allows us to remove the tracing_enabled file. The reason that
the remove is in a different change set and not removed here is
in case we find some lonely userspace tool that requires the file
to exist. Then the removal patch will get reverted, but this one
will not.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Remove unused function unregister_tracer()
Steven Rostedt [Sat, 12 May 2012 00:54:53 +0000 (20:54 -0400)]
tracing: Remove unused function unregister_tracer()

The function register_tracer() is only used by kernel core code,
that never needs to remove the tracer. As trace_events have become
the main way to add new tracing to the kernel, the need to
unregister a tracer has diminished. Remove the unused function
unregister_tracer(). If a need arises where we need it, then we
can always add it back.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Separate open function from set_event and available_events
Steven Rostedt [Thu, 3 May 2012 18:57:28 +0000 (14:57 -0400)]
tracing: Separate open function from set_event and available_events

The open function used by available_events is the same as set_event even
though it uses different seq functions. This causes a side effect of
writing into available_events clearing all events, even though
available_events is suppose to be read only.

There's no reason to keep a single function for just the open and have
both use different functions for everything else. It is a little
confusing and causes strange behavior. Just have each have their own
function.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agoring-buffer: Change unsigned long type of ring_buffer_oldest_event_ts() to u64
Yoshihiro YUNOMAE [Thu, 11 Oct 2012 23:27:54 +0000 (16:27 -0700)]
ring-buffer: Change unsigned long type of ring_buffer_oldest_event_ts() to u64

ring_buffer_oldest_event_ts() should return a value of u64 type, because
ring_buffer_per_cpu->buffer_page->buffer_data_page->time_stamp is u64 type.

Link: http://lkml.kernel.org/r/1349998076-15495-5-git-send-email-dhsharp@google.com
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Yoshihiro YUNOMAE <yoshihiro.yunomae.ez@hitachi.com>
Signed-off-by: David Sharp <dhsharp@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Reset ring buffer when changing trace_clocks
David Sharp [Thu, 11 Oct 2012 23:27:52 +0000 (16:27 -0700)]
tracing: Reset ring buffer when changing trace_clocks

Because the "tsc" clock isn't in nanoseconds, the ring buffer must be
reset when changing clocks so that incomparable timestamps don't end up
in the same trace.

Tested: Confirmed switching clocks resets the trace buffer.

Google-Bug-Id: 6980623
Link: http://lkml.kernel.org/r/1349998076-15495-3-git-send-email-dhsharp@google.com
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: David Sharp <dhsharp@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agolinux/kernel.h: Remove duplicate trace_printk declaration
Michal Hocko [Thu, 25 Oct 2012 13:41:51 +0000 (15:41 +0200)]
linux/kernel.h: Remove duplicate trace_printk declaration

!CONFIG_TRACING both declares and defines (empty) trace_printk.
The first one is not redundant so it can be removed.

Link: http://lkml.kernel.org/r/1351172511-18125-1-git-send-email-mhocko@suse.cz
Signed-off-by: Michal Hocko <mhocko@suse.cz>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Cleanup unnecessary function declarations
Vaibhav Nagarnaik [Thu, 7 Jun 2012 23:46:25 +0000 (16:46 -0700)]
tracing: Cleanup unnecessary function declarations

The functions defined in include/trace/syscalls.h are not used directly
since struct ftrace_event_class was introduced. Remove them from the
header file and rearrange the ftrace_event_class declarations in
trace_syscalls.c.

Link: http://lkml.kernel.org/r/1339112785-21806-2-git-send-email-vnagarnaik@google.com
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Trivial cleanup
David Sharp [Thu, 7 Jun 2012 23:46:24 +0000 (16:46 -0700)]
tracing: Trivial cleanup

Remove ftrace_format_syscall() declaration; it is neither defined nor
used. Also update a comment and formatting.

Link: http://lkml.kernel.org/r/1339112785-21806-1-git-send-email-vnagarnaik@google.com
Signed-off-by: David Sharp <dhsharp@google.com>
Signed-off-by: Vaibhav Nagarnaik <vnagarnaik@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Cache comms only after an event occurred
Steven Rostedt [Thu, 11 Oct 2012 16:14:25 +0000 (12:14 -0400)]
tracing: Cache comms only after an event occurred

Whenever an event is registered, the comm of tasks are saved at
every task switch instead of saving them at every event. But if
an event isn't executed much, the comm cache will be filled up
by tasks that did not record the event and you lose out on the comms
that did.

Here's an example, if you enable the following events:

echo 1 > /debug/tracing/events/kvm/kvm_cr/enable
echo 1 > /debug/tracing/events/net/net_dev_xmit/enable

Note, there's no kvm running on this machine so the first event will
never be triggered, but because it is enabled, the storing of comms
will continue. If we now disable the network event:

echo 0 > /debug/tracing/events/net/net_dev_xmit/enable

and look at the trace:

cat /debug/tracing/trace
            sshd-2672  [001] ..s2   375.731616: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=242 rc=0
            sshd-2672  [001] ..s1   375.731617: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=242 rc=0
            sshd-2672  [001] ..s2   375.859356: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=242 rc=0
            sshd-2672  [001] ..s1   375.859357: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=242 rc=0
            sshd-2672  [001] ..s2   375.947351: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=242 rc=0
            sshd-2672  [001] ..s1   375.947352: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=242 rc=0
            sshd-2672  [001] ..s2   376.035383: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=242 rc=0
            sshd-2672  [001] ..s1   376.035383: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=242 rc=0
            sshd-2672  [001] ..s2   377.563806: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=226 rc=0
            sshd-2672  [001] ..s1   377.563807: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=226 rc=0
            sshd-2672  [001] ..s2   377.563834: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6be0 len=114 rc=0
            sshd-2672  [001] ..s1   377.563842: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6be0 len=114 rc=0

We see that process 2672 which triggered the events has the comm "sshd".
But if we run hackbench for a bit and look again:

cat /debug/tracing/trace
           <...>-2672  [001] ..s2   375.731616: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=242 rc=0
           <...>-2672  [001] ..s1   375.731617: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=242 rc=0
           <...>-2672  [001] ..s2   375.859356: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=242 rc=0
           <...>-2672  [001] ..s1   375.859357: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=242 rc=0
           <...>-2672  [001] ..s2   375.947351: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=242 rc=0
           <...>-2672  [001] ..s1   375.947352: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=242 rc=0
           <...>-2672  [001] ..s2   376.035383: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=242 rc=0
           <...>-2672  [001] ..s1   376.035383: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=242 rc=0
           <...>-2672  [001] ..s2   377.563806: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6de0 len=226 rc=0
           <...>-2672  [001] ..s1   377.563807: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6de0 len=226 rc=0
           <...>-2672  [001] ..s2   377.563834: net_dev_xmit: dev=eth0 skbaddr=ffff88005cbb6be0 len=114 rc=0
           <...>-2672  [001] ..s1   377.563842: net_dev_xmit: dev=br0 skbaddr=ffff88005cbb6be0 len=114 rc=0

The stored "sshd" comm has been flushed out and we get a useless "<...>".

But by only storing comms after a trace event occurred, we can run
hackbench all day and still get the same output.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Have tracing_sched_wakeup_trace() use standard unlock_commit
Steven Rostedt [Thu, 11 Oct 2012 15:14:14 +0000 (11:14 -0400)]
tracing: Have tracing_sched_wakeup_trace() use standard unlock_commit

The functon tracing_sched_wakeup_trace() does an open coded unlock
commit and save stack. This is what the trace_nowake_buffer_unlock_commit()
is for.

Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Enable comm recording if trace_printk() is used
Steven Rostedt [Thu, 11 Oct 2012 14:15:05 +0000 (10:15 -0400)]
tracing: Enable comm recording if trace_printk() is used

If comm recording is not enabled when trace_printk() is used then
you just get this type of output:

[ adding trace_printk("hello! %d", irq); in do_IRQ ]

           <...>-2843  [001] d.h.    80.812300: do_IRQ: hello! 14
           <...>-2734  [002] d.h2    80.824664: do_IRQ: hello! 14
           <...>-2713  [003] d.h.    80.829971: do_IRQ: hello! 14
           <...>-2814  [000] d.h.    80.833026: do_IRQ: hello! 14

By enabling the comm recorder when trace_printk is enabled:

       hackbench-6715  [001] d.h.   193.233776: do_IRQ: hello! 21
            sshd-2659  [001] d.h.   193.665862: do_IRQ: hello! 21
          <idle>-0     [001] d.h1   193.665996: do_IRQ: hello! 21

Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Expand ring buffer when trace_printk() is used
Steven Rostedt [Thu, 11 Oct 2012 01:44:34 +0000 (21:44 -0400)]
tracing: Expand ring buffer when trace_printk() is used

Since tracing is not used by 99% of Linux users, even though tracing
may be configured in, it does not make sense to allocate 1.4 Megs
per CPU for the ring buffers if they are not used. Thus, on boot up
the ring buffers are set to a minimal size until something needs the
and they are expanded.

