perf python: Support the PERF_RECORD_SWITCH event

To test it check tools/perf/python/twatch.py, after following the
instructions there to enable context_switch, output looks like:

  [root@zoo linux]# tools/perf/python/twatch.py
  cpu: 1, pid: 31463, tid: 31463 { type: context_switch, next_prev_pid: 31463, next_prev_tid: 31463, switch_out: 0 }
  cpu: 2, pid: 31463, tid: 31496 { type: context_switch, next_prev_pid: 31463, next_prev_tid: 31496, switch_out: 0 }
  cpu: 2, pid: 31463, tid: 31496 { type: context_switch, next_prev_pid: 31463, next_prev_tid: 31496, switch_out: 1 }
  cpu: 3, pid: 31463, tid: 31527 { type: context_switch, next_prev_pid: 31463, next_prev_tid: 31527, switch_out: 0 }
  cpu: 1, pid: 31463, tid: 31463 { type: context_switch, next_prev_pid: 31463, next_prev_tid: 31463, switch_out: 1 }
  cpu: 3, pid: 31463, tid: 31527 { type: context_switch, next_prev_pid: 31463, next_prev_tid: 31527, switch_out: 1 }
  cpu: 1, pid: 31463, tid: 31463 { type: context_switch, next_prev_pid: 31463, next_prev_tid: 31463, switch_out: 0 }
  ^CTraceback (most recent call last):
    File "tools/perf/python/twatch.py", line 67, in <module>
      main(context_switch = 1, thread = 31463)
    File "tools/perf/python/twatch.py", line 40, in main
      evlist.poll(timeout = -1)
  KeyboardInterrupt
  [root@zoo linux]#

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Guy Streeter <streeter@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-1ukistmpamc5z717k80ctcp2@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix handling read result using a signed variable

The function can return negative value, assigning it to unsigned
variable can cause memory corruption.

The problem has been detected using proposed semantic patch
scripts/coccinelle/tests/unsigned_lesser_than_zero.cocci [1].

[1]: http://permalink.gmane.org/gmane.linux.kernel/2038576

Signed-off-by: Andrzej Hajda <a.hajda@samsung.com>
Cc: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: kernel-janitors@vger.kernel.org
Link: http://lkml.kernel.org/r/1444122017-16856-1-git-send-email-a.hajda@samsung.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Use hpp_dimension__add_output to register hpp columns

The perf_hpp__init currently does not respect sorting dimensions and the
setup_sorting function could endup queueing same format twice. That
screwed up the perf_hpp__list and got stuck in loop within
perf_hpp__setup_output_field function.

  $ perf report -F +overhead

  0x00000000004c1355 in perf_hpp__is_sort_entry (format=format@entry=0x880440 <perf_hpp.format>) at util/sort.c:1506
  1506    {

     #0  0x00000000004c1355 in perf_hpp__is_sort_entry (format=format@entry=0x880440 <perf_hpp.format>) at util/sort.c:1506
     #1  0x00000000004c139d in perf_hpp__same_sort_entry (a=a@entry=0x880440 <perf_hpp.format>, b=b@entry=0x2bb2fe0) at util/sort.c:1380
     #2  0x00000000004f8d3c in perf_hpp__setup_output_field () at ui/hist.c:554
     #3  0x00000000004c1d1e in setup_sorting () at util/sort.c:1984
     #4  0x000000000042efbf in cmd_report (argc=0, argv=0x7ffea5a0e790, prefix=<optimized out>) at builtin-report.c:874
     #5  0x0000000000476f13 in run_builtin (p=p@entry=0x875628 <commands+168>, argc=argc@entry=3, argv=argv@entry=0x7ffea5a0e790) at perf.c:385
     #6  0x000000000047710b in handle_internal_command (argc=3, argv=0x7ffea5a0e790) at perf.c:445
     #7  0x0000000000477176 in run_argv (argcp=argcp@entry=0x7ffea5a0e5fc, argv=argv@entry=0x7ffea5a0e5f0) at perf.c:489
     #8  0x00000000004773e7 in main (argc=3, argv=0x7ffea5a0e790) at perf.c:606

Using hpp_dimension__add_output function to register the output column.
It will also mark the dimension as taken and omit above stuck.

Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444134312-29136-4-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Introduce hpp_dimension__add_output function

This function will allow to register output column from ui code and
respect taken sort/output dimensions.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444134312-29136-3-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Get rid of superfluos call to reset_dimensions

There's no need to call reset_dimensions within __setup_output_field
function. It's already called in its caller setup_sorting right before
perf_hpp__init, which will be changed in following patch to respect
taken dimension.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444134312-29136-2-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/x86/intel/uncore: Fix multi-segment problem of perf_event_intel_uncore

In multi-segment system, uncore devices may belong to buses whose segment
number is other than 0:

  ....
  0000:ff:10.5 System peripheral: Intel Corporation Xeon E5 v3/Core i7 Scratchpad & Semaphore Registers (rev 03)
  ...
  0001:7f:10.5 System peripheral: Intel Corporation Xeon E5 v3/Core i7 Scratchpad & Semaphore Registers (rev 03)
  ...
  0001:bf:10.5 System peripheral: Intel Corporation Xeon E5 v3/Core i7 Scratchpad & Semaphore Registers (rev 03)
  ...
  0001:ff:10.5 System peripheral: Intel Corporation Xeon E5 v3/Core i7 Scratchpad & Semaphore Registers (rev 03
  ...

In that case, relation of bus number and physical id may be broken
because "uncore_pcibus_to_physid" doesn't take account of PCI segment.
For example, bus 0000:ff and 0001:ff uses the same entry of
"uncore_pcibus_to_physid" array.

This patch fixes this problem by introducing the segment-aware pci2phy_map instead.

Signed-off-by: Taku Izumi <izumi.taku@jp.fujitsu.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: acme@kernel.org
Cc: hpa@zytor.com
Link: http://lkml.kernel.org/r/1443096621-4119-1-git-send-email-izumi.taku@jp.fujitsu.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf/x86: Add Intel cstate PMUs support

This patch adds new PMUs to support cstate related free running
(read-only) counters. These counters may be used simultaneously by other
tools, such as turbostat. However, it still make sense to implement them
in perf. Because we can conveniently collect them together with other
events, and allow to use them from tools without special MSR access
code.

These counters include CORE_C*_RESIDENCY and PKG_C*_RESIDENCY.
According to counters' scope and category, two PMUs are registered with
the perf_event core subsystem.

 - 'cstate_core': The counter is available for each physical core. The
                  counters include CORE_C*_RESIDENCY.

 - 'cstate_pkg':  The counter is available for each physical package. The
                  counters include PKG_C*_RESIDENCY.

The events are exposed in sysfs for use by perf stat and other tools.
The files are:

  /sys/devices/cstate_core/events/c*-residency
  /sys/devices/cstate_pkg/events/c*-residency

These events only support system-wide mode counting.
The /sys/devices/cstate_*/cpumask file can be used by tools to figure
out which CPUs to monitor by default.

The PMU type (attr->type) is dynamically allocated and is available from
/sys/devices/core_misc/type and /sys/device/cstate_*/type.

Sampling is not supported.

Here is an example.

 - To caculate the fraction of time when the core is running in C6 state
   CORE_C6_time% = CORE_C6_RESIDENCY / TSC

 # perf stat -x, -e"cstate_core/c6-residency/,msr/tsc/" -C0 -- taskset -c 0 sleep 5

   11838820015,,cstate_core/c6-residency/,5175919658,100.00
   11877130740,,msr/tsc/,5175922010,100.00

 For sleep, 99.7% of time we ran in C6 state.

 # perf stat -x, -e"cstate_core/c6-residency/,msr/tsc/" -C0 -- taskset -c 0 busyloop

   1253316,,cstate_core/c6-residency/,4360969154,100.00
   10012635248,,msr/tsc/,4360972366,100.00

 For busyloop, 0.01% of time we ran in C6 state.

Signed-off-by: Kan Liang <kan.liang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: acme@kernel.org
Cc: eranian@google.com
Link: http://lkml.kernel.org/r/1443443404-8581-1-git-send-email-kan.liang@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

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:

User visible changes:

  - Switch the default callchain output mode to 'graph,0.5,caller', to make it
    look like the default for other tools, reducing the learning curve for
    people used to 'caller' based viewing. (Arnaldo Carvalho de Melo)

  - Implement column based horizontal scrolling in the hists browser (top, report),
    making it possible to use the TUI for things like 'perf mem report' where
    there are many more columns than can fit in a terminal. (Arnaldo Carvalho de Melo)

  - Support sorting by symbol_iaddr with perf.data files produced by
    'perf mem record'. (Don Zickus)

