Skip to content

L
linux
Switch branch/tag
  1. 23 Oct, 2013 14 commits
  3. 21 Oct, 2013 18 commits
  5. 18 Oct, 2013 1 commit
  7. 17 Oct, 2013 3 commits
    • Arnaldo Carvalho de Melo's avatar
      perf trace: Improve messages related to /proc/sys/kernel/perf_event_paranoid · a8f23d8f
      Arnaldo Carvalho de Melo authored 
      kernel/events/core.c has:

  /*
   * perf event paranoia level:
   *  -1 - not paranoid at all
   *   0 - disallow raw tracepoint access for unpriv
   *   1 - disallow cpu events for unpriv
   *   2 - disallow kernel profiling for unpriv
   */
  int sysctl_perf_event_paranoid __read_mostly = 1;

So, with the default being 1, a non-root user can trace his stuff:

  [acme@zoo ~]$ cat /proc/sys/kernel/perf_event_paranoid
  1
  [acme@zoo ~]$ yes > /dev/null &
  [1] 15338
  [acme@zoo ~]$ trace -p 15338 | head -5
       0.005 ( 0.005 ms): write(fd: 1</dev/null>, buf: 0x7fe6db765000, count: 4096 ) = 4096
       0.045 ( 0.001 ms): write(fd: 1</dev/null>, buf: 0x7fe6db765000, count: 4096 ) = 4096
       0.085 ( 0.001 ms): write(fd: 1</dev/null>, buf: 0x7fe6db765000, count: 4096 ) = 4096
       0.125 ( 0.001 ms): write(fd: 1</dev/null>, buf: 0x7fe6db765000, count: 4096 ) = 4096
       0.165 ( 0.001 ms): write(fd: 1</dev/null>, buf: 0x7fe6db765000, count: 4096 ) = 4096
  [acme@zoo ~]$
  [acme@zoo ~]$ trace --duration 1 sleep 1
    1002.148 (1001.218 ms): nanosleep(rqtp: 0x7fff46c79250                           ) = 0
  [acme@zoo ~]$
  [acme@zoo ~]$ trace -- usleep 1 | tail -5
       0.905 ( 0.002 ms): brk(                                                     ) = 0x1c82000
       0.910 ( 0.003 ms): brk(brk: 0x1ca3000                                       ) = 0x1ca3000
       0.913 ( 0.001 ms): brk(                                                     ) = 0x1ca3000
       0.990 ( 0.059 ms): nanosleep(rqtp: 0x7fffe31a3280                           ) = 0
       0.995 ( 0.000 ms): exit_group(
  [acme@zoo ~]$

But can't do system wide tracing:

  [acme@zoo ~]$ trace
  Error:	Operation not permitted.
  Hint:	Check /proc/sys/kernel/perf_event_paranoid setting.
  Hint:	For system wide tracing it needs to be set to -1.
  Hint:	The current value is 1.
  [acme@zoo ~]$

  [acme@zoo ~]$ trace --cpu 0
  Error:	Operation not permitted.
  Hint:	Check /proc/sys/kernel/perf_event_paranoid setting.
  Hint:	For system wide tracing it needs to be set to -1.
  Hint:	The current value is 1.
  [acme@zoo ~]$

If the paranoid level is >= 2, i.e. turn this perf stuff off for !root users:

  [acme@zoo ~]$ sudo sh -c 'echo 2 > /proc/sys/kernel/perf_event_paranoid'
  [acme@zoo ~]$ cat /proc/sys/kernel/perf_event_paranoid
  2
  [acme@zoo ~]$
  [acme@zoo ~]$ trace usleep 1
  Error:	Permission denied.
  Hint:	Check /proc/sys/kernel/perf_event_paranoid setting.
  Hint:	For your workloads it needs to be <= 1
  Hint:	For system wide tracing it needs to be set to -1.
  Hint:	The current value is 2.
  [acme@zoo ~]$
  [acme@zoo ~]$ trace
  Error:	Permission denied.
  Hint:	Check /proc/sys/kernel/perf_event_paranoid setting.
  Hint:	For your workloads it needs to be <= 1
  Hint:	For system wide tracing it needs to be set to -1.
  Hint:	The current value is 2.
  [acme@zoo ~]$
  [acme@zoo ~]$ trace --cpu 1
  Error:	Permission denied.
  Hint:	Check /proc/sys/kernel/perf_event_paranoid setting.
  Hint:	For your workloads it needs to be <= 1
  Hint:	For system wide tracing it needs to be set to -1.
  Hint:	The current value is 2.
  [acme@zoo ~]$

If the user manages to get what he/she wants, convincing root not
to be paranoid at all...

  [root@zoo ~]# echo -1 > /proc/sys/kernel/perf_event_paranoid
  [root@zoo ~]# cat /proc/sys/kernel/perf_event_paranoid
  -1
  [root@zoo ~]#

  [acme@zoo ~]$ ps -eo user,pid,comm | grep Xorg
  root       729 Xorg
  [acme@zoo ~]$
  [acme@zoo ~]$ trace -a --duration 0.001 -e \!select,ioctl,writev | grep Xorg  | head -5
      23.143 ( 0.003 ms): Xorg/729 setitimer(which: REAL, value: 0x7fffaadf16e0 ) = 0
      23.152 ( 0.004 ms): Xorg/729 read(fd: 31, buf: 0x2544af03, count: 4096     ) = 8
      23.161 ( 0.002 ms): Xorg/729 read(fd: 31, buf: 0x2544af03, count: 4096     ) = -1 EAGAIN Resource temporarily unavailable
      23.175 ( 0.002 ms): Xorg/729 setitimer(which: REAL, value: 0x7fffaadf16e0 ) = 0
      23.235 ( 0.002 ms): Xorg/729 setitimer(which: REAL, value: 0x7fffaadf16e0 ) = 0
  [acme@zoo ~]$

Cc: Adrian Hunter <adrian.hunter@intel.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: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-di28olfwd28rvkox7v3hqhu1@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      a8f23d8f
    • Arnaldo Carvalho de Melo's avatar
      perf tools: Introduce filename__read_int helper · 97a07f10
      Arnaldo Carvalho de Melo authored 
      Just opens a file and calls atoi() in at most its first 64 bytes.