This works well for events and tracers (function, etc), but for the
asynchronous use of trace_printk() which can write to the ring buffer
at any time, does not expand the buffers.

On boot up a check is made to see if any trace_printk() is used to
see if the trace_printk() temp buffer pages should be allocated. This
same code can be used to expand the buffers as well.

Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agoring-buffer: Add a 'dropped events' counter
Slava Pestov [Fri, 15 Jul 2011 21:23:58 +0000 (14:23 -0700)]
ring-buffer: Add a 'dropped events' counter

The existing 'overrun' counter is incremented when the ring
buffer wraps around, with overflow on (the default). We wanted
a way to count requests lost from the buffer filling up with
overflow off, too. I decided to add a new counter instead
of retro-fitting the existing one because it seems like a
different statistic to count conceptually, and also because
of how the code was structured.

Link: http://lkml.kernel.org/r/1310765038-26399-1-git-send-email-slavapestov@google.com
Signed-off-by: Slava Pestov <slavapestov@google.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Change tracer's integer flags to bool
Hiraku Toyooka [Tue, 2 Oct 2012 08:27:10 +0000 (17:27 +0900)]
tracing: Change tracer's integer flags to bool

print_max and use_max_tr in struct tracer are "int" variables and
used like flags. This is wasteful, so change the type to "bool".

Link: http://lkml.kernel.org/r/20121002082710.9807.86393.stgit@falsita
Signed-off-by: Hiraku Toyooka <hiraku.toyooka.gu@hitachi.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Allow tracers to start at core initcall
Steven Rostedt [Fri, 5 Oct 2012 16:13:07 +0000 (12:13 -0400)]
tracing: Allow tracers to start at core initcall

There's times during debugging that it is helpful to see traces of early
boot functions. But the tracers are initialized at device_initcall()
which is quite late during the boot process. Setting the kernel command
line parameter ftrace=function will not show anything until the function
tracer is initialized. This prevents being able to trace functions before
device_initcall().

There's no reason that the tracers need to be initialized so late in the
boot process. Move them up to core_initcall() as they still need to come
after early_initcall() which initializes the tracing buffers.

Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agotracing: Replace strict_strto* with kstrto*
Daniel Walter [Wed, 26 Sep 2012 20:08:38 +0000 (22:08 +0200)]
tracing: Replace strict_strto* with kstrto*

 * remove old string conversions with kstrto*

Link: http://lkml.kernel.org/r/20120926200838.GC1244@0x90.at
Signed-off-by: Daniel Walter <sahne@0x90.at>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
9 years agoperf/x86: Fix sparse warnings
Peter Huewe [Mon, 29 Oct 2012 20:48:17 +0000 (21:48 +0100)]
perf/x86: Fix sparse warnings

FYI, there are new sparse warnings:

 arch/x86/kernel/cpu/perf_event.c:1356:18: sparse: symbol 'events_attr' was not declared. Should it be static?

This patch makes it static and also adds the static keyword to
fix arch/x86/kernel/cpu/perf_event.c:1344:9: warning: symbol
'events_sysfs_show' was not declared.

Signed-off-by: Peter Huewe <peterhuewe@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Cc: fengguang.wu@intel.com
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/n/tip-lerdpXlnruh0yvWs2owwuizl@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Tue, 30 Oct 2012 07:35:33 +0000 (08:35 +0100)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements, fixes and code move from Arnaldo Carvalho de Melo:

 * Initialize 'page_size' variable in the python binding, this was sent
   for perf/urgent by mistake, then when merging Ingo removed it, fixing
   the problem for perf/urgent, but when perf/urgent was merged with
   perf/core, where that initialization is needed, made the python
   binding mmap call to fail, fix it by initializing page_size again.

 * Add a browser for 'perf script' and make it available from the report
   and annotate browsers. It does filtering to find the scripts that
   handle events found in the perf.data file used. From Feng Tang

 * Move some functions from symbol.c to more appropriate files, creating
   dso.[ch] in the process, no code changes. From Jiri Olsa

 * Fix mmap error output message for when perf_mmap fails and returns
   !-EPERM, where the default for mmap_pages, INT_MAX, was causing a
   !power of 2 error message, fix from Jiri Olsa.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf python: Initialize 'page_size' variable
Arnaldo Carvalho de Melo [Tue, 16 Oct 2012 17:51:04 +0000 (14:51 -0300)]
perf python: Initialize 'page_size' variable

The commit 0c1fe6b:

 'perf tools: Have the page size value available for all tools'

Broke the python binding because the global variable 'page_size' is
initialized on the main() routine, that is not called when using
just the python binding, causing evlist.mmap() to fail because it
expects that variable to be initialized to the system's page size.

Fix it by initializing it on the binding init routine.

Reported-by: David Ahern <dsahern@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-vrvp3azmbfzexnpmkhmvtzzc@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf record: Fix mmap error output condition
Jiri Olsa [Sat, 20 Oct 2012 14:33:19 +0000 (16:33 +0200)]
perf record: Fix mmap error output condition

The mmap_pages default value is not power of 2 (UINT_MAX).

Together with perf_evlist__mmap function returning error value different
from EPERM, we get misleading error message: "--mmap_pages/-m value must
be a power of two."

Fixing this by adding extra check for UINT_MAX value for this error
condition.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1350743599-4805-12-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf header: Add is_perf_magic() func
Feng Tang [Tue, 30 Oct 2012 03:56:07 +0000 (11:56 +0800)]
perf header: Add is_perf_magic() func

With this function, other modules can basically check whether a file is
a legal perf data file by checking its first 8 bytes against all
possible perf magic numbers.

Change the function name from check_perf_magic to more meaningful
is_perf_magic as suggested by acme.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1351569369-26732-7-git-send-email-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf hists browser: Integrate script browser into main hists browser
Feng Tang [Tue, 30 Oct 2012 03:56:06 +0000 (11:56 +0800)]
perf hists browser: Integrate script browser into main hists browser

Integrate the script browser into "perf report" framework, users can use
function key 'r' or the drop down menu to list all perf scripts and
select one of them, just like they did for the annotation.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1351569369-26732-6-git-send-email-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf annotate browser: Integrate script browser into annotation browser
Feng Tang [Tue, 30 Oct 2012 03:56:05 +0000 (11:56 +0800)]
perf annotate browser: Integrate script browser into annotation browser

Integrate the script browser into annotation, users can press function
key 'r' to list all perf scripts and select one of them to run that
script, the output will be shown in a separate browser.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1351569369-26732-5-git-send-email-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf scripts browser: Add a browser for perf script
Feng Tang [Tue, 30 Oct 2012 03:56:04 +0000 (11:56 +0800)]
perf scripts browser: Add a browser for perf script

Create a script browser, so that user can check all the available
scripts for current perf data file and run them inside the main perf
report or annotation browsers, for all perf samples or for samples
belong to one thread/symbol.

Please be noted: current script browser is only for report use, and
doesn't cover the record phase, IOW it must run against one existing
perf data file.

The work flow is, users can use function key to list all the available
scripts for current perf data file in system and chose one, which will
be executed with popen("perf script -s xxx.xx",) and all the output
lines are put into one ui browser, pressing 'q' or left arrow key will
make it return to previous browser.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1351569369-26732-4-git-send-email-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf script: Add more filter to find_scripts()
Feng Tang [Tue, 30 Oct 2012 03:56:03 +0000 (11:56 +0800)]
perf script: Add more filter to find_scripts()

As suggested by Arnaldo, many scripts have their own usages and need
capture specific events or tracepoints, so only those scripts whose
target events match the events in current perf data file should be
listed in the script browser menu.

This patch will add the event match checking, by opening "xxx-record"
script to cherry pick out all events name and comparing them with
those inside the perf data file.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1351569369-26732-3-git-send-email-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Add a global variable "const char *input_name"
Feng Tang [Tue, 30 Oct 2012 03:56:02 +0000 (11:56 +0800)]
perf tools: Add a global variable "const char *input_name"

Currently many perf commands annotate/evlist/report/script/lock etc all
support "-i" option to chose a specific perf data, and all of them
create a local "input_name" to save the file name for that perf data.

Since most of these commands need it, we can add a global variable for
it, also it can some other benefits:

1. When calling script browser inside hists/annotation browser, it needs
to know the perf data file name to run that script.

2. For further feature like runtime switching to another perf data file,
this variable can also help.

Signed-off-by: Feng Tang <feng.tang@intel.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1351569369-26732-2-git-send-email-feng.tang@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Move dso_* related functions into dso object
Jiri Olsa [Sat, 27 Oct 2012 21:18:32 +0000 (23:18 +0200)]
perf tools: Move dso_* related functions into dso object

Moving dso_* related functions into dso object.

Keeping symbol loading related functions still in the symbol object as
it seems more convenient.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351372712-21104-6-git-send-email-jolsa@redhat.com
[ committer note: Use "symbol.h" instead of <symbol.h> to make it build with O= ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Move strxfrchar into string object
Jiri Olsa [Sat, 27 Oct 2012 21:18:31 +0000 (23:18 +0200)]
perf tools: Move strxfrchar into string object

Moving strxfrchar function into string object.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351372712-21104-5-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Move hex2u64 into util object
Jiri Olsa [Sat, 27 Oct 2012 21:18:30 +0000 (23:18 +0200)]
perf tools: Move hex2u64 into util object

Moving hex2u64 function into util object.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351372712-21104-4-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Move BUILD_ID_SIZE into build-id object
Jiri Olsa [Sat, 27 Oct 2012 21:18:29 +0000 (23:18 +0200)]
perf tools: Move BUILD_ID_SIZE into build-id object

Moving BUILD_ID_SIZE define into build-id object, plus include related
changes.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351372712-21104-3-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Move build_id__sprintf into build-id object
Jiri Olsa [Sat, 27 Oct 2012 21:18:28 +0000 (23:18 +0200)]
perf tools: Move build_id__sprintf into build-id object

Moving build_id__sprintf function into build-id object.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Reviewed-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Namhyung Kim <namhyung@kernel.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351372712-21104-2-git-send-email-jolsa@redhat.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Move parse_events error printing to parse_events_options
Andi Kleen [Fri, 26 Oct 2012 20:30:06 +0000 (13:30 -0700)]
perf tools: Move parse_events error printing to parse_events_options

The callers of parse_events usually have their own error handling.  Move
the fprintf for a bad event to parse_events_options, which is the only
one who should need it.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1351283415-13170-25-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Fri, 26 Oct 2012 14:52:45 +0000 (16:52 +0200)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements from Arnaldo Carvalho de Melo:

 * perf inject changes to allow showing where a task sleeps, from Andrew Vagin.

 * Makefile improvements from Namhyung Kim.

 * Add --pre and --post command hooks in 'stat', from Peter Zijlstra.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf stat: Add --pre and --post command
Peter Zijlstra [Tue, 23 Oct 2012 11:40:14 +0000 (13:40 +0200)]
perf stat: Add --pre and --post command

In order to measure kernel builds, one has to do some pre/post cleanup
work in order to do the repeat build.

So provide --pre and --post command hooks to allow doing just that.

  perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' \
-- make -s -j64 O=defconfig-build/ bzImage

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Ingo Molnar <mingo@kernel.org>
Cc: Stephane Eranian <eranian@gmail.com>
Link: http://lkml.kernel.org/r/1350992414.13456.5.camel@twins
[ committer note: Added respective entries in Documentation/perf-stat.txt ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf inject: Mark a dso if it's used
Andrew Vagin [Tue, 7 Aug 2012 12:56:05 +0000 (16:56 +0400)]
perf inject: Mark a dso if it's used

Otherwise they will be not written in an output file.

Signed-off-by: Andrew Vagin <avagin@openvz.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344344165-369636-5-git-send-email-avagin@openvz.org
[ committer note: Fixed up wrt changes made in the immediate previous patches ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf inject: Merge sched_stat_* and sched_switch events
Andrew Vagin [Tue, 7 Aug 2012 12:56:04 +0000 (16:56 +0400)]
perf inject: Merge sched_stat_* and sched_switch events

You may want to know where and how long a task is sleeping. A callchain
may be found in sched_switch and a time slice in stat_iowait, so I add
handler in perf inject for merging this events.

My code saves sched_switch event for each process and when it meets
stat_iowait, it reports the sched_switch event, because this event
contains a correct callchain. By another words it replaces all
stat_iowait events on proper sched_switch events.

I use the next sequence of commands for testing:

  perf record -e sched:sched_stat_sleep -e sched:sched_switch \
      -e sched:sched_process_exit -g -o ~/perf.data.raw \
      ~/test-program
  perf inject -v -s -i ~/perf.data.raw -o ~/perf.data
  perf report --stdio -i ~/perf.data
   100.00% foo  [kernel.kallsyms]  [k] __schedule
                |
                --- __schedule
                    schedule
                   |
                   |--79.75%-- schedule_hrtimeout_range_clock
                   |          schedule_hrtimeout_range
                   |          poll_schedule_timeout
                   |          do_select
                   |          core_sys_select
                   |          sys_select
                   |          system_call_fastpath
                   |          __select
                   |          __libc_start_main
                   |
                    --20.25%-- do_nanosleep
                              hrtimer_nanosleep
                              sys_nanosleep
                              system_call_fastpath
                              __GI___libc_nanosleep
                              __libc_start_main

 And here is test-program.c:

 #include<unistd.h>
 #include<time.h>
 #include<sys/select.h>

 int main()
 {
struct timespec ts1;
struct timeval tv1;
int i;
long s;

for (i = 0; i <  10; i++) {
ts1.tv_sec = 0;
ts1.tv_nsec = 10000000;
nanosleep(&ts1, NULL);

tv1.tv_sec = 0;
tv1.tv_usec = 40000;
select(0, NULL, NULL, NULL,&tv1);
}
return 1;
 }

Signed-off-by: Andrew Vagin <avagin@openvz.org>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344344165-369636-4-git-send-email-avagin@openvz.org
[ committer note: Made it use evsel->handler ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf inject: Work with files
Andrew Vagin [Tue, 7 Aug 2012 12:56:02 +0000 (16:56 +0400)]
perf inject: Work with files

Before this patch "perf inject" can only handle data from pipe.

I want to use "perf inject" for reworking events. Look at my following patch.

v2: add information about new options in tools/perf/Documentation/

Signed-off-by: Andrew Vagin <avagin@openvz.org>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1344344165-369636-2-git-send-email-avagin@openvz.org
[ committer note: fixed it up to cope with 5852a445ded57a002439e & f62d3f0 ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Fix LIBELF_MMAP checking
Namhyung Kim [Fri, 26 Oct 2012 08:55:52 +0000 (17:55 +0900)]
perf tools: Fix LIBELF_MMAP checking

Currently checking mmap support in libelf failed due to wrong flags.

    CHK libelf
    CHK libdw
    CHK libunwind
    CHK -DLIBELF_MMAP
/tmp/ccYJwdR0.o: In function `main':
:(.text+0x18): undefined reference to `elf_begin'
collect2: error: ld returned 1 exit status

This cannot happen since we checked the elf_begin() when checking
libelf and it succeeded.

Fix it by using a same flag with libelf checking.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Borislav Petkov <bp@amd64.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351241752-2919-5-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Always show CHK message when doing try-cc
Namhyung Kim [Fri, 26 Oct 2012 08:55:51 +0000 (17:55 +0900)]
perf tools: Always show CHK message when doing try-cc

It might be useful to see what's happening behind us rather than just
waiting few seconds during the config checking.

Also align the CHK message with other ones.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Borislav Petkov <bp@amd64.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351241752-2919-4-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Convert invocation of MAKE into SUBDIR
Namhyung Kim [Fri, 26 Oct 2012 08:55:50 +0000 (17:55 +0900)]
perf tools: Convert invocation of MAKE into SUBDIR

This will show directory change info in a consistent form.  Also it can
be converted again into David Howell's descend command.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Borislav Petkov <bp@amd64.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351241752-2919-3-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Cleanup doc related targets
Namhyung Kim [Fri, 26 Oct 2012 08:55:49 +0000 (17:55 +0900)]
perf tools: Cleanup doc related targets

Documentation targets handling rules are duplicate.  Consolidate them
with DOC_TARGETS and INSTALL_DOC_TARGETS.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Borislav Petkov <bp@amd64.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1351241752-2919-2-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agotools lib traceevent: Do not generate dependency for system header files
Namhyung Kim [Fri, 26 Oct 2012 08:55:48 +0000 (17:55 +0900)]
tools lib traceevent: Do not generate dependency for system header files

Ingo reported (again!) that 'make clean' on perf/traceevent does not
work due to some reason with system header file. Quotes Ingo:

 "Note that the old dependency related build failure thought to be
  fixed in commit 860df5833e46 is back:

   make[1]: *** No rule to make target
   `/usr/lib/gcc/x86_64-redhat-linux/4.7.0/include/stddef.h', needed by `.trace-seq.d'.  Stop.

  'make clean' itself does not work in libtraceevent:

   comet:~/tip/tools/lib/traceevent> make clean
   make: *** No rule to make target `/usr/lib/gcc/x86_64-redhat-linux/4.7.0/include/stddef.h', needed by `.trace-seq.d'.  Stop.

  So I had to clean it out manually:

   comet:~/tip/tools/lib/traceevent> git ls-files --others | xargs rm
   comet:~/tip/tools/lib/traceevent>

  and then things build fine."

Try to fix it by excluding system headers from dependency generation.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Reported-by: Ingo Molnar <mingo@kernel.org>
Cc: Borislav Petkov <bp@amd64.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/1351241752-2919-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Fri, 26 Oct 2012 08:30:49 +0000 (10:30 +0200)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core trace improvements from Arnaldo Carvalho de Melo:

 * Don't stop synthesizing threads when one vanishes, this is for
   the existing threads when we start a tool like trace.

 * Use sched:sched_stat_runtime to provide a thread summary, this
   produces the same output as the 'trace summary' subcommand of
   tglx's original "trace" tool.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf trace: Use sched:sched_stat_runtime to provide a thread summary
Arnaldo Carvalho de Melo [Wed, 17 Oct 2012 20:13:12 +0000 (17:13 -0300)]
perf trace: Use sched:sched_stat_runtime to provide a thread summary

[root@sandy ~]# perf trace --sched --duration 0.100 --pid `pidof firefox`
<SNIP>
 17079.847 ( 0.009 ms): 17643 poll(ufds: 140037623086496, nfds: 11, timeout_msecs: 0) = 0 Timeout
 17079.892 ( 0.010 ms): 17643 read(fd: 4, buf: 140038178943092, count: 4096         ) = -1 EAGAIN Resource temporarily unavailable
 17079.921 ( 0.013 ms): 17643 poll(ufds: 140037623086496, nfds: 11, timeout_msecs: 0) = 0 Timeout
 17079.949 ( 0.009 ms): 17643 read(fd: 4, buf: 140038178943092, count: 4096         ) = -1 EAGAIN Resource temporarily unavailable
^C
 _____________________________________________________________________
 __)    Summary of events    (__

              [ task - pid ]     [ events ] [ ratio ]  [ runtime ]
 _____________________________________________________________________

             firefox - 17643 :      18013   [ 72.2% ]    359.110 ms
             firefox - 17663 :         41   [  0.2% ]     21.439 ms
             firefox - 17664 :       6840   [ 27.4% ]    133.642 ms
             firefox - 17667 :         46   [  0.2% ]      0.682 ms
[root@sandy ~]#

This is equivalent to the 'perf trace summary' subcomand in the tmp.perf/trace2
branch.

Another example, setting a huge duration filter to get just a system
wide summary:

[root@sandy ~]# perf trace --duration 10000.0 --sched
^C
 _____________________________________________________________________
 __)    Summary of events    (__

              [ task - pid ]     [ events ] [ ratio ]  [ runtime ]
 _____________________________________________________________________

           scsi_eh_1 - 258   :         15   [  0.0% ]      0.133 ms
        kworker/0:1H - 322   :         13   [  0.0% ]      0.032 ms
         jbd2/dm-0-8 - 384   :          4   [  0.0% ]      0.115 ms
         flush-253:0 - 470   :          1   [  0.0% ]      0.027 ms
             firefox - 950   :       4783   [  0.1% ]     24.863 ms
             firefox - 992   :       1883   [  0.1% ]      6.808 ms
             firefox - 995   :         35   [  0.0% ]      0.111 ms
         ksoftirqd/6 - 4362  :          2   [  0.0% ]      0.005 ms
         ksoftirqd/7 - 4365  :          1   [  0.0% ]      0.007 ms
                Xorg - 4671  :        148   [  0.0% ]      0.912 ms
     gnome-settings- - 4846  :         14   [  0.0% ]      0.086 ms
     seahorse-daemon - 4847  :         14   [  0.0% ]      0.092 ms
         gnome-panel - 4875  :         46   [  0.0% ]      0.159 ms
     gnome-power-man - 4918  :         16   [  0.0% ]      0.065 ms
     gvfs-afc-volume - 4992  :         77   [  0.0% ]      0.136 ms
     gnome-screensav - 5114  :         24   [  0.0% ]      0.128 ms
               xchat - 8082  :        466   [  0.0% ]      2.019 ms
            synergyc - 8369  :        941   [  0.0% ]      3.291 ms
            synergyc - 8371  :         85   [  0.0% ]      1.817 ms
         jbd2/dm-4-8 - 9352  :          4   [  0.0% ]      0.109 ms
             rpcbind - 9786  :          3   [  0.0% ]      0.017 ms
        rtkit-daemon - 12802 :         10   [  0.0% ]      0.038 ms
        rtkit-daemon - 12803 :          8   [  0.0% ]      0.000 ms
       udisks-daemon - 13020 :         27   [  0.0% ]      0.240 ms
         kworker/7:0 - 14651 :        669   [  0.0% ]      2.616 ms
         kworker/5:1 - 16220 :          2   [  0.0% ]      0.069 ms
         kworker/4:0 - 19776 :         13   [  0.0% ]      0.176 ms
             openvpn - 20131 :        133   [  0.0% ]      0.762 ms
     plugin-containe - 20508 :      60658   [  1.7% ]    131.153 ms
        npviewer.bin - 20520 :      72208   [  2.0% ]    138.945 ms
        npviewer.bin - 20542 :         35   [  0.0% ]      0.074 ms
        npviewer.bin - 20543 :         30   [  0.0% ]      0.074 ms
        npviewer.bin - 20547 :         35   [  0.0% ]      0.092 ms
        npviewer.bin - 20552 :         35   [  0.0% ]      0.093 ms
                sshd - 20645 :         32   [  0.0% ]      0.071 ms
        npviewer.bin - 21053 :         35   [  0.0% ]      0.074 ms
        npviewer.bin - 21054 :         35   [  0.0% ]      0.097 ms
         kworker/0:2 - 21169 :        149   [  0.0% ]      1.143 ms
         kworker/3:0 - 22171 :        113   [  0.0% ]     96.892 ms
         flush-253:4 - 22410 :          1   [  0.0% ]      0.028 ms
         kworker/6:0 - 24581 :         25   [  0.0% ]      0.275 ms
         kworker/1:0 - 25572 :          4   [  0.0% ]      0.103 ms
         kworker/2:1 - 26299 :        138   [  0.0% ]      1.440 ms
         kworker/0:0 - 26325 :          1   [  0.0% ]      0.003 ms
                perf - 26330 :    3506967   [ 96.1% ]   6648.310 ms
[root@sandy ~]#

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/n/tip-mzuli0srnxyi1o029py6537x@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf trace: Count number of events for each thread and globally
Arnaldo Carvalho de Melo [Wed, 17 Oct 2012 20:09:46 +0000 (17:09 -0300)]
perf trace: Count number of events for each thread and globally

The nr_events in trace__run was local, but we will need it in other
trace methods, move it to struct trace.

We'll also need the number of events per thread, so introduce a
nr_events method for that in struct thread_trace.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ksutaz0mtejnf7e6az3ca1td@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Don't stop synthesizing threads when one vanishes
Arnaldo Carvalho de Melo [Wed, 17 Oct 2012 21:50:13 +0000 (18:50 -0300)]
perf tools: Don't stop synthesizing threads when one vanishes

The perf_event__synthesize_threads routine synthesizes all the existing
threads in the system, because we don't have any kernel facilities to
ask for PERF_RECORD_{FORK,MMAP,COMM} for existing threads.

It was returning an error as soon as one thread couldn't be synthesized,
which is a bit extreme when, for instance, a forkish workload is
running, like a kernel compile.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-i7oas1eodpoer2bx38fwyasv@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoMerge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Thu, 25 Oct 2012 07:42:03 +0000 (09:42 +0200)]
Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

 * Align the 'Ok'/'FAILED!' test results in 'perf test.

 * Support interrupted syscalls in 'trace'

 * Add an event duration column and filter in 'trace'.

 * There are references to the man pages in some tools, so try to build
   Documentation when installing, warning the user if that is not possible,
   from Borislav Petkov.

 * Give user better message if precise is not supported, from David Ahern.

 * Try to find cross-built objdump path by using the session environment
   information in the perf.data file header, from Irina Tirdea, original
   patch and idea by Namhyung Kim.

 * Diplays more output on features check for make V=1, so that one can figure
   out what is happening by looking at gcc output, etc. From Jiri Olsa.

 * Account the nr_entries in rblist properly, fix by Suzuki K. Poulose.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Try to build Documentation when installing
Borislav Petkov [Fri, 28 Sep 2012 17:47:07 +0000 (19:47 +0200)]
perf tools: Try to build Documentation when installing

There's a portion in the "perf list" output refering to the exact
specification of raw hardware events.

Since this description is in the perf-list manpage, try to build and
install the man pages, warning the user when that is not possible
due to missing packages (xmlto and asciidoc).

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/n/tip-ij71ysszkdvz3fy3wr331bke@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf trace: Print the name of a syscall when failing to read its info
Arnaldo Carvalho de Melo [Wed, 24 Oct 2012 20:44:13 +0000 (18:44 -0200)]
perf trace: Print the name of a syscall when failing to read its info

When failing to read the tracepoint event format, like currently with
sys_execve, that is not defined via SYSCALL_DEFINE macros and thus
doesn't have an entry in:

  $ ls -d /sys/kernel/debug/tracing/events/syscalls/sys_enter_*exec*
  /sys/kernel/debug/tracing/events/syscalls/sys_enter_kexec_load
  $

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/n/tip-`ranpwd
Link: http://lkml.kernel.org/n/tip-q3ak0j8b81yxylykq5wp2uwi@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Pretty print errno for some more functions
Arnaldo Carvalho de Melo [Thu, 18 Oct 2012 22:06:13 +0000 (19:06 -0300)]
perf tools: Pretty print errno for some more functions

This time: access, open and socket.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-e19dmpz8zxqo2uebxnp7ilkf@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf trace: Add duration filter
Arnaldo Carvalho de Melo [Mon, 8 Oct 2012 12:56:00 +0000 (09:56 -0300)]
perf trace: Add duration filter

Example:

[acme@sandy linux]$ perf trace --duration 0.025 usleep 1
     2.221 ( 0.958 ms): 6724 execve(arg0: 140733557168278, arg1: 140733557178768, arg2: 16134304, arg3: 140733557167840, arg4: 7955998171588342573, arg5: 6723) = -2
     3.690 ( 1.443 ms): 6724 execve(arg0: 140733557168295, arg1: 140733557178768, arg2: 16134304, arg3: 140733557167840, arg4: 7955998171588342573, arg5: 6723) = 0
     3.979 ( 0.048 ms): 6724 open(filename: 208733843841, flags: 0, mode: 1                        ) = 3
     4.071 ( 0.075 ms): 6724 open(filename: 139744419925673, flags: 0, mode: 0                     ) = 3
     4.318 ( 0.056 ms): 6724 nanosleep(rqtp: 140734030404608, rmtp: 0                              ) = 0
[acme@sandy linux]$ perf trace --duration 0.100 usleep 1
     1.143 ( 1.021 ms): 6726 execve(arg0: 140736323962279, arg1: 140736323972752, arg2: 34926752, arg3: 140736323961824, arg4: 7955998171588342573, arg5: 6725) = 0
[acme@sandy linux]$

Cherry picked from tmp.perf/trace2 branch.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/n/tip-oslw2j2958we9qf0ctra4whd@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf trace: Add an event duration column
Arnaldo Carvalho de Melo [Wed, 24 Oct 2012 19:24:47 +0000 (17:24 -0200)]
perf trace: Add an event duration column

 # perf trace usleep 1 | tail -10
     0.453 ( 0.002 ms): mmap(addr: 0, len: 4096, prot: 3, flags: 34, fd: 4294967295, off: 0   ) = -763342848
     0.456 ( 0.001 ms): mmap(addr: 0, len: 4096, prot: 3, flags: 34, fd: 4294967295, off: 0   ) = -763346944
     0.459 ( 0.001 ms): arch_prctl(option: 4098, arg2: 140126839658240, arg3: 140126839652352, arg4: 34, arg5: 4294967295) = 0
     0.473 ( 0.003 ms): mprotect(start: 208741634048, len: 16384, prot: 1                     ) = 0
     0.477 ( 0.003 ms): mprotect(start: 208735956992, len: 4096, prot: 1                      ) = 0
     0.483 ( 0.004 ms): munmap(addr: 140126839664640, len: 91882                              ) = 0
     0.540 ( 0.001 ms): brk(brk: 0                                                            ) = 31928320
     0.542 ( 0.002 ms): brk(brk: 32063488                                                     ) = 32063488
     1.456 ( 0.901 ms): nanosleep(rqtp: 140735472817168, rmtp: 0                              ) = 0
     1.462 ( 0.000 ms): exit_group(error_code: 0
 #

This also comes from the tmp.perf/trace2 branch.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/n/tip-g9akh5hjw2kvjerpo9xror6f@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf trace: Support interrupted syscalls
Arnaldo Carvalho de Melo [Sat, 6 Oct 2012 21:43:19 +0000 (18:43 -0300)]
perf trace: Support interrupted syscalls

Using the same strategies as in the tmp.perf/trace2, i.e. the 'trace'
tool implemented by tglx, just updated to the current codebase.

Example:

[root@sandy linux]# perf trace usleep 1  | tail
     2.003: mmap(addr: 0, len: 4096, prot: 3, flags: 34, fd: 4294967295, off: 0   ) = -2128396288
     2.017: mmap(addr: 0, len: 4096, prot: 3, flags: 34, fd: 4294967295, off: 0   ) = -2128400384
     2.029: arch_prctl(option: 4098, arg2: 140146949441280, arg3: 140146949435392, arg4: 34, arg5: 4294967295) = 0
     2.084: mprotect(start: 208741634048, len: 16384, prot: 1                     ) = 0
     2.098: mprotect(start: 208735956992, len: 4096, prot: 1                      ) = 0
     2.122: munmap(addr: 140146949447680, len: 91882                              ) = 0
     2.359: brk(brk: 0                                                            ) = 28987392
     2.371: brk(brk: 29122560                                                     ) = 29122560
     2.490: nanosleep(rqtp: 140735694241504, rmtp: 0                              ) = 0
     2.507: exit_group(error_code: 0
[root@sandy linux]#

For now the timestamp and duration are always on, will be selectable.

Also if multiple threads are being monitored, its tid will appear.

The ret output continues to be interpreted a la strace.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ly9ulroru4my5isn0xe9gr0m@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf test: Align the 'Ok'/'FAILED!' test results
Arnaldo Carvalho de Melo [Wed, 24 Oct 2012 17:44:41 +0000 (15:44 -0200)]
perf test: Align the 'Ok'/'FAILED!' test results

And also print 'FAILED!' in red.

Suggested-by: Ingo Molnar <mingo@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-rkisq85w24il3e2yl3nzumhu@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Give user better message if precise is not supported
David Ahern [Thu, 13 Sep 2012 20:59:15 +0000 (14:59 -0600)]
perf tools: Give user better message if precise is not supported

Platforms (e.g., VM's) without support for precise mode get a confusing
error message. e.g.,
$ perf record -e cycles:p -a -- sleep 1

  Error: sys_perf_event_open() syscall returned with 95 (Operation not
  supported).  /bin/dmesg may provide additional information.

  No hardware sampling interrupt available. No APIC? If so then you can
  boot the kernel with the "lapic" boot parameter to force-enable it.
  sleep: Terminated

which is not clear that precise mode might be the root problem. With this
patch:

$ perf record -e cycles:p -fo /tmp/perf.data -- sleep 1
  Error:
  'precise' request may not be supported. Try removing 'p' modifier
  sleep: Terminated

v2: softened message to 'may not be' supported per Robert's suggestion

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Richter <robert.richter@amd.com>
Link: http://lkml.kernel.org/r/1347569955-54626-4-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoAccount the nr_entries in rblist properly
Suzuki K. Poulose [Fri, 31 Aug 2012 07:09:18 +0000 (12:39 +0530)]
Account the nr_entries in rblist properly

The nr_entries in rblist is never decremented when an element
is deleted. Also, use rblist__remove_node to delete a node in
rblist__delete(). This would keep the nr_entries sane.

Signed-off-by: Suzuki K. Poulose <suzuki@in.ibm.com>
Acked-by: David S. Ahern <dsahern@gmail.com>
Cc: David S. Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/20120831070834.14806.87398.stgit@suzukikp.in.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Try to find cross-built objdump path
Irina Tirdea [Mon, 15 Oct 2012 23:33:38 +0000 (02:33 +0300)]
perf tools: Try to find cross-built objdump path

As we have architecture information of saved perf.data file, we can try
to find cross-built objdump path.

The triplets include support for Android (arm, x86 and mips
architectures).

Signed-off-by: Irina Tirdea <irina.tirdea@intel.com>
Originally-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung.kim@lge.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/1350344020-8071-5-git-send-email-irina.tirdea@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Remove noise in python version feature test
Arnaldo Carvalho de Melo [Tue, 9 Oct 2012 18:15:25 +0000 (15:15 -0300)]
perf tools: Remove noise in python version feature test

Now that the feature tests honours the V=1 make verbosity switch, add a
return to the main() routine in the python version test, to avoid this
distraction:

CHK python version
<stdin>: In function 'main':
<stdin>:5: warning: control reaches end of non-void function

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Namhyung Kim <namhyung@gmail.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-999no5yxlc2oqo9xjeez5zmv@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agoperf tools: Diplays more output on features check for make V=1
Jiri Olsa [Tue, 9 Oct 2012 15:50:01 +0000 (17:50 +0200)]
perf tools: Diplays more output on features check for make V=1

Adding more verbose output for compile time features checking, to ease
up debuging of feature detection failures.

Suggested-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/n/tip-fbjha6xs5soyaiek8j4142xg@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
9 years agox86: Allow tracing of functions in arch/x86/kernel/rtc.c
David Vrabel [Mon, 8 Oct 2012 12:07:30 +0000 (13:07 +0100)]
x86: Allow tracing of functions in arch/x86/kernel/rtc.c

Move native_read_tsc() to tsc.c to allow profiling to be
re-enabled for rtc.c.

Signed-off-by: David Vrabel <david.vrabel@citrix.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/1349698050-6560-1-git-send-email-david.vrabel@citrix.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf test: Add automated tests for pmu sysfs translated events
Jiri Olsa [Wed, 10 Oct 2012 12:53:18 +0000 (14:53 +0200)]
perf test: Add automated tests for pmu sysfs translated events

Add automated tests for all events found under PMU/events
directory. Tested events are in the 'cpu/event=xxx/u' format,
where 'xxx' is substituted by every event found.

The 'event=xxx' term is translated to the cpu specific term.
We only check that the event is created (not the real config
numbers) and that the modifier is properly set.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349873598-12583-9-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf tools: Add support to specify hw event as PMU event term
Jiri Olsa [Wed, 10 Oct 2012 12:53:17 +0000 (14:53 +0200)]
perf tools: Add support to specify hw event as PMU event term

Add a way to specify hw event as PMU event term like:

 'cpu/event=cpu-cycles/u'
 'cpu/event=instructions,.../u'
 'cpu/cycles,.../u'

The 'event=cpu-cycles' term is replaced/translated by the hw events
term translation, which is exposed by sysfs 'events' group attribute.

Add parser bits, the rest is already handled by the PMU alias code.

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349873598-12583-8-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf tools: Fix PMU object alias initialization
Jiri Olsa [Wed, 10 Oct 2012 12:53:16 +0000 (14:53 +0200)]
perf tools: Fix PMU object alias initialization

The pmu_lookup should return pmus that do not expose the 'events'
group attribute in sysfs. Also it should fail when any other error
during 'events' lookup is hit (pmu_aliases fails).

Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349873598-12583-7-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf/x86: Add hardware events translations for Intel P6 cpus
Jiri Olsa [Wed, 10 Oct 2012 12:53:15 +0000 (14:53 +0200)]
perf/x86: Add hardware events translations for Intel P6 cpus

Add support for Intel P6 processors to display 'events' sysfs
directory (/sys/devices/cpu/events/) with hw event translations:

  # ls /sys/devices/cpu/events/
  branch-instructions
  branch-misses
  bus-cycles
  cache-misses
  cache-references
  cpu-cycles
  instructions
  ref-cycles
  stalled-cycles-backend
  stalled-cycles-frontend

Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349873598-12583-6-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf/x86: Add hardware events translations for AMD cpus
Jiri Olsa [Wed, 10 Oct 2012 12:53:14 +0000 (14:53 +0200)]
perf/x86: Add hardware events translations for AMD cpus

Add support for AMD processors to display 'events' sysfs
directory (/sys/devices/cpu/events/) with hw event translations:

  # ls  /sys/devices/cpu/events/
  branch-instructions
  branch-misses
  bus-cycles
  cache-misses
  cache-references
  cpu-cycles
  instructions
  ref-cycles
  stalled-cycles-backend
  stalled-cycles-frontend

Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349873598-12583-5-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf/x86: Add hardware events translations for Intel cpus
Jiri Olsa [Wed, 10 Oct 2012 12:53:13 +0000 (14:53 +0200)]
perf/x86: Add hardware events translations for Intel cpus

Add support for Intel processors to display 'events' sysfs
directory (/sys/devices/cpu/events/) with hw event translations:

  # ls  /sys/devices/cpu/events/
  branch-instructions
  branch-misses
  bus-cycles
  cache-misses
  cache-references
  cpu-cycles
  instructions
  ref-cycles
  stalled-cycles-backend
  stalled-cycles-frontend

Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349873598-12583-4-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf/x86: Filter out undefined events from sysfs events attribute
Jiri Olsa [Wed, 10 Oct 2012 12:53:12 +0000 (14:53 +0200)]
perf/x86: Filter out undefined events from sysfs events attribute

The sysfs events group attribute currently shows all hw events,
including also undefined ones.

This patch filters out all undefined events out of the sysfs events
group attribute, so they don't even show up.

Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349873598-12583-3-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf/x86: Make hardware event translations available in sysfs
Jiri Olsa [Wed, 10 Oct 2012 12:53:11 +0000 (14:53 +0200)]
perf/x86: Make hardware event translations available in sysfs

Add support to display hardware events translations available
through the sysfs. Add 'events' group attribute under the sysfs
x86 PMU record with attribute/file for each hardware event.

This patch adds only backbone for PMUs to display config under
'events' directory. The specific PMU support itself will come
in next patches, however this is how the sysfs group will look
like:

  # ls  /sys/devices/cpu/events/
  branch-instructions
  branch-misses
  bus-cycles
  cache-misses
  cache-references
  cpu-cycles
  instructions
  ref-cycles
  stalled-cycles-backend
  stalled-cycles-frontend

The file - hw event ID mapping is:

  file                      hw event ID
  ---------------------------------------------------------------
  cpu-cycles                PERF_COUNT_HW_CPU_CYCLES
  instructions              PERF_COUNT_HW_INSTRUCTIONS
  cache-references          PERF_COUNT_HW_CACHE_REFERENCES
  cache-misses              PERF_COUNT_HW_CACHE_MISSES
  branch-instructions       PERF_COUNT_HW_BRANCH_INSTRUCTIONS
  branch-misses             PERF_COUNT_HW_BRANCH_MISSES
  bus-cycles                PERF_COUNT_HW_BUS_CYCLES
  stalled-cycles-frontend   PERF_COUNT_HW_STALLED_CYCLES_FRONTEND
  stalled-cycles-backend    PERF_COUNT_HW_STALLED_CYCLES_BACKEND
  ref-cycles                PERF_COUNT_HW_REF_CPU_CYCLES

Each file in the 'events' directory contains the term translation
for the symbolic hw event for the currently running cpu model.

  # cat /sys/devices/cpu/events/stalled-cycles-backend
  event=0xb1,umask=0x01,inv,cmask=0x01

Suggested-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Jiri Olsa <jolsa@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1349873598-12583-2-git-send-email-jolsa@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoMerge branch 'perf/urgent' into perf/core
Ingo Molnar [Wed, 24 Oct 2012 08:20:57 +0000 (10:20 +0200)]
Merge branch 'perf/urgent' into perf/core

Pick up v3.7-rc2 and fixes before applying more patches.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf, cpu hotplug: Use cached value of smp_processor_id()
Srivatsa S. Bhat [Tue, 16 Oct 2012 07:58:17 +0000 (13:28 +0530)]
perf, cpu hotplug: Use cached value of smp_processor_id()

The perf_cpu_notifier() macro invokes smp_processor_id()
multiple times. Optimize it by using a local variable.

Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: peterz@infradead.org
Cc: acme@ghostprotocols.net
Link: http://lkml.kernel.org/r/20121016075817.3572.76733.stgit@srivatsabhat.in.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoperf, cpu hotplug: Run CPU_STARTING notifiers with irqs disabled
Srivatsa S. Bhat [Tue, 16 Oct 2012 07:58:10 +0000 (13:28 +0530)]
perf, cpu hotplug: Run CPU_STARTING notifiers with irqs disabled

The CPU_STARTING notifiers are supposed to be run with irqs
disabled. But the perf_cpu_notifier() macro invokes them without
doing that. Fix it.

Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: peterz@infradead.org
Cc: acme@ghostprotocols.net
Link: http://lkml.kernel.org/r/20121016075809.3572.47848.stgit@srivatsabhat.in.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agox86/perf: Fix virtualization sanity check
Andre Przywara [Tue, 9 Oct 2012 15:38:35 +0000 (17:38 +0200)]
x86/perf: Fix virtualization sanity check

In check_hw_exists() we try to detect non-emulated MSR accesses
by writing an arbitrary value into one of the PMU registers
and check if it's value after a readout is still the same.
This algorithm silently assumes that the register does not contain
the magic value already, which is wrong in at least one situation.

Fix the algorithm to really do a read-modify-write cycle. This fixes
a warning under Xen under some circumstances on AMD family 10h CPUs.

The reasons in more details actually sound like a story from
Believe It or Not!:

First you need an AMD family 10h/12h CPU. These do not reset the
PERF_CTR registers on a reboot.
Now you boot bare metal Linux, which goes successfully through this
check, but leaves the magic value of 0xabcd in the register. You
don't use the performance counters, but do a reboot (warm reset).
Then you choose to boot Xen. The check will be triggered with a
recent Linux kernel as Dom0 again, trying to write 0xabcd into the
MSR. Xen silently drops the write (expected), but the subsequent read
will return the value in the register, which just happens to be the
expected magic value. Thus the test misleadingly succeeds, leaving
the kernel in the belief that the PMU is available. This will trigger
the following message:

[    0.020294] ------------[ cut here ]------------
[    0.020311] WARNING: at arch/x86/xen/enlighten.c:730 xen_apic_write+0x15/0x17()
[    0.020318] Hardware name: empty
[    0.020323] Modules linked in:
[    0.020334] Pid: 1, comm: swapper/0 Not tainted 3.3.8 #7
[    0.020340] Call Trace:
[    0.020354]  [<ffffffff81050379>] warn_slowpath_common+0x80/0x98
[    0.020369]  [<ffffffff810503a6>] warn_slowpath_null+0x15/0x17
[    0.020378]  [<ffffffff810034df>] xen_apic_write+0x15/0x17
[    0.020392]  [<ffffffff8101cb2b>] perf_events_lapic_init+0x2e/0x30
[    0.020410]  [<ffffffff81ee4dd0>] init_hw_perf_events+0x250/0x407
[    0.020419]  [<ffffffff81ee4b80>] ? check_bugs+0x2d/0x2d
[    0.020430]  [<ffffffff81002181>] do_one_initcall+0x7a/0x131
[    0.020444]  [<ffffffff81edbbf9>] kernel_init+0x91/0x15d
[    0.020456]  [<ffffffff817caaa4>] kernel_thread_helper+0x4/0x10
[    0.020471]  [<ffffffff817c347c>] ? retint_restore_args+0x5/0x6
[    0.020481]  [<ffffffff817caaa0>] ? gs_change+0x13/0x13
[    0.020500] ---[ end trace a7919e7f17c0a725 ]---

The new code will change every of the 16 low bits read from the
register and tries to write and read-back that modified number
from the MSR.

Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Avi Kivity <avi@redhat.com>
Link: http://lkml.kernel.org/r/1349797115-28346-2-git-send-email-andre.przywara@amd.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoMerge tag 'perf-urgent-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
Ingo Molnar [Wed, 24 Oct 2012 05:47:40 +0000 (07:47 +0200)]
Merge tag 'perf-urgent-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/urgent

Pull perf/urgent fixes from Arnaldo Carvalho de Melo:

 * Validate syscall id before growing syscall table in 'trace', fixing potential
   excessive memory usage.

 * Validate perf_sample.raw_data, making 'trace' more robust, avoiding some
   potential SEGFAULTs when reading tracepoint fields.

 * Fix exclude_guest parse events 'perf test's, from Jiri Olsa.

 * Do not flush maps on COMM, that is sent by the kernel when a process is
   exec'ed, but also when a process changes its name. Since we were assuming
   a COMM always meant an EXEC, we were losing track of a process maps by
   flushing its maps. Fix from Luigi Semenzato.

 * A recent patch introduced a problem by not initializing what should be
   the first kind of pager to use, 'man', instead it was being left as zero
   which means no pager. This caused 'perf subcmd --help' to produce no output.
   Fix from Namhyung Kim.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
9 years agoMerge tag 'stable/for-linus-3.7-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 24 Oct 2012 02:17:27 +0000 (05:17 +0300)]
Merge tag 'stable/for-linus-3.7-rc2-tag' of git://git./linux/kernel/git/konrad/xen

Pull xen bug-fixes from Konrad Rzeszutek Wilk:
 - Fix mysterious SIGSEGV or SIGKILL in applications due to corrupting
   of the %eip when returning from a signal handler.
 - Fix various ARM compile issues after the merge fallout.
 - Continue on making more of the Xen generic code usable by ARM
   platform.
 - Fix SR-IOV passthrough to mirror multifunction PCI devices.
 - Fix various compile warnings.
 - Remove hypercalls that don't exist anymore.

* tag 'stable/for-linus-3.7-rc2-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
  xen: dbgp: Fix warning when CONFIG_PCI is not enabled.
  xen: arm: comment on why 64-bit xen_pfn_t is safe even on 32 bit
  xen: balloon: use correct type for frame_list
  xen/x86: don't corrupt %eip when returning from a signal handler
  xen: arm: make p2m operations NOPs
  xen: balloon: don't include e820.h
  xen: grant: use xen_pfn_t type for frame_list.
  xen: events: pirq_check_eoi_map is X86 specific
  xen: XENMEM_translate_gpfn_list was remove ages ago and is unused.
  xen: sysfs: fix build warning.
  xen: sysfs: include err.h for PTR_ERR etc
  xen: xenbus: quirk uses x86 specific cpuid
  xen PV passthru: assign SR-IOV virtual functions to separate virtual slots
  xen/xenbus: Fix compile warning.
  xen/x86: remove duplicated include from enlighten.c

9 years agoalpha: separate thread-synchronous flags
Al Viro [Sat, 20 Oct 2012 14:52:23 +0000 (15:52 +0100)]
alpha: separate thread-synchronous flags

... and fix the race in updating unaligned control ones

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
9 years agoMerge tag 'kvm-3.7-2' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Wed, 24 Oct 2012 01:08:42 +0000 (04:08 +0300)]
Merge tag 'kvm-3.7-2' of git://git./virt/kvm/kvm

Pull kvm fixes from Avi Kivity:
 "KVM updates for 3.7-rc2"

* tag 'kvm-3.7-2' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM guest: exit idleness when handling KVM_PV_REASON_PAGE_NOT_PRESENT
  KVM: apic: fix LDR calculation in x2apic mode
  KVM: MMU: fix release noslot pfn

9 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 24 Oct 2012 01:07:51 +0000 (04:07 +0300)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Ingo Molnar:
 "Most of these are uprobes race fixes from Oleg, and their preparatory
  cleanups.  (It's larger than what I'd normally send for an -rc kernel,
  but they looked significant enough to not delay them.)

  There's also an oprofile fix and an uncore PMU fix."

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (22 commits)
  perf/x86: Disable uncore on virtualized CPUs
  oprofile, x86: Fix wrapping bug in op_x86_get_ctrl()
  ring-buffer: Check for uninitialized cpu buffer before resizing
  uprobes: Fix the racy uprobe->flags manipulation
  uprobes: Fix prepare_uprobe() race with itself
  uprobes: Introduce prepare_uprobe()
  uprobes: Fix handle_swbp() vs unregister() + register() race
  uprobes: Do not delete uprobe if uprobe_unregister() fails
  uprobes: Don't return success if alloc_uprobe() fails
  uprobes/x86: Only rep+nop can be emulated correctly
  uprobes: Simplify is_swbp_at_addr(), remove stale comments
  uprobes: Kill set_orig_insn()->is_swbp_at_addr()
  uprobes: Introduce copy_opcode(), kill read_opcode()
  uprobes: Kill set_swbp()->is_swbp_at_addr()
  uprobes: Restrict valid_vma(false) to skip VM_SHARED vmas
  uprobes: Change valid_vma() to demand VM_MAYEXEC rather than VM_EXEC
  uprobes: Change write_opcode() to use FOLL_FORCE
  uprobes: Move clear_thread_flag(TIF_UPROBE) to uprobe_notify_resume()
  uprobes: Kill UTASK_BP_HIT state
  uprobes: Fix UPROBE_SKIP_SSTEP checks in handle_swbp()
  ...

9 years agoMerge branch 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 24 Oct 2012 01:07:02 +0000 (04:07 +0300)]
Merge branch 'core-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull core kernel fixes from Ingo Molnar:
 "Two small fixes"

* 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  Documentation: Reflect the new location of the NMI watchdog info
  nohz: Fix idle ticks in cpu summary line of /proc/stat

9 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Linus Torvalds [Wed, 24 Oct 2012 01:05:56 +0000 (04:05 +0300)]
Merge branch 'for-linus' of git://git./linux/kernel/git/s390/linux

Pull s390 fixes from Martin Schwidefsky:
 "Among the usual minor bug fixes the more interesting patches are the
  perf counters for the latest machine, the missing select to enable
  transparent huge pages and a build fix for the UAPI rework."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390,uapi: do not use uapi/asm-generic/kvm_para.h
  s390/cache: fix data/instruction cache output
  s390: fix linker script for 31 bit builds
  s390/thp: select HAVE_ARCH_TRANSPARENT_HUGEPAGE
  s390/kdump: Use 64 bit mode for 0x10000 entry point
  perf_cpum_cf: Add support for counters available with IBM zEC12
  s390/css: stop stsch loop after cc 3
  s390/cio: use generic bitmap functions
  s390/chpid: make headers usable (again)

9 years agoMerge branch 'stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux...
Linus Torvalds [Wed, 24 Oct 2012 01:05:15 +0000 (04:05 +0300)]
Merge branch 'stable' of git://git./linux/kernel/git/cmetcalf/linux-tile

Pull tile fixes from Chris Metcalf:
 "This fixes one issue with compiler flags that can cause modules not to
  load, and cleans up some warnings with ELF_R_xxx defines."

* 'stable' of git://git.kernel.org/pub/scm/linux/kernel/git/cmetcalf/linux-tile:
  arch/tile: avoid build warnings from duplicate ELF_R_xxx #defines
  arch/tile: avoid generating .eh_frame information in modules

9 years agoMerge tag 'please-pull-uapi-fix' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Wed, 24 Oct 2012 01:03:21 +0000 (04:03 +0300)]
Merge tag 'please-pull-uapi-fix' of git://git./linux/kernel/git/aegl/linux

Pull ia64 fix from Tony Luck:
 "Fix from dhowells for UAPI fallout"

* tag 'please-pull-uapi-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux:
  UAPI: Make arch/ia64/include/asm/kvm_para.h generic

9 years agoarch/tile: avoid build warnings from duplicate ELF_R_xxx #defines
Chris Metcalf [Fri, 19 Oct 2012 20:29:43 +0000 (16:29 -0400)]
arch/tile: avoid build warnings from duplicate ELF_R_xxx #defines

These are now provided in <asm-generic/module.h>, so clean up warnings
by not re-defining them in module.c.

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
9 years agoarch/tile: avoid generating .eh_frame information in modules
Chris Metcalf [Fri, 19 Oct 2012 15:43:11 +0000 (11:43 -0400)]
arch/tile: avoid generating .eh_frame information in modules

The tile tool chain uses the .eh_frame information for backtracing.
The vmlinux build drops any .eh_frame sections at link time, but when
present in kernel modules, it causes a module load failure due to the
presence of unsupported pc-relative relocations.  When compiling to
use compiler feedback support, the compiler by default omits .eh_frame
information, so we don't see this problem.  But when not using feedback,
we need to explicitly suppress the .eh_frame.

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Cc: stable@vger.kernel.org
9 years agoMerge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Tue, 23 Oct 2012 05:51:07 +0000 (08:51 +0300)]
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "Fixes for intel and nouveau mainly.

   - intel: disable HSW by default, sdvo fixes, link train regression
     fix
   - nouveau: acpi rom loading regression fix, with a few other fixes
     from the rework
   -core: just other minor fixes and race fixes for ttm."

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux: (24 commits)
  drm/ttm: Fix a theoretical race in ttm_bo_cleanup_refs()
  drm/ttm: Fix a theoretical race
  drm: platform: Don't initialize driver-private data
  drm/debugfs: remove redundant info from gem_names
  drm: fb: cma: Fail gracefully on allocation failure
  drm: fb: cma: Fix typo in debug message
  drm/nouveau/clock: fix missing pll type/addr when matching default entry
  drm/nouveau/fb: fix reporting of memory type on GF8+ IGPs
  drm/nv41/vm: don't init hw pciegart on boards with agp bridge
  drm/nouveau/bios: fetch full 4KiB block to determine ACPI ROM image size
  drm/nouveau: validate vbios size
  drm/nouveau: warn when trying to free mm which is still in use
  drm/nouveau: fix nouveau_mm/nouveau_mm_node leak
  drm/nouveau/bios: improve error handling when reading the vbios from ACPI
  drm/nouveau: handle same-fb page flips
  drm/i915: Initialize obj->pages before use by i915_gem_object_do_bit17_swizzle()
  drm/i915: Add no-lvds quirk for Supermicro X7SPA-H
  drm/i915: Insert i915_preliminary_hw_support variable.
  drm/i915: shut up spurious WARN in the gtt fault handler
  Revert "drm/i915: Try harder to complete DP training pattern 1"
  ...

9 years agoMerge tag 'jfs-3.7-2' of git://github.com/kleikamp/linux-shaggy
Linus Torvalds [Tue, 23 Oct 2012 05:49:34 +0000 (08:49 +0300)]
Merge tag 'jfs-3.7-2' of git://github.com/kleikamp/linux-shaggy

Pull jfs fix from Dave Kleikamp:
 "Bug fix: Fix FITRIM argument handling"

* tag 'jfs-3.7-2' of git://github.com/kleikamp/linux-shaggy:
  jfs: Fix FITRIM argument handling

9 years agoMerge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso...
Linus Torvalds [Tue, 23 Oct 2012 05:48:26 +0000 (08:48 +0300)]
Merge tag 'ext4_for_linus' of git://git./linux/kernel/git/tytso/ext4

Pull ext4 fixes from Ted Ts'o:
 "Various bug fixes for ext4.  The most serious of them fixes a security
  bug (CVE-2012-4508) which leads to stale data exposure when we have
  fallocate racing against writes to files undergoing delayed
  allocation.  We also have two fixes for the metadata checksum feature,
  the most serious of which can cause the superblock to have a invalid
  checksum after a power failure."

* tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: Avoid underflow in ext4_trim_fs()
  ext4: Checksum the block bitmap properly with bigalloc enabled
  ext4: fix undefined bit shift result in ext4_fill_flex_info
  ext4: fix metadata checksum calculation for the superblock
  ext4: race-condition protection for ext4_convert_unwritten_extents_endio
  ext4: serialize fallocate with ext4_convert_unwritten_extents

9 years agoMerge tag 'nfs-for-3.7-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Tue, 23 Oct 2012 05:47:38 +0000 (08:47 +0300)]
Merge tag 'nfs-for-3.7-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

Pull NFS client bugfixes from Trond Myklebust:
 - Do not call pnfs_return_layout() from an rpciod context
 - nfs4_ds_disconnect can cause Oopses.  Kill it...
 - Fix the return value for nfs_callback_start_svc
 - Fix a number of compile warnings

* tag 'nfs-for-3.7-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  NFSv4: Fix the return value for nfs_callback_start_svc
  NFSv4.1: Declare osd_pri_2_pnfs_err(), objio_init_read/write to be static
  NFSv4: fs/nfs/nfs4getroot.c needs to include "internal.h"
  NFSv4.1: Use kcalloc() to allocate zeroed arrays instead of kzalloc()
  NFSv4.1: Do not call pnfs_return_layout() from an rpciod context
  NFSv4.1: Kill nfs4_ds_disconnect()

9 years agoMerge tag 'regmap-fix-mmio' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
Linus Torvalds [Tue, 23 Oct 2012 05:39:38 +0000 (08:39 +0300)]
Merge tag 'regmap-fix-mmio' of git://git./linux/kernel/git/broonie/regmap

Pull regmap fix from Mark Brown:
 "regmap: Fix for dependencies for MMIO

  Trivial dependency issue, not noticed before as the only user of MMIO
  also needs I2C."

* tag 'regmap-fix-mmio' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie/regmap:
  regmap: select REGMAP if REGMAP_MMIO and REGMAP_IRQ enabled

9 years agodrm/ttm: Fix a theoretical race in ttm_bo_cleanup_refs()
Thomas Hellstrom [Mon, 22 Oct 2012 12:51:26 +0000 (12:51 +0000)]
drm/ttm: Fix a theoretical race in ttm_bo_cleanup_refs()

In theory, that function could release the lru lock between
checking for bo on ddestroy list and a successful reserve if the bo
was already reserved, and the function was called with waiting reserves
allowed.
However, all current reservers of a bo on the ddestroy list would
atomically take the bo off the list after a successful reserve so this
race should not have been hit, so no need to backport for stable.

This patch also fixes a case found by Maarten Lankhorst where
ttm_mem_evict_first called with no_wait_gpu would incorrectly
spin waiting for bo idle if trying to evict a busy buffer that
also sits on the ddestroy list.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
9 years agodrm/ttm: Fix a theoretical race
Thomas Hellstrom [Mon, 22 Oct 2012 12:51:25 +0000 (12:51 +0000)]
drm/ttm: Fix a theoretical race

The ttm_mem_evict_first function could theoretically drop the
lru lock without retrying if a reservation from off the LRU list
ended up waiting.
However, since currently there are no users that could cause a wait
in that situation so this is not suitable for stable

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>