  - Display DATA_SRC sample type bit, i.e. when running 'perf evlist -v' the
    "DATA_SRC" wasn't appearing when set, fix it to look like: (Jiri Olsa)

      cpu/mem-loads/pp: ...SNIP... sample_type: IP|TID|TIME|ADDR|CPU|PERIOD|DATA_SRC

  - Introduce the 'P' event modifier, meaning 'max precision level, please', i.e.:

     $ perf record -e cycles:P usleep 1

    Is now similar to:

     $ perf record usleep 1

    Useful, for instance, when specifying multiple events. (Jiri Olsa)

  - Make 'perf -v' and 'perf -h' work. (Jiri Olsa)

  - Fail properly when pattern matching fails to find a tracepoint, i.e.
    '-e non:existent' was being correctly handled, with a proper error message
    about that not being a valid event, but '-e non:existent*' wasn't,
    fix it. (Jiri Olsa)

Infrastructure changes:

  - Separate arch specific entries in 'perf test' and add an 'Intel CQM' one
    to be fun on x86 only. (Matt Fleming)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf tools: Fail properly in case pattern matching fails to find tracepoint

Currently we dont fail properly when pattern matching fails to find any
tracepoint.

Current behaviour:

  $ perf record -e 'sched:krava*' sleep 1
  WARNING: event parser found nothinginvalid or unsupported event: 'sched:krava*'
  Run 'perf list' for a list of valid events

  usage: perf record [<options>] [<command>]
     or: perf record [<options>] -- <command> [<options>]

This patch change:

  $ perf record -e 'sched:krava*' sleep 1
  event syntax error: 'sched:krava*'
                       \___ unknown tracepoint

  Error:  File /sys/kernel/debug/tracing/events/sched/krava* not found.
  Hint:   Perhaps this kernel misses some CONFIG_ setting to enable this feature?.

  Run 'perf list' for a list of valid events

   usage: perf record [<options>] [<command>]
      or: perf record [<options>] -- <command> [<options>]

Reported-by: Daniel Bristot de Oliveira <bristot@redhat.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444073477-3181-1-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Implement horizontal scrolling

Do it using the recently introduced ui_brower scrolling mode, setting
ui_browser.columns to the number of sort columns and then, when
rendering each line, skipping as many initial columns as the user
pressed the right arrow.

As the user presses the left arrow, the ui_browser code will remove the
scrolling counter and the left scrolling takes place.

The right arrow key was an alias for ENTER, so people used to press it
may get a bit annoyed at first, sorry! Ditto for ESC and the left key.

Callchains can be left as is or we can, when rendering the Symbol
column, store the at what position on the screen it is and then
using ui_browser__gotorc() to print it from there, i.e. the callchain
would move around with the symbol.

Leaving it as is, i.e. at a fixed position, close to the left, saves
precious screen real state for it, so I'm inclined to leave it as is
now.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Chandler Carruth <chandlerc@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ccqq9sabgfge5dwbqjwh71ij@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf ui browser: Optional horizontal scrolling key binding

If the classes derived from ui_browser want to do some sort of
horizontal scrolling, they have just to set ui_browser->columns to
the number of columns available.

Those columns can be the number of characters on the screen, if what is
desired is to scroll character by character, or the number of columns in
a spreadsheet like table.

This is what the hist_browser will do, skipping ui_browser->horiz_scroll
columns when rendering each of its lines.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-q6a22bpmpgcr1awgzrmd4jrs@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf callchain: Switch default to 'graph,0.5,caller'

Which is the most common default found in other similar tools.

Requested-by: Ingo Molnar <mingo@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Chandler Carruth <chandlerc@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://www.youtube.com/watch?v=nXaxk27zwlk
Link: http://lkml.kernel.org/n/tip-v8lq36aispvdwgxdmt9p9jd9@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Add Intel CQM test

Peter reports that it's possible to trigger a WARN_ON_ONCE() in the
Intel CQM code by combining a hardware event and an Intel CQM
(software) event into a group. Unfortunately, the perf tools are not
able to create this bundle and we need to manually construct a test
case.

For posterity, record Peter's proof of concept test case in tools/perf
so that it presents a model for how we can perform architecture
specific tests, or "arch tests", in perf in the future.

The particular issue triggered in the test case is that when the
counter for the hardware event overflows and triggers a PMI we'll read
both the hardware event and the software event counters.
Unfortunately, for CQM that involves performing an IPI to read the CQM
event counters on all sockets, which in NMI context triggers the
WARN_ON_ONCE().

Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kanaka Juvva <kanaka.d.juvva@intel.com>
Cc: Vikas Shivappa <vikas.shivappa@intel.com>
Cc: Vince Weaver <vince@deater.net>
Link: http://lkml.kernel.org/r/1437490509-15373-1-git-send-email-matt@codeblueprint.co.uk
Link: http://lkml.kernel.org/n/tip-3p4ra0u8vzm7m289a1m799kf@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Move x86 tests into arch directory

Move out the x86-specific tests into tools/perf/arch/x86/tests and
define an 'arch_tests' array, which is the list of tests that only apply
to the build architecture.

We can also now begin to get rid of some of the #ifdef code that is
present in the generic perf tests.

Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kanaka Juvva <kanaka.d.juvva@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Vikas Shivappa <vikas.shivappa@intel.com>
Cc: Vince Weaver <vince@deater.net>
Link: http://lkml.kernel.org/n/tip-9s68h4ptg06ah0lgnjz55mqn@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Add arch tests

Tests that only make sense for some architectures currently live in
the same place as the generic tests. Move out the x86-specific tests
into tools/perf/arch/x86/tests and define an 'arch_tests' array, which
is the list of tests that only apply to the build architecture.

The main idea is to encourage developers to add arch tests to build
out perf's test coverage, without dumping everything in
tools/perf/tests.

Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kanaka Juvva <kanaka.d.juvva@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Vikas Shivappa <vikas.shivappa@intel.com>
Cc: Vince Weaver <vince@deater.net>
Link: http://lkml.kernel.org/n/tip-p4uc1c15ssbj8xj7ku5slpa6@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Handle -h and -v options

Adding handling for '-h' and '-v' options to invoke help and version
command respectively.

Current behaviour is:

   $ perf -v
   Unknown option: -v

    Usage: perf [--version] [--help] [OPTIONS] COMMAND [ARGS]
   $ perf -h
   Unknown option: -h

    Usage: perf [--version] [--help] [OPTIONS] COMMAND [ARGS]

New behaviour:

  $ perf -h

   usage: perf [--version] [--help] [OPTIONS] COMMAND [ARGS]

   The most commonly used perf commands are:
     annotate        Read perf.data (created by perf record) and display annotated code
     archive         Create archive with object files with build-ids found in perf.data file
     bench           General framework for benchmark suites
   ...

  $ perf -v
  perf version 4.3.rc3.gc99e32

Updated man page.

Requested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-10-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Setup proper width for symbol_iaddr field

We need to properly initialize column width for symbol_iaddr field, so
all symbols could fit in the column.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-9-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add support for sorting on the iaddr

Sorting on 'symbol' gives to broad a resolution as it can cover a range
of IP address.  Use the iaddr instead to get proper sorting on IP
addresses.  Need to use the 'mem_sort' feature of perf record.

New sort option is: symbol_iaddr, header label is 'Code Symbol'.

  $ perf mem report --stdio -F +symbol_iaddr
  # Overhead       Samples  Code Symbol              Local Weight
  # ........  ............  ........................ ............
  #
      54.08%             1  [k] nmi_handle           192
       4.51%             1  [k] finish_task_switch   16
       3.66%             1  [.] malloc               13
       3.10%             1  [.] __strcoll_l          11

Signed-off-by: Don Zickus <dzickus@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-8-git-send-email-jolsa@kernel.org
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests: Add parsing test for 'P' modifier

We cant test 'P' modifier gets properly parsed, the functionality test
itself is beyond this suite.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-7-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Introduce 'P' modifier to request max precision

The 'P' will cause the event to get maximum possible detected precise
level.

Following record:
  $ perf record -e cycles:P ...

will detect maximum precise level for 'cycles' event and use it.

Commiter note:

Testing it:

  $ perf record -e cycles:P usleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.013 MB perf.data (9 samples) ]
  $ perf evlist
  cycles:P
  $ perf evlist -v
  cycles:P: size: 112, { sample_period, sample_freq }: 4000, sample_type:
  IP|TID|TIME|PERIOD, disabled: 1, inherit: 1, mmap: 1, comm: 1, freq: 1,
  enable_on_exec: 1, task: 1, precise_ip: 2, sample_id_all: 1, mmap2: 1,
  comm_exec: 1
  $

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-6-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Export perf_event_attr__set_max_precise_ip()

It'll be used in following patch.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-5-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Fix sizeof_sym_hist overflow issue

The annotated_source::sizeof_sym_hist could easily overflow int size,
resulting in crash in __symbol__inc_addr_samples.

Changing its type int size_t as was probably intended from beginning
based on the initialization code in symbol__alloc_hist.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-4-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Display DATA_SRC sample type bit

Adding DATA_SRC bit_name call to display sample_type properly.

   $ perf evlist -v
   cpu/mem-loads/pp: ...SNIP... sample_type: IP|TID|TIME|ADDR|CPU|PERIOD|DATA_SRC, ...

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-3-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib api fs: No need to use PATH_MAX + 1

Because there's no point, PATH_MAX is big enough.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Don Zickus <dzickus@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1444068369-20978-2-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

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:

User visible changes:

 - Do event name substring search as last resort in 'perf list'.
   (Arnaldo Carvalho de Melo)

   E.g.:

    # perf list clock

    List of pre-defined events (to be used in -e):

     cpu-clock                                          [Software event]
     task-clock                                         [Software event]

     uncore_cbox_0/clockticks/                          [Kernel PMU event]
     uncore_cbox_1/clockticks/                          [Kernel PMU event]

     kvm:kvm_pvclock_update                             [Tracepoint event]
     kvm:kvm_update_master_clock                        [Tracepoint event]
     power:clock_disable                                [Tracepoint event]
     power:clock_enable                                 [Tracepoint event]
     power:clock_set_rate                               [Tracepoint event]
     syscalls:sys_enter_clock_adjtime                   [Tracepoint event]
     syscalls:sys_enter_clock_getres                    [Tracepoint event]
     syscalls:sys_enter_clock_gettime                   [Tracepoint event]
     syscalls:sys_enter_clock_nanosleep                 [Tracepoint event]
     syscalls:sys_enter_clock_settime                   [Tracepoint event]
     syscalls:sys_exit_clock_adjtime                    [Tracepoint event]
     syscalls:sys_exit_clock_getres                     [Tracepoint event]
     syscalls:sys_exit_clock_gettime                    [Tracepoint event]
     syscalls:sys_exit_clock_nanosleep                  [Tracepoint event]
     syscalls:sys_exit_clock_settime                    [Tracepoint event]

 - Reduce min 'perf stat --interval-print/-I' to 10ms. (Kan Liang)

   perf stat --interval in action:

   # perf stat -e cycles -I 50 -a usleep $((200 * 1000))
   print interval < 100ms. The overhead percentage could be high in some cases. Please proceed with caution.
   #   time                    counts unit events
      0.050233636         48,240,396      cycles
      0.100557098         35,492,594      cycles
      0.150804687         39,295,112      cycles
      0.201032269         33,101,961      cycles
      0.201980732            786,379      cycles
  #

 - Allow for max_stack greater than PERF_MAX_STACK_DEPTH, as when
   synthesizing callchains from Intel PT data. (Adrian Hunter)

 - Allow probing on kmodules without DWARF. (Masami Hiramatsu)

 - Fix a segfault when processing a perf.data file with callchains using
   "perf report --call-graph none". (Namhyung Kim)

 - Fix unresolved COMMs in 'perf top' when -s comm is used. (Namhyung Kim)

 - Register idle thread in 'perf top'. (Namhyung Kim)

 - Change 'record.samples' type to unsigned long long, fixing output of
   number of samples in 32-bit architectures. (Yang Shi)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf stat: Reduce min --interval-print to 10ms

The --interval-print parameter was limited to 100ms. However, for
example, 10ms is required to do sophisticated bandwidth analysis using
uncore events.

The test shows that the overhead of the system-wide uncore monitoring
with 10ms interval is only ~2%. So this patch reduces the minimal
interval-print allowd to 10ms.

But 10ms may not work well for all cases. For example, when the
cpus/threads number is very large, for system-wide core event monitoring
the overhead could be high.

To handle this issue, a warning will be displayed when the
interval-print is set between 10ms to 100ms. So users can make a
decision according to their specific cases.

 # perf stat -e uncore_imc_1/cas_count_read/ -a --interval-print 10 -- sleep 1

 print interval < 100ms. The overhead percentage could be high in some
 cases. Please proceed with caution.
 #           time             counts unit events
      0.010200451               0.10 MiB  uncore_imc_1/cas_count_read/
      0.020475117               0.02 MiB  uncore_imc_1/cas_count_read/
      0.030692800               0.01 MiB  uncore_imc_1/cas_count_read/
      0.040948161               0.02 MiB  uncore_imc_1/cas_count_read/
      0.051159564               0.00 MiB  uncore_imc_1/cas_count_read/

Signed-off-by: Kan Liang <kan.liang@intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/1443776674-42511-1-git-send-email-kan.liang@intel.com
[ Added warning about overhead when using sub 100ms intervals to the man page ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Change 'record.samples' type to unsigned long long

When run "perf record -e", the number of samples showed up is wrong on some
32 bit systems, i.e. powerpc and arm.

For example, run the below commands on 32 bit powerpc:

  perf probe -x /lib/libc.so.6 malloc
  perf record -e probe_libc:malloc -a ls perf.data
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.036 MB perf.data (13829241621624967218 samples) ]

Actually, "perf script" just shows 21 samples. The number of samples is also
absurd since samples is long type, but it is printed as PRIu64.

Build test ran on x86-64, x86, aarch64, arm, mips, ppc and ppc64.

Signed-off-by: Yang Shi <yang.shi@linaro.org>
Cc: linaro-kernel@lists.linaro.org
Link: http://lkml.kernel.org/r/1443563383-4064-1-git-send-email-yang.shi@linaro.org
[ Bumped the 'hits' var used together with record.samples to 'unsigned long long' too ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf probe: Allow probing on kmodules without dwarf

Allow probing on kernel modules when 'perf' is built without debuginfo
support.

Currently perf-probe --module requires linking with libdw, but this
doesn't make sense.

E.g.
  ----
  # make NO_DWARF=1
  # ./perf probe -m pcspkr pcspkr_event%return
    Error: unknown switch `m'
  ----

With this patch
  ----
  # ./perf probe -m pcspkr pcspkr_event%return
  Added new event:
    probe:pcspkr_event   (on pcspkr_event%return in pcspkr)

  You can now use it in all perf tools, such as:

          perf record -e probe:pcspkr_event -aR sleep 1
  ----

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20151002125832.18617.78721.stgit@localhost.localdomain
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf list: Honour 'event_glob' whem printing selectable PMUs

Some PMUs, like the 'intel_bts' one can be used as an event name, i.e.:

$ perf record -e intel_bts:// usleep 1

Is a valid event name.

But the code printing such PMUs was not honouring the 'event_glob'
parameter, so the following line was always appearing:

  $ intel_bts//                                        [Kernel PMU event]

Fix it:

  $ [acme@felicio linux]$ perf list data

  List of pre-defined events (to be used in -e):

    uncore_imc/data_reads/                             [Kernel PMU event]
    uncore_imc/data_writes/                            [Kernel PMU event]

  $

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-ajb71858n7q7ao77b8pyy74w@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf list: Do event name substring search as last resort when no events found

Before:

  # perf list _alloc_ | head -10
  #

After:

  # perf list _alloc_ | head -10
    ext4:ext4_alloc_da_blocks                          [Tracepoint event]
    ext4:ext4_get_implied_cluster_alloc_exit           [Tracepoint event]
    kmem:kmem_cache_alloc_node                         [Tracepoint event]
    kmem:mm_page_alloc_extfrag                         [Tracepoint event]
    kmem:mm_page_alloc_zone_locked                     [Tracepoint event]
    xen:xen_mmu_alloc_ptpage                           [Tracepoint event]
  #

And it works for all types of events:

  # perf list br

  List of pre-defined events (to be used in -e):

    branch-instructions OR branches                    [Hardware event]
    branch-misses                                      [Hardware event]

    branch-load-misses                                 [Hardware cache event]
    branch-loads                                       [Hardware cache event]

    branch-instructions OR cpu/branch-instructions/    [Kernel PMU event]
    branch-misses OR cpu/branch-misses/                [Kernel PMU event]

    filelock:break_lease_block                         [Tracepoint event]
    filelock:break_lease_noblock                       [Tracepoint event]
    filelock:break_lease_unblock                       [Tracepoint event]
    syscalls:sys_enter_brk                             [Tracepoint event]
    syscalls:sys_exit_brk                              [Tracepoint event]

  #

Suggested-by: Ingo Molnar <mingo@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-qieivl18jdemoaghgndj36e6@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf callchain: Allow for max_stack greater than PERF_MAX_STACK_DEPTH

Adjust the validation to allow for max_stack greater than
PERF_MAX_STACK_DEPTH.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-18-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Fix a bug on "--call-graph none" option

The patch f9db0d0f1b2c ("perf callchain: Allow disabling call graphs
per event") added an ability to enable/disable callchain recording per
event.  But it had a problem when the enablement setting is changed at
'perf report' time using -g/--call-graph option.

For example, the following scenario will get a segfault.

  $ perf record -ag sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.500 MB perf.data (2555 samples) ]

  $ perf report -g none
  perf: Segmentation fault
  -------- backtrace --------
  perf[0x53a98a]
  /usr/lib/libc.so.6(+0x335af)[0x7f4e91df95af]

This is because callchain_param.sort() callback was not set but it
tried to call the function as it had the PERF_SAMPLE_CALLCHAIN bit.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Fixes: f9db0d0f1b2c ("perf callchain: Allow disabling call graphs per event")
Link: http://lkml.kernel.org/r/1443587640-24242-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Register idle thread

The perf top didn't add the idle/swapper thread to the machine's thread
list and its comm was displayed as ':0'.  Fix it.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443577526-3240-3-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Fix unresolved comm when -s comm is used

The perf top uses 'dso,symbol' sort keys by default so it overlooked a
problem in task's comm resolving.  When the sort key contains 'comm',
some task's comm is not shown properly.  This is because the
perf_top__mmap_read_idx() checks the cpumode value improperly.

The cpumode value of non-sample events are 0 (PERF_RECORD_MISC_CPUMODE_
UNKNOWN) so the events will be ignored by the switch statement.  This patch
allows it for non-sample events.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443577526-3240-2-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Allocate area for sample_id_hdr in a synthesized comm event

A previous patch added a synthesized comm event for forked child process
but it missed that the event should contain area for sample_id_hdr at
the end.  It worked by accident since the perf_event union contains
bigger event structs like mmap_events.

This patch fixes it by dynamically allocating event struct including
those area like in perf_event__synthesize_thread_map().

Reported-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443577526-3240-1-git-send-email-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/x86/intel/uncore: Do not use macro DEFINE_PCI_DEVICE_TABLE()

The DEFINE_PCI_DEVICE_TABLE() macro is deprecated. Use
'struct pci_device_id' instead of DEFINE_PCI_DEVICE_TABLE(),
with the goal of getting rid of this macro completely.

This Coccinelle semantic patch performs this transformation:

@@
identifier a;
declarer name DEFINE_PCI_DEVICE_TABLE;
initializer i;
@@
- DEFINE_PCI_DEVICE_TABLE(a)
+ const struct pci_device_id a[] = i;

Signed-off-by: Vaishali Thakkar <vthakkar1994@gmail.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/20151001085201.GA16939@localhost
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge tag 'perf-core-for-mingo' of git://git./linux/kernel/git/acme/linux into perf/core

Pull perf/core improvements and fixes:

User visible changes:

  - By default use the most precise "cycles" hw counter available, i.e.
    when the user doesn't specify any event, it will try using cycles:ppp,
    cycles:pp, etc. (Arnaldo Carvalho de Melo)

  - Remove blank lines, headers when piping output in 'perf list', so that it can
    be sanely used with 'wc -l', etc. (Arnaldo Carvalho de Melo)

  - Amend documentation about max_stack and synthesized callchains. (Adrian Hunter)

  - Fix 'perf probe -l' for probes added to kernel module functions. (Masami Hiramatsu)

Build fixes:

  - Fix shadowed declarations that break the build on older distros. (Jiri Olsa)

  - Fix build break on powerpc due to sample_reg_masks. (Sukadev Bhattiprolu)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: By default use the most precise "cycles" hw counter available

If the user doesn't specify any event, try the most precise "cycles"
available, i.e. start by "cycles:ppp" and go on removing "p" till it
works.

E.g.

  $ perf record usleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.017 MB perf.data (11 samples) ]
  $ perf evlist
  cycles:pp
  $ perf evlist -v
  cycles:pp: size: 112, { sample_period, sample_freq }: 4000, sample_type:
  IP|TID|TIME|PERIOD, disabled: 1, inherit: 1, mmap: 1, comm: 1, freq: 1,
  enable_on_exec: 1, task: 1, precise_ip: 2, sample_id_all: 1,
  exclude_guest: 1, mmap2: 1, comm_exec: 1
  $ grep 'model name' /proc/cpuinfo | head -1
  model name : Intel(R) Core(TM) i7-3667U CPU @ 2.00GHz
  $

When 'cycles' appears explicitely is specified this will not be tried,
i.e. the user has full control of the level of precision to be used:

  $ perf record -e cycles usleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.016 MB perf.data (9 samples) ]
  $ perf evlist
  cycles
  $ perf evlist -v
  cycles: size: 112, { sample_period, sample_freq }: 4000, sample_type:
  IP|TID|TIME|PERIOD, disabled: 1, inherit: 1, mmap: 1, comm: 1, freq: 1,
  enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2:
  1, comm_exec: 1
  $

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Chandler Carruth <chandlerc@gmail.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://www.youtube.com/watch?v=nXaxk27zwlk
Link: http://lkml.kernel.org/n/tip-b1ywebmt22pi78vjxau01wth@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf list: Remove blank lines, headers when piping output

So that one can, for instance, use it with wc -l:

  # perf list *:*write* | wc -l
  60

Or to look for the "bio" tracepoints, without 'perf list' headers:

  # perf list *:*bio* | head
    block:block_bio_backmerge                          [Tracepoint event]
    block:block_bio_bounce                             [Tracepoint event]
    block:block_bio_complete                           [Tracepoint event]
    block:block_bio_frontmerge                         [Tracepoint event]
    block:block_bio_queue                              [Tracepoint event]
    block:block_bio_remap                              [Tracepoint event]
  #

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-ts7sc0x8u4io4cifzkup4j44@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf probe: Improve error message when %return is on inlined function

perf probe shows more precisely message when it finds given
%return target function is inlined.

Without this fix:
  ----
  # ./perf probe -V getname_flags%return
  Return probe must be on the head of a real function.
  Debuginfo analysis failed.
    Error: Failed to show vars.
  ----

With this fix:
  ----
  # ./perf probe -V getname_flags%return
  Failed to find "getname_flags%return",
   because getname_flags is an inlined function and has no return point.
  Debuginfo analysis failed.
    Error: Failed to show vars.
  ----

Suggested-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20150930164137.3733.55055.stgit@localhost.localdomain
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf probe: Fix a segfault bug in debuginfo_cache

perf probe --list will get a segfault if the first kprobe event is on a
module and the second or latter one is on the kernel.

e.g.
  ----
  # ./perf probe -q -m pcspkr pcspkr_event
  # ./perf probe -q vfs_read
  # ./perf probe -l
  Segmentation fault (core dumped)
  ----

This is because the debuginfo_cache fails to handle NULL module name,
which causes segfault on strcmp. (Note that strcmp("something", NULL)
always causes segfault)

To fix this debuginfo_cache__open always translates the NULL module name
to "kernel" (this is correct, because NULL module name means opening the
debuginfo for the kernel)

  ----
  # ./perf probe -l
    probe:pcspkr_event   (on pcspkr_event@drivers/input/misc/pcspkr.c
    in pcspkr)
    probe:vfs_read       (on vfs_read@ksrc/linux-3/fs/read_write.c)
  ----

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20150930164135.3733.23993.stgit@localhost.localdomain
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf probe: Show correct source lines of probes on kmodules

Perf probe always failed to find appropriate line numbers because of
failing to find .text start address offset from debuginfo.

e.g.
  ----
  # ./perf probe -m pcspkr pcspkr_event:5
  Added new events:
    probe:pcspkr_event   (on pcspkr_event:5 in pcspkr)
    probe:pcspkr_event_1 (on pcspkr_event:5 in pcspkr)

  You can now use it in all perf tools, such as:

          perf record -e probe:pcspkr_event_1 -aR sleep 1

  # ./perf probe -l
  Failed to find debug information for address ffffffffa031f006
  Failed to find debug information for address ffffffffa031f016
    probe:pcspkr_event   (on pcspkr_event+6 in pcspkr)
    probe:pcspkr_event_1 (on pcspkr_event+22 in pcspkr)
  ----

This fixes the above issue as below.
1. Get the relative address of the symbol in .text by using
   map->start.
2. Adjust the address by adding the offset of .text section
   in the kernel module binary.

With this fix, perf probe -l shows lines correctly.
  ----
  # ./perf probe -l
    probe:pcspkr_event   (on pcspkr_event:5@drivers/input/misc/pcspkr.c in pcspkr)
    probe:pcspkr_event_1 (on pcspkr_event:5@drