safe/jmp/linux-2.6
14 years agoperf_counter: Turn off by default
Ingo Molnar [Thu, 11 Jun 2009 14:13:24 +0000 (16:13 +0200)]
perf_counter: Turn off by default

Perfcounters were enabled by default to help testing - but now that we
are submitting it upstream, make it default-disabled.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Add counter->id to the throttle event
Peter Zijlstra [Thu, 11 Jun 2009 12:57:55 +0000 (14:57 +0200)]
perf_counter: Add counter->id to the throttle event

So as to be able to distuinguish between multiple counters.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Better align code
Ingo Molnar [Thu, 11 Jun 2009 12:44:26 +0000 (14:44 +0200)]
perf_counter: Better align code

Whitespace and comment bits. Also update copyrights.

[ Impact: cleanup ]

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>

14 years agoperf_counter: Rename L2 to LL cache
Peter Zijlstra [Thu, 11 Jun 2009 12:19:11 +0000 (14:19 +0200)]
perf_counter: Rename L2 to LL cache

The top (fastest) and last level (biggest) caches are the most
interesting ones, performance wise.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
[ Fixed the Nehalem LL table to LLC Reference/Miss events ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Standardize event names
Peter Zijlstra [Thu, 11 Jun 2009 12:06:28 +0000 (14:06 +0200)]
perf_counter: Standardize event names

Pure renames only, to PERF_COUNT_HW_* and PERF_COUNT_SW_*.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Rename enums
Peter Zijlstra [Thu, 11 Jun 2009 11:19:29 +0000 (13:19 +0200)]
perf_counter: Rename enums

Rename the perf enums to be in the 'perf_' namespace and strictly
enumerate the ABI bits.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Clean up u64 usage
Ingo Molnar [Thu, 11 Jun 2009 12:16:15 +0000 (14:16 +0200)]
perf_counter tools: Clean up u64 usage

A build error slipped in:

 builtin-report.c: In function ‘hist_entry__fprintf’:
 builtin-report.c:711: error: format ‘%12d’ expects type ‘int’, but argument 3 has type ‘uint64_t’

Because we got a bit sloppy with those types. uint64_t really sucks,
because there's no printf format for it. So standardize on __u64
instead - for all types that go to or come from the ABI (which is __u64),
or for values that need to be large enough even on 32-bit.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Rename perf_counter_limit sysctl
Peter Zijlstra [Thu, 11 Jun 2009 09:25:05 +0000 (11:25 +0200)]
perf_counter: Rename perf_counter_limit sysctl

Rename perf_counter_limit to perf_counter_max_sample_rate and
prohibit creation of counters with a known higher sample
frequency.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: More paranoia settings
Peter Zijlstra [Thu, 11 Jun 2009 09:18:36 +0000 (11:18 +0200)]
perf_counter: More paranoia settings

Rename the perf_counter_priv knob to perf_counter_paranoia (because
priv can be read as private, as opposed to privileged) and provide
one more level:

 0 - permissive
 1 - restrict cpu counters to privilidged contexts
 2 - restrict kernel-mode code counting and profiling

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: powerpc: Implement generalized cache events for POWER processors
Paul Mackerras [Thu, 11 Jun 2009 04:55:42 +0000 (14:55 +1000)]
perf_counter: powerpc: Implement generalized cache events for POWER processors

This adds tables of event codes for the generalized cache events for
all the currently supported powerpc processors: POWER{4,5,5+,6,7} and
PPC970*, plus powerpc-specific code to use these tables when a
generalized cache event is requested.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <18992.36430.933526.742969@drongo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counters: powerpc: Add support for POWER7 processors
Paul Mackerras [Thu, 11 Jun 2009 04:54:01 +0000 (14:54 +1000)]
perf_counters: powerpc: Add support for POWER7 processors

This adds the back-end for the PMU on POWER7 processors.  POWER7
has 4 fully-programmable counters and two fixed-function counters
(which do respect the freeze conditions, can generate interrupts,
and are writable, unlike PMC5/6 on POWER5+/6).

Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <18992.36329.189378.17992@drongo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Accurate period data
Peter Zijlstra [Wed, 10 Jun 2009 19:34:59 +0000 (21:34 +0200)]
perf_counter: Accurate period data

We currently log hw.sample_period for PERF_SAMPLE_PERIOD, however this is
incorrect. When we adjust the period, it will only take effect the next
cycle but report it for the current cycle. So when we adjust the period
for every cycle, we're always wrong.

Solve this by keeping track of the last_period.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Introduce struct for sample data
Peter Zijlstra [Wed, 10 Jun 2009 19:02:22 +0000 (21:02 +0200)]
perf_counter: Introduce struct for sample data

For easy extension of the sample data, put it in a structure.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Normalize data using per sample period data
Peter Zijlstra [Wed, 10 Jun 2009 19:45:22 +0000 (21:45 +0200)]
perf_counter tools: Normalize data using per sample period data

When we use variable period sampling, add the period to the sample
data and use that to normalize the samples.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Annotate exit ctx recursion
Peter Zijlstra [Wed, 10 Jun 2009 20:53:37 +0000 (22:53 +0200)]
perf_counter: Annotate exit ctx recursion

Ever since Paul fixed it to unclone the context before taking the
ctx->lock this became a false positive, annotate it away.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Propagate signals properly
Peter Zijlstra [Wed, 10 Jun 2009 13:55:59 +0000 (15:55 +0200)]
perf_counter tools: Propagate signals properly

Currently report and stat catch SIGINT (and others) without altering
their exit state. This means that things like:

   while :; do perf stat ./foo ; done

Loops become hard-to-interrupt, because bash never sees perf terminate
due to interruption. Fix this.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Small frequency related fixes
Peter Zijlstra [Wed, 10 Jun 2009 13:03:06 +0000 (15:03 +0200)]
perf_counter tools: Small frequency related fixes

Create the counter in a disabled state and only enable it after we
mmap() the buffer, this allows us to see the first few samples (and
observe the frequency ramp).

Furthermore, print the period in the verbose report.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: More aggressive frequency adjustment
Peter Zijlstra [Wed, 10 Jun 2009 11:40:57 +0000 (13:40 +0200)]
perf_counter: More aggressive frequency adjustment

Also employ the overflow handler to adjust the frequency, this results
in a stable frequency in about 40~50 samples, instead of that many ticks.

This also means we can start sampling at a sample period of 1 without
running head-first into the throttle.

It relies on sched_clock() to accurately measure the time difference
between the overflow NMIs.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter/x86: Fix the model number of Intel Core2 processors
Yong Wang [Wed, 10 Jun 2009 09:06:12 +0000 (17:06 +0800)]
perf_counter/x86: Fix the model number of Intel Core2 processors

Fix the model number of Intel Core2 processors according to the
documentation: Intel Processor Identification with the CPUID
Instruction: http://www.intel.com/support/processors/sb/cs-009861.htm

Signed-off-by: Yong Wang <yong.y.wang@intel.com>
Also-Reported-by: Arnd Bergmann <arnd@arndb.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20090610090612.GA26580@ywang-moblin2.bj.intel.com>
[ Added two more model numbers suggested by Arnd Bergmann ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter, x86: Correct some event and umask values for Intel processors
Yong Wang [Tue, 9 Jun 2009 13:15:53 +0000 (21:15 +0800)]
perf_counter, x86: Correct some event and umask values for Intel processors

Correct some event and UMASK values according to Intel SDM,
in the Nehalem and Atom tables.

Signed-off-by: Yong Wang <yong.y.wang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20090609131553.GA12489@ywang-moblin2.bj.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Standardize color printing
Ingo Molnar [Mon, 8 Jun 2009 21:15:28 +0000 (23:15 +0200)]
perf_counter tools: Standardize color printing

The rule is:

 - high overhead: red
 -  mid overhead: green
 -  low overhead: normal (white/black)

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Add support for profiling JIT generated code
Pekka Enberg [Mon, 8 Jun 2009 18:12:48 +0000 (21:12 +0300)]
perf report: Add support for profiling JIT generated code

This patch adds support for profiling JIT generated code to 'perf
report'. A JIT compiler is required to generate a "/tmp/perf-$PID.map"
symbols map that is parsed when looking and displaying symbols.

Thanks to Peter Zijlstra for his help with this patch!

Example "perf report" output with the Jato JIT:

 #
 # (40311 samples)
 #
 # Overhead           Command  Shared Object              Symbol
 # ........  ................  .........................  ......
 #
     97.80%              jato  /tmp/perf-11915.map        [.] Fibonacci.fib(I)I
      0.56%              jato  00000000b7fa023b           0x000000b7fa023b
      0.45%              jato  /tmp/perf-11915.map        [.] Fibonacci.main([Ljava/lang/String;)V
      0.38%              jato  [kernel]                   [k] get_page_from_freelist
      0.06%              jato  [kernel]                   [k] kunmap_atomic
      0.05%              jato  ./jato                     [.] utf8Hash
      0.04%              jato  ./jato                     [.] executeJava
      0.04%              jato  ./jato                     [.] defineClass

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: a.p.zijlstra@chello.nl
Cc: acme@redhat.com
LKML-Reference: <Pine.LNX.4.64.0906082111590.12407@melkki.cs.Helsinki.FI>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Add mmap event hooks to mprotect()
Peter Zijlstra [Mon, 8 Jun 2009 18:11:57 +0000 (21:11 +0300)]
perf_counter: Add mmap event hooks to mprotect()

Some JIT compilers allocate memory for generated code with
posix_memalign() + mprotect() so we need to hook into mprotect()
to make sure 'perf' is aware that we're executing code in
anonymous memory.

[ penberg@cs.helsinki.fi: move the hook to sys_mprotect() ]
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
LKML-Reference: <Pine.LNX.4.64.0906082111030.12407@melkki.cs.Helsinki.FI>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter, x86: Clean up hw_cache_event ids copies
Thomas Gleixner [Mon, 8 Jun 2009 17:10:25 +0000 (19:10 +0200)]
perf_counter, x86: Clean up hw_cache_event ids copies

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter, x86: Implement generalized cache event types, add AMD support
Thomas Gleixner [Mon, 8 Jun 2009 20:33:10 +0000 (22:33 +0200)]
perf_counter, x86: Implement generalized cache event types, add AMD support

Fill in amd_hw_cache_event_id[] with the AMD CPU specific events,
for family 0x0f, 0x10 and 0x11.

There's apparently no distinction between load and store events, so
we only fill in the load events.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Clean up x86 boot messages
Ingo Molnar [Fri, 29 May 2009 09:25:09 +0000 (11:25 +0200)]
perf_counter: Clean up x86 boot messages

Standardize and tidy up all the messages we print during
perfcounter initialization.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter, x86: Implement generalized cache event types, add Atom support
Thomas Gleixner [Mon, 8 Jun 2009 07:30:41 +0000 (09:30 +0200)]
perf_counter, x86: Implement generalized cache event types, add Atom support

Fill in core2_hw_cache_event_id[] with the Atom model specific events.

The events can be used in all the tools via the -e (--event) parameter,
for example "-e l1-misses" or -"-e l2-accesses" or "-e l2-write-misses".

( Note: these are straight from the Intel manuals - not tested yet.)

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter, x86: Implement generalized cache event types, add Core2 support
Thomas Gleixner [Mon, 8 Jun 2009 05:42:04 +0000 (07:42 +0200)]
perf_counter, x86: Implement generalized cache event types, add Core2 support

Fill in core2_hw_cache_event_id[] with the Core2 model specific events.

The events can be used in all the tools via the -e (--event) parameter,
for example "-e l1-misses" or -"-e l2-accesses" or "-e l2-write-misses".

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf stat: Print out instructins/cycle metric
Ingo Molnar [Sun, 7 Jun 2009 16:14:46 +0000 (18:14 +0200)]
perf stat: Print out instructins/cycle metric

Before:

     7549326754  cycles               #    3201.811 M/sec
    10007594937  instructions         #    4244.408 M/sec

After:

     7542051194  cycles               #    3201.996 M/sec
    10007743852  instructions         #    4248.811 M/sec # 1.327 per cycle

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Print more expressive message in case of file open error
Ingo Molnar [Sun, 7 Jun 2009 15:58:23 +0000 (17:58 +0200)]
perf report: Print more expressive message in case of file open error

Before:

 $ perf report
 failed to open file: No such file or directory

After:

 $ perf report
  failed to open file: perf.data  (try 'perf record' first)

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Handle kernels with !CONFIG_PERF_COUNTER
Ingo Molnar [Sun, 7 Jun 2009 15:46:24 +0000 (17:46 +0200)]
perf_counter tools: Handle kernels with !CONFIG_PERF_COUNTER

If perf is run on a !CONFIG_PERF_COUNTER kernel right now it
bails out with no messages or with confusing messages.

Standardize this case some more and explain the situation.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf record: Fall back to cpu-clock-ticks if no PMU
Ingo Molnar [Sun, 7 Jun 2009 15:39:02 +0000 (17:39 +0200)]
perf record: Fall back to cpu-clock-ticks if no PMU

On architectures/CPUs without PMU support but with perfcounters
enabled 'perf record' currently fails because it cannot create a
cycle based hw-perfcounter.

Fall back to the cpu-clock-tick sw-perfcounter in this case, which
is hrtimer based and will always work (as long as perfcounters
are enabled).

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf top: Fall back to cpu-clock-tick hrtimer sampling if no cycle counter available
Ingo Molnar [Sun, 7 Jun 2009 15:31:52 +0000 (17:31 +0200)]
perf top: Fall back to cpu-clock-tick hrtimer sampling if no cycle counter available

On architectures/CPUs without PMU support but with perfcounters
enabled 'perf top' currently fails because it cannot create a
cycle based hw-perfcounter.

Fall back to the cpu-clock-tick sw-perfcounter in this case, which
is hrtimer based and will always work (as long as perfcounters
is enabled).

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf stat: Continue even on counter creation error
Ingo Molnar [Sun, 7 Jun 2009 15:06:46 +0000 (17:06 +0200)]
perf stat: Continue even on counter creation error

Before:

 $ perf stat ~/hackbench 5

 error: syscall returned with -1 (No such device)

After:

 $ perf stat ~/hackbench 5
 Time: 1.640

 Performance counter stats for '/home/mingo/hackbench 5':

    6524.570382  task-clock-ticks     #       3.838 CPU utilization factor
          35704  context-switches     #       0.005 M/sec
            191  CPU-migrations       #       0.000 M/sec
           8958  page-faults          #       0.001 M/sec
  <not counted>  cycles
  <not counted>  instructions
  <not counted>  cache-references
  <not counted>  cache-misses

 Wall-clock time elapsed:  1699.999995 msecs

Also add -v (--verbose) option to allow the printing of failed
counter opens.

Plus dont print 'inf' if wall-time is zero (due to jiffies granularity),
instead skip the printing of the CPU utilization factor.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf top: Wait for a minimal set of events before reading first snapshot
Frederic Weisbecker [Sat, 6 Jun 2009 21:10:43 +0000 (23:10 +0200)]
perf top: Wait for a minimal set of events before reading first snapshot

The first snapshot reading often occur before any events have
been read in the mapped perfcounter files.

Just wait until we have at least one event before starting the
snapshot, or the delay before the first set of entries to be
displayed may be long in case of low refresh rate.

Note: we could also use a semaphore to wait before
"print_entries" number of eveents is reached, but again this
value is tunable and we can't ensure we will even reach it.
Also we could base on a default mimimum set of entries for the
first refresh, say 15, but again, the minimal sample is
tunable, and we could end up displaying nothing until we have a
minimal default set of events, which can take some time in case
of high samples filters.

Hence this simple solution which partially covers the default
case.

[ Impact: fix display artifacts in perf top ]

Signed-off-by: Frederic Weisbecker <fweisbeec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <1244322643-6447-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf annotate: Fix command line help text
Ingo Molnar [Sat, 6 Jun 2009 19:25:29 +0000 (21:25 +0200)]
perf annotate: Fix command line help text

Arjan noticed this bug in the perf annotate help output:

    -s, --symbol <file>   symbol to annotate

that should be <symbol> instead.

Reported-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Initialize a stack variable before use
Arjan van de Ven [Sat, 6 Jun 2009 19:22:33 +0000 (21:22 +0200)]
perf_counter tools: Initialize a stack variable before use

the "perf report" utility crashed in some circumstances
because the "sym" stack variable was not initialized before used
(as also proven by valgrind).

With this fix both the crash goes away and valgrind no longer complains.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf annotate: Automatically pick up vmlinux in the local directory
Ingo Molnar [Sat, 6 Jun 2009 19:17:03 +0000 (21:17 +0200)]
perf annotate: Automatically pick up vmlinux in the local directory

Right now kernel debug info does not get resolved by default, because
we dont know where to look for the vmlinux.

The -k option can be used for that - but if no option is given, pick
up vmlinux files in the current directory - in case a kernel hacker
runs profiling from the source directory that the kernel was built in.

The real solution would be to embedd the location (and perhaps the
date/timestamp) of the vmlinux file in /proc/kallsyms, so that
tools can pick it up automatically.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Fix error condition in parse_aliases()
Ingo Molnar [Sat, 6 Jun 2009 19:04:17 +0000 (21:04 +0200)]
perf_counter tools: Fix error condition in parse_aliases()

gcc warned about this bug:

util/parse-events.c: In function ‘parse_generic_hw_symbols’:
util/parse-events.c:175: warning: comparison is always false due to limited range of data type
util/parse-events.c:182: warning: comparison is always false due to limited range of data type
util/parse-events.c:190: warning: comparison is always false due to limited range of data type

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Warning fixes on 32-bit
Arjan van de Ven [Sat, 6 Jun 2009 18:36:38 +0000 (20:36 +0200)]
perf_counter tools: Warning fixes on 32-bit

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Move from Documentation/perf_counter/ to tools/perf/
Ingo Molnar [Sat, 6 Jun 2009 18:33:43 +0000 (20:33 +0200)]
perf_counter tools: Move from Documentation/perf_counter/ to tools/perf/

Several people have suggested that 'perf' has become a full-fledged
tool that should be moved out of Documentation/. Move it to the
(new) tools/ directory.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoMerge branch 'linus' into perfcounters/core
Ingo Molnar [Sat, 6 Jun 2009 18:21:25 +0000 (20:21 +0200)]
Merge branch 'linus' into perfcounters/core

Merge reason: Pick up the latest fixes before the -v8 perfcounters
      release.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Add 'perf annotate' feature
Ingo Molnar [Sat, 6 Jun 2009 13:48:52 +0000 (15:48 +0200)]
perf_counter tools: Add 'perf annotate' feature

Add new perf sub-command to display annotated source code:

 $ perf annotate decode_tree_entry

------------------------------------------------
 Percent | Source code & Disassembly of /home/mingo/git/git
------------------------------------------------
         :
         : /home/mingo/git/git:     file format elf64-x86-64
         :
         :
         : Disassembly of section .text:
         :
         : 00000000004a0da0 <decode_tree_entry>:
         : *modep = mode;
         : return str;
         : }
         :
         : static void decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size)
         : {
    3.82 :   4a0da0: 41 54                 push   %r12
         : const char *path;
         : unsigned int mode, len;
         :
         : if (size < 24 || buf[size - 21])
    0.17 :   4a0da2: 48 83 fa 17           cmp    $0x17,%rdx
         : *modep = mode;
         : return str;
         : }
         :
         : static void decode_tree_entry(struct tree_desc *desc, const char *buf, unsigned long size)
         : {
    0.00 :   4a0da6: 49 89 fc              mov    %rdi,%r12
    0.00 :   4a0da9: 55                    push   %rbp
    3.37 :   4a0daa: 53                    push   %rbx
         : const char *path;
         : unsigned int mode, len;
         :
         : if (size < 24 || buf[size - 21])
    0.08 :   4a0dab: 76 73                 jbe    4a0e20 <decode_tree_entry+0x80>
    0.00 :   4a0dad: 80 7c 16 eb 00        cmpb   $0x0,-0x15(%rsi,%rdx,1)
    3.48 :   4a0db2: 75 6c                 jne    4a0e20 <decode_tree_entry+0x80>
         : static const char *get_mode(const char *str, unsigned int *modep)
         : {
         : unsigned char c;
         : unsigned int mode = 0;
         :
         : if (*str == ' ')
    1.94 :   4a0db4: 0f b6 06              movzbl (%rsi),%eax
    0.39 :   4a0db7: 3c 20                 cmp    $0x20,%al
    0.00 :   4a0db9: 74 65                 je     4a0e20 <decode_tree_entry+0x80>
         : return NULL;
         :
         : while ((c = *str++) != ' ') {
    0.06 :   4a0dbb: 89 c2                 mov    %eax,%edx
         : if (c < '0' || c > '7')
    1.99 :   4a0dbd: 31 ed                 xor    %ebp,%ebp
         : unsigned int mode = 0;
         :
         : if (*str == ' ')
         : return NULL;
         :
         : while ((c = *str++) != ' ') {
    1.74 :   4a0dbf: 48 8d 5e 01           lea    0x1(%rsi),%rbx
         : if (c < '0' || c > '7')
    0.00 :   4a0dc3: 8d 42 d0              lea    -0x30(%rdx),%eax
    0.17 :   4a0dc6: 3c 07                 cmp    $0x7,%al
    0.00 :   4a0dc8: 76 0d                 jbe    4a0dd7 <decode_tree_entry+0x37>
    0.00 :   4a0dca: eb 54                 jmp    4a0e20 <decode_tree_entry+0x80>
    0.00 :   4a0dcc: 0f 1f 40 00           nopl   0x0(%rax)
   16.57 :   4a0dd0: 8d 42 d0              lea    -0x30(%rdx),%eax
    0.14 :   4a0dd3: 3c 07                 cmp    $0x7,%al
    0.00 :   4a0dd5: 77 49                 ja     4a0e20 <decode_tree_entry+0x80>
         : return NULL;
         : mode = (mode << 3) + (c - '0');
    3.12 :   4a0dd7: 0f b6 c2              movzbl %dl,%eax
         : unsigned int mode = 0;
         :
         : if (*str == ' ')
         : return NULL;
         :
         : while ((c = *str++) != ' ') {
    0.00 :   4a0dda: 0f b6 13              movzbl (%rbx),%edx
   16.74 :   4a0ddd: 48 83 c3 01           add    $0x1,%rbx
         : if (c < '0' || c > '7')
         : return NULL;
         : mode = (mode << 3) + (c - '0');

The first column is the percentage of samples that arrived on that
particular line - relative to the total cost of the function.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Prepare for 'perf annotate'
Ingo Molnar [Sat, 6 Jun 2009 13:19:13 +0000 (15:19 +0200)]
perf_counter tools: Prepare for 'perf annotate'

Prepare for the 'perf annotate' implementation by splitting off
builtin-annotate.c from builtin-report.c.

( We keep this commit separate to ease the later librarization
  of the facilities that perf-report and perf-annotate shares. )

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Tidy up manpage details
Ingo Molnar [Tue, 15 Apr 2008 20:39:31 +0000 (22:39 +0200)]
perf_counter tools: Tidy up manpage details

Also fix a misalignment in usage string printing.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Uniform help printouts
Ingo Molnar [Fri, 13 Mar 2009 02:20:49 +0000 (03:20 +0100)]
perf_counter tools: Uniform help printouts

Also add perf list to command-list.txt.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Add help for perf list
Thomas Gleixner [Sat, 6 Jun 2009 12:56:33 +0000 (14:56 +0200)]
perf_counter tools: Add help for perf list

Also update other areas of the help texts.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Fix cache-event printout
Ingo Molnar [Sat, 6 Jun 2009 11:58:12 +0000 (13:58 +0200)]
perf_counter tools: Fix cache-event printout

Also standardize the cache printout (so that it can be pasted back
into the command) and sort out the aliases.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Add 'perf list' to list available events
Thomas Gleixner [Sat, 6 Jun 2009 10:24:17 +0000 (12:24 +0200)]
perf_counter tools: Add 'perf list' to list available events

perf list: List all the available event types which can be used in
-e (--event) options.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Implement generalized cache event types
Ingo Molnar [Fri, 5 Jun 2009 18:22:46 +0000 (20:22 +0200)]
perf_counter: Implement generalized cache event types

Extend generic event enumeration with the PERF_TYPE_HW_CACHE
method.

This is a 3-dimensional space:

       { L1-D, L1-I, L2, ITLB, DTLB, BPU } x
       { load, store, prefetch } x
       { accesses, misses }

User-space passes in the 3 coordinates and the kernel provides
a counter. (if the hardware supports that type and if the
combination makes sense.)

Combinations that make no sense produce a -EINVAL.
Combinations that are not supported by the hardware produce -ENOTSUP.

Extend the tools to deal with this, and rewrite the event symbol
parsing code with various popular aliases for the units and
access methods above. So 'l1-cache-miss' and 'l1d-read-ops' are
both valid aliases.

( x86 is supported for now, with the Nehalem event table filled in,
  and with Core2 and Atom having placeholder tables. )

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Separate out attr->type from attr->config
Ingo Molnar [Sat, 6 Jun 2009 07:58:57 +0000 (09:58 +0200)]
perf_counter: Separate out attr->type from attr->config

Counter type is a frequently used value and we do a lot of
bit juggling by encoding and decoding it from attr->config.

Clean this up by creating a separate attr->type field.

Also clean up the various similarly complex user-space bits
all around counter attribute management.

The net improvement is significant, and it will be easier
to add a new major type (which is what triggered this cleanup).

(This changes the ABI, all tools are adapted.)
(PowerPC build-tested.)

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf top: Fix zero or negative refresh delay
Frederic Weisbecker [Fri, 5 Jun 2009 17:31:01 +0000 (19:31 +0200)]
perf top: Fix zero or negative refresh delay

If perf top is executed with a zero value for the refresh rate,
we get a division by zero exception while computing samples_per_sec.

Also a zero refresh rate is not possible, neither do we want to
accept negative values.

[ Impact: fix division by zero in perf top ]

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1244223061-5399-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf record: Set frequency correctly
Ingo Molnar [Fri, 5 Jun 2009 16:37:22 +0000 (18:37 +0200)]
perf record: Set frequency correctly

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Sample and display frequency adjustment changes
Ingo Molnar [Fri, 5 Jun 2009 16:07:51 +0000 (18:07 +0200)]
perf_counter tools: Sample and display frequency adjustment changes

To allow the debugging of frequency-adjusting counters, sample
those adjustments and display them in perf report -D.

Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Fix frequency adjustment for < HZ
Peter Zijlstra [Fri, 5 Jun 2009 16:01:29 +0000 (18:01 +0200)]
perf_counter: Fix frequency adjustment for < HZ

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Add PERF_SAMPLE_PERIOD
Peter Zijlstra [Fri, 5 Jun 2009 13:05:43 +0000 (15:05 +0200)]
perf_counter: Add PERF_SAMPLE_PERIOD

In order to allow easy tracking of the period, also provide means of
adding it to the sample data.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Change PERF_SAMPLE_CONFIG into PERF_SAMPLE_ID
Peter Zijlstra [Fri, 5 Jun 2009 12:44:52 +0000 (14:44 +0200)]
perf_counter: Change PERF_SAMPLE_CONFIG into PERF_SAMPLE_ID

The purpose of PERF_SAMPLE_CONFIG was to identify the counters,
since then we've added counter ids, use those instead.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Remove -march=native
Ingo Molnar [Fri, 5 Jun 2009 15:56:21 +0000 (17:56 +0200)]
perf_counter tools: Remove -march=native

Turns out that neither PowerPC nor older x86 compilers know this switch
...

and since it does not make a measurable difference, just omit it.

Reported-by: Paul Mackerras <paulus@samba.org>
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Clarify events/samples naming
Ingo Molnar [Fri, 5 Jun 2009 12:29:10 +0000 (14:29 +0200)]
perf_counter tools: Clarify events/samples naming

A number of places said 'events' while they should say 'samples'.

Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Display user/kernel differentiator
Ingo Molnar [Fri, 5 Jun 2009 12:13:18 +0000 (14:13 +0200)]
perf report: Display user/kernel differentiator

Before:

    25.96%  copy_user_generic_string
    15.23%  two_op
    15.19%  one_op
     6.92%  enough_duration
     1.23%  alloc_pages_current
     1.14%  acpi_os_read_port
     1.08%  _spin_lock

After:

    25.96%  [k] copy_user_generic_string
    15.23%  [.] two_op
    15.19%  [.] one_op
     6.92%  [.] enough_duration
     1.23%  [k] alloc_pages_current
     1.14%  [k] acpi_os_read_port
     1.08%  [k] _spin_lock

The '[k]' differentiator is a quick clue that it's a kernel symbol,
without having to bring in the full dso column.

Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Deal with maps
Peter Zijlstra [Fri, 5 Jun 2009 12:04:59 +0000 (14:04 +0200)]
perf report: Deal with maps

In order to deal with [vdso] maps generalize the ip->symbol path
a bit and allow to override some bits with custom functions.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Generate mmap events for install_special_mapping()
Peter Zijlstra [Fri, 5 Jun 2009 12:04:55 +0000 (14:04 +0200)]
perf_counter: Generate mmap events for install_special_mapping()

In order to track the vdso also generate mmap events for
install_special_mapping().

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agox86: Set context.vdso before installing the mapping
Peter Zijlstra [Fri, 5 Jun 2009 12:04:51 +0000 (14:04 +0200)]
x86: Set context.vdso before installing the mapping

In order to make arch_vma_name() work from inside
install_special_mapping() we need to set the context.vdso
before calling it.

( This is needed for performance counters to be able to track
  this special executable area. )

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf record, top: Implement --freq
Ingo Molnar [Fri, 5 Jun 2009 11:27:02 +0000 (13:27 +0200)]
perf record, top: Implement --freq

Support frequency-based profiling and make it the default.

(Also add a Hz printout in perf top.)

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf record: Split out counter creation into a helper function
Ingo Molnar [Fri, 5 Jun 2009 11:18:41 +0000 (13:18 +0200)]
perf record: Split out counter creation into a helper function

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Fix incorrect printf formats
Yong Wang [Fri, 5 Jun 2009 03:37:35 +0000 (11:37 +0800)]
perf_counter tools: Fix incorrect printf formats

Otherwise the code does not compile on 32-bit boxes.

builtin-report.c: In function 'map__fprintf':
builtin-report.c:240: error: format '%lx' expects type 'long unsigned int', but argument 3 has type 'uint64_t'
builtin-report.c:240: error: format '%lx' expects type 'long unsigned int', but argument 4 has type 'uint64_t'
builtin-report.c:240: error: format '%lx' expects type 'long unsigned int', but argument 5 has type 'uint64_t'

Signed-off-by: Yong Wang <yong.y.wang@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <20090605033735.GA20451@ywang-moblin2.bj.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Fix lockup with interrupting counters
Paul Mackerras [Fri, 5 Jun 2009 02:36:28 +0000 (12:36 +1000)]
perf_counter: Fix lockup with interrupting counters

Commit 8e3747c1 ("perf_counter: Change data head from u32 to u64")
changed the type of 'head' in struct perf_mmap_data from atomic_t
to atomic_long_t, but missed converting one use of atomic_read on
it to atomic_long_read.  The effect of using atomic_read rather than
atomic_long_read on powerpc (and other big-endian architectures) is
that we get the high half of the 64-bit quantity, resulting in the
cmpxchg retry loop in perf_output_begin spinning forever as soon as
data->head becomes non-zero.  On little-endian architectures such as
x86 we would get the low half, resulting in a lockup once data->head
becomes greater than 4G.

This fixes it by using atomic_long_read rather than atomic_read.

[ Impact: fix perfcounter lockup on PowerPC / big-endian systems ]

Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <18984.33964.21541.743096@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/anholt...
Linus Torvalds [Thu, 4 Jun 2009 22:23:51 +0000 (15:23 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/anholt/drm-intel

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/anholt/drm-intel:
  drm/i915: Remove a bad BUG_ON in the fence management code.

14 years agoMerge branch 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied...
Linus Torvalds [Thu, 4 Jun 2009 22:23:39 +0000 (15:23 -0700)]
Merge branch 'drm-fixes' of git://git./linux/kernel/git/airlied/drm-2.6

* 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6:
  drm: ignore EDID with really tiny modes.
  drm: don't associate _DRM_DRIVER maps with a master
  drm/i915: intel_lvds.c fix section mismatch
  drm: Hook up DPMS property handling in drm_crtc.c. Add drm_helper_connector_dpms.
  drm: set permissions on edid file to 0444
  drm: add newlines to text sysfs files
  drm/radeon: fix ring free alignment calculations
  drm: fix irq naming for kms drivers.

14 years agodrivers/char/mem.c: avoid OOM lockup during large reads from /dev/zero
Salman Qazi [Thu, 4 Jun 2009 22:20:39 +0000 (15:20 -0700)]
drivers/char/mem.c: avoid OOM lockup during large reads from /dev/zero

While running 20 parallel instances of dd as follows:

  #!/bin/bash
  for i in `seq 1 20`; do
           dd if=/dev/zero of=/export/hda3/dd_$i bs=1073741824 count=1 &
  done
  wait

on a 16G machine, we noticed that rather than just killing the processes,
the entire kernel went down.  Stracing dd reveals that it first does an
mmap2, which makes 1GB worth of zero page mappings.  Then it performs a
read on those pages from /dev/zero, and finally it performs a write.

The machine died during the reads.  Looking at the code, it was noticed
that /dev/zero's read operation had been changed by
557ed1fa2620dc119adb86b34c614e152a629a80 ("remove ZERO_PAGE") from giving
zero page mappings to actually zeroing the page.

The zeroing of the pages causes physical pages to be allocated to the
process.  But, when the process exhausts all the memory that it can, the
kernel cannot kill it, as it is still in the kernel mode allocating more
memory.  Consequently, the kernel eventually crashes.

To fix this, I propose that when a fatal signal is pending during
/dev/zero read operation, we simply return and let the user process die.

Signed-off-by: Salman Qazi <sqazi@google.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
[ Modified error return and comment trivially.  - Linus]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agoperf_counter tools: Fix warn_unused_result warnings
Frederic Weisbecker [Thu, 4 Jun 2009 20:15:58 +0000 (22:15 +0200)]
perf_counter tools: Fix warn_unused_result warnings

Fix warnings for return values that we don't care about:

 util/quote.c:222: attention : ignoring return value of ‘fwrite’, declared with attribute warn_unused_result
 util/quote.c:235: attention : ignoring return value of ‘fwrite’, declared with attribute warn_unused_result
 util/quote.c: In function ‘write_name_quotedpfx’:
 util/quote.c:290: attention : ignoring return value of ‘fwrite’, declared with attribute warn_unused_result

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1244146558-8635-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Add -vvv to print the list of threads and its mmaps
Arnaldo Carvalho de Melo [Thu, 4 Jun 2009 16:54:00 +0000 (13:54 -0300)]
perf report: Add -vvv to print the list of threads and its mmaps

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Sleep before refresh using poll in perf top
Frederic Weisbecker [Thu, 4 Jun 2009 18:48:04 +0000 (20:48 +0200)]
perf_counter: Sleep before refresh using poll in perf top

perf top is refreshed every delay_secs the thread runs in such
loop:

while (sleep(delay_secs)) {
print_sym_table();
}

At the end of print_sym_table(), poll is used without sleep delay
to check if we have something from stdin.

It means that this check is done only every delay_secs, which can
be higher that 2 secs if the user defined a custom refresh rate.

We can drop sleep() here and directly use poll to wait between
refresh periods, so that the reaction after the user stops perf top
after typing "Enter" is immediate and doesn't suffer from the
delay_secs latency.

Nb: poll doesn't add any overhead that can parasite perf top measures
since it sleeps the entire timeout here.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <1244141284-7507-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agolguest: fix 'unhandled trap 13' with CONFIG_CC_STACKPROTECTOR
Rusty Russell [Wed, 3 Jun 2009 05:22:24 +0000 (14:52 +0930)]
lguest: fix 'unhandled trap 13' with CONFIG_CC_STACKPROTECTOR

We don't set up the canary; let's disable stack protector on boot.c so
we can get into lguest_init, then set it up.  As a side effect,
switch_to_new_gdt() sets up %fs for us properly too.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
14 years agoperf_counter tools: Use fork and remove munmap events
Peter Zijlstra [Thu, 4 Jun 2009 14:53:49 +0000 (16:53 +0200)]
perf_counter tools: Use fork and remove munmap events

Use fork events to clone comm and map data and remove everything
munmap related

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Remove munmap stuff
Peter Zijlstra [Thu, 4 Jun 2009 15:08:58 +0000 (17:08 +0200)]
perf_counter: Remove munmap stuff

In name of keeping it simple, only track mmap events. Userspace
will have to remove old overlapping maps when it encounters them.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Add fork event
Peter Zijlstra [Thu, 4 Jun 2009 14:53:44 +0000 (16:53 +0200)]
perf_counter: Add fork event

Create a fork event so that we can easily clone the comm and
dso maps without having to generate all those events.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf stat: Update help text
Ingo Molnar [Thu, 4 Jun 2009 14:33:00 +0000 (16:33 +0200)]
perf stat: Update help text

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Bail out if there are unrecognized options/arguments
Ingo Molnar [Thu, 4 Jun 2009 14:24:37 +0000 (16:24 +0200)]
perf report: Bail out if there are unrecognized options/arguments

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Dont output in color on !tty
Ingo Molnar [Thu, 4 Jun 2009 13:40:25 +0000 (15:40 +0200)]
perf_counter tools: Dont output in color on !tty

Dont emit ASCII color characters if the terminal is not a tty,
such as when perf report gets redirected into a file.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Add color terminal output support
Ingo Molnar [Thu, 4 Jun 2009 13:19:47 +0000 (15:19 +0200)]
perf_counter tools: Add color terminal output support

Add Git's color printing library to util/color.[ch].

Add it to perf report, with a trivial example to print high-overhead
entries in red, low-overhead entries in green.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Add consistent spacing rules
Peter Zijlstra [Thu, 4 Jun 2009 13:16:56 +0000 (15:16 +0200)]
perf report: Add consistent spacing rules

Make the sort header and the print function have the same column width.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Print out the total number of events
Ingo Molnar [Thu, 4 Jun 2009 12:21:16 +0000 (14:21 +0200)]
perf report: Print out the total number of events

So that the statistical quality of the profile can be estimated at a glance.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Print out symbol parsing errors only if --verbose
Ingo Molnar [Thu, 4 Jun 2009 12:13:04 +0000 (14:13 +0200)]
perf_counter tools: Print out symbol parsing errors only if --verbose

Also, add a suggestion to 'perf report', if the default sort order is
used.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Simplify symbol output
Peter Zijlstra [Thu, 4 Jun 2009 13:00:45 +0000 (15:00 +0200)]
perf report: Simplify symbol output

The DSO can be printed already - no need to repeat it in the
symbol field.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter tools: Build with native optimization
Ingo Molnar [Thu, 4 Jun 2009 11:58:13 +0000 (13:58 +0200)]
perf_counter tools: Build with native optimization

Build the tools with -march=native by default.

No measurable difference in speed though, compared to the
default, on a Nehalem testbox.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agodrm/i915: Remove a bad BUG_ON in the fence management code.
Eric Anholt [Thu, 4 Jun 2009 11:18:14 +0000 (11:18 +0000)]
drm/i915: Remove a bad BUG_ON in the fence management code.

This could be triggered by a gtt mapping fault on 965 that decides to
remove the fence from another object that happens to be active currently.
Since the other object doesn't rely on the fence reg for its execution, we
don't wait for it to finish.  We'll soon be not waiting on 915 most of the
time as well, so just drop the BUG_ON.

Signed-off-by: Eric Anholt <eric@anholt.net>
14 years agoperf record/report: Fix PID/COMM handling
Ingo Molnar [Thu, 4 Jun 2009 11:41:22 +0000 (13:41 +0200)]
perf record/report: Fix PID/COMM handling

Fix two bugs causing lost comm mappings:

 - initial PID is not 0 but getpid()

 - when we are unable to handle an mmap event, dont assume the event
   itself is broken - try to parse the stream. This way we wont lose
   comm events.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: Documentation update
Yong Wang [Wed, 3 Jun 2009 08:42:25 +0000 (16:42 +0800)]
perf_counter: Documentation update

The 'nmi' bit is no longer there.

Signed-off-by: Yong Wang <yong.y.wang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Arjan van de Ven <arjan@linux.intel.com>
LKML-Reference: <20090603084225.GA6553@ywang-moblin2.bj.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf_counter: powerpc: Use new identifier names in powerpc-specific code
Paul Mackerras [Wed, 3 Jun 2009 23:49:59 +0000 (09:49 +1000)]
perf_counter: powerpc: Use new identifier names in powerpc-specific code

Commit b23f3325 ("perf_counter: Rename various fields") fixed up
most of the uses of the renamed fields, but missed one instance
of "record_type" in powerpc-specific code which needs to be changed
to "sample_type", and a "PERF_RECORD_ADDR" in the same statement that
needs to be changed to "PERF_SAMPLE_ADDR", causing compilation
errors on powerpc.  This fixes it.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <18983.3111.770392.800486@cargo.ozlabs.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf top: Reduce default filter threshold
Ingo Molnar [Thu, 4 Jun 2009 06:53:05 +0000 (08:53 +0200)]
perf top: Reduce default filter threshold

On idle systems 'perf top' comes up empty by default, because the event
count filter is set to 100.

Reduce it to 5 instead.

Also add an option to limit the number of functions displayed.

Reported-by: Steven Rostedt <rostedt@goodmis.org>
Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agoperf report: Fix rbtree bug
Arnaldo Carvalho de Melo [Thu, 4 Jun 2009 02:02:33 +0000 (23:02 -0300)]
perf report: Fix rbtree bug

Ingo Molnar reported:

> FYI, i just got this crash (segfault) in perf report after
> collecting a long profile from Xorg:
>
> Starting program: /home/mingo/tip/Documentation/perf_counter/perf report
> [Thread debugging using libthread_db enabled]
> Detaching after fork from child process 20008.
> [New Thread 0x7f92fd62a6f0 (LWP 20005)]
>
> Program received signal SIGSEGV, Segmentation fault.
> 0x000000000041031a in __rb_erase_color (node=0x142c090, parent=0x0,
> root=0x881918)
>     at util/rbtree.c:143
> 143 if (parent->rb_left == node)

It was a problem introduced in this cset:

 perf report: Fix comm sorting - 8229289b607682f90b946ad2c319526303c17700

This patch should fix it.

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Stephane Eranian <eranian@googlemail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
14 years agodrm: ignore EDID with really tiny modes.
Adam Jackson [Thu, 4 Jun 2009 00:20:34 +0000 (10:20 +1000)]
drm: ignore EDID with really tiny modes.

Some EDIDs lie and report tiny modes that aren't possible. Ignore
these modes.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm: don't associate _DRM_DRIVER maps with a master
Ben Skeggs [Tue, 26 May 2009 00:35:52 +0000 (10:35 +1000)]
drm: don't associate _DRM_DRIVER maps with a master

A driver will use the _DRM_DRIVER map flag to indicate that it wants
to be responsible for removing the map itself, bypassing the DRM's
automagic cleanup code.

Since the multi-master changes this has been broken, resulting in some
drivers having their registers unmapped before it's finished with them.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm/i915: intel_lvds.c fix section mismatch
Jaswinder Singh Rajput [Wed, 3 Jun 2009 23:41:19 +0000 (09:41 +1000)]
drm/i915: intel_lvds.c fix section mismatch

intel_no_lvds[] does not require __initdata as it is used only by

void intel_lvds_init(struct drm_device *dev).

Signed-off-by: Jaswinder Singh Rajput <jaswinder@kernel.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm: Hook up DPMS property handling in drm_crtc.c. Add drm_helper_connector_dpms.
Keith Packard [Sun, 31 May 2009 03:42:28 +0000 (20:42 -0700)]
drm: Hook up DPMS property handling in drm_crtc.c. Add drm_helper_connector_dpms.

Making the drm_crtc.c code recognize the DPMS property and invoke the
connector->dpms function doesn't remove any capability from the driver while
reducing code duplication.

That just highlighted the problem with the existing DPMS functions which
could turn off the connector, but failed to turn off any relevant crtcs. The
new drm_helper_connector_dpms function manages all of that, using the
drm_helper-specific crtc and encoder dpms functions, automatically computing
the appropriate DPMS level for each object in the system.

This fixes the current troubles in the i915 driver which left PLLs, pipes
and planes running while in DPMS_OFF mode or even while they were unused.

Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm: set permissions on edid file to 0444
Keith Packard [Sun, 31 May 2009 03:42:26 +0000 (20:42 -0700)]
drm: set permissions on edid file to 0444

Without initializing the sysfs attributes for the edid file,
it was created with mode 0, making it difficult for applications to use.

Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm: add newlines to text sysfs files
Keith Packard [Sun, 31 May 2009 03:42:25 +0000 (20:42 -0700)]
drm: add newlines to text sysfs files

The contents of various simple text files in sysfs should end with
a newline to make them easier to read from the console.

Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm/radeon: fix ring free alignment calculations
Dave Airlie [Wed, 3 Jun 2009 21:08:13 +0000 (07:08 +1000)]
drm/radeon: fix ring free alignment calculations

fd.o bz#21849

We were aligning to +16 dwords, instead of to the next 16dword
boundary in the ring. Fix the calculation to go to the next 16dword
boundary when space checking.

Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agodrm: fix irq naming for kms drivers.
Dave Airlie [Tue, 2 Jun 2009 06:50:35 +0000 (16:50 +1000)]
drm: fix irq naming for kms drivers.

allocating devname in the i915 driver was a hack originally and I
forgot to figure out how to do this properly back then.

So this is the cleaner version that just picks devname or driver name
in the irq code.

It removes the devname allocs from the i915 driver.

Signed-off-by: Dave Airlie <airlied@redhat.com>