To read things like /proc/sys/kernel/perf_event_paranoid.

Cc: Adrian Hunter <adrian.hunter@intel.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: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-669q04c5tou5pnt8jtiz6y2r@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      97a07f10
    • Arnaldo Carvalho de Melo's avatar
      perf evlist: Introduce perf_evlist__strerror_tp method · 6ef068cb
      Arnaldo Carvalho de Melo authored 
      Out of 'perf trace', should be used by other tools that uses
tracepoints.

Cc: Adrian Hunter <adrian.hunter@intel.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: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ramkumar Ramachandra <artagnon@gmail.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-lyvtxhchz4ga8fwht15x8wou@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      6ef068cb
  9. 16 Oct, 2013 4 commits
    • Arnaldo Carvalho de Melo's avatar
      perf trace: Improve event processing exit · ba209f85
      Arnaldo Carvalho de Melo authored 
      We need to differentiate SIGCHLD from SIGINT, the later should cause as
immediate as possible exit, while the former should wait to process the
events that may be perceived in the ring buffer after the SIGCHLD is
handled.

Cc: Adrian Hunter <adrian.hunter@intel.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: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-vf6n57ewm3mjy2sz6r491hus@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      ba209f85
    • Arnaldo Carvalho de Melo's avatar
      perf trace: Use vfs_getname hook if available · c522739d
      Arnaldo Carvalho de Melo authored 
      Initially it tries to find a probe:vfs_getname that should be setup
with:

 perf probe 'vfs_getname=getname_flags:65 pathname=result->name:string'

or with slight changes to cope with code flux in the getname_flags code.

In the future, if a "vfs:getname" tracepoint becomes available, then it
will be preferred.

This is not strictly required and more expensive method of reading the
/proc/pid/fd/ symlink will be used when the fd->path array entry is not
populated by a previous vfs_getname + open syscall ret sequence.

As with any other 'perf probe' probe the setup must be done just once
and the probe will be left inactive, waiting for users, be it 'perf
trace' of any other tool.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Mike Galbraith <efault@gmx.de>
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-ujg8se8glq5izmu8cdkq15po@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      c522739d
    • Arnaldo Carvalho de Melo's avatar
      perf trace: Split fd -> pathname array handling · 97119f37
      Arnaldo Carvalho de Melo authored 
      So that the part that grows the array as needed is untied from the code
that reads the /proc/pid/fd symlink and can be used for the vfs_getname
hook that will set the fd -> path translation too, when available.

Cc: Adrian Hunter <adrian.hunter@intel.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: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ydo5rumyv9hdc1vsfmqamugs@git.kernel.orgSigned-off-by: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
      97119f37
    • Peter Zijlstra's avatar
      perf/x86: Optimize intel_pmu_pebs_fixup_ip() · 9536c8d2
      Peter Zijlstra authored 
      There's been reports of high NMI handler overhead, highlighted by
such kernel messages:

  [ 3697.380195] perf samples too long (10009 > 10000), lowering kernel.perf_event_max_sample_rate to 13000
  [ 3697.389509] INFO: NMI handler (perf_event_nmi_handler) took too long to run: 9.331 msecs

Don Zickus analyzed the source of the overhead and reported:

 > While there are a few places that are causing latencies, for now I focused on
 > the longest one first.  It seems to be 'copy_user_from_nmi'
 >
 > intel_pmu_handle_irq ->
 >	intel_pmu_drain_pebs_nhm ->
 >		__intel_pmu_drain_pebs_nhm ->
 >			__intel_pmu_pebs_event ->
 >				intel_pmu_pebs_fixup_ip ->
 >					copy_from_user_nmi
 >
 > In intel_pmu_pebs_fixup_ip(), if the while-loop goes over 50, the sum of
 > all the copy_from_user_nmi latencies seems to go over 1,000,000 cycles
 > (there are some cases where only 10 iterations are needed to go that high
 > too, but in generall over 50 or so).  At this point copy_user_from_nmi
 > seems to account for over 90% of the nmi latency.

The solution to that is to avoid having to call copy_from_user_nmi() for
every instruction.

Since we already limit the max basic block size, we can easily
pre-allocate a piece of memory to copy the entire thing into in one
go.

Don reported this test result:

 > Your patch made a huge difference in improvement.  The
 > copy_from_user_nmi() no longer hits the million of cycles.  I still
 > have a batch of 100,000-300,000 cycles.  My longest NMI paths used
 > to be dominated by copy_from_user_nmi, now it is not (I have to dig
 > up the new hot path).
Reported-and-tested-by: default avatarDon Zickus <dzickus@redhat.com>
Cc: jmario@redhat.com
Cc: acme@infradead.org
Cc: dave.hansen@linux.intel.com
Cc: eranian@google.com
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: default avatarPeter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20131016105755.GX10651@twins.programming.kicks-ass.netSigned-off-by: default avatarIngo Molnar <mingo@kernel.org>
      9536c8d2
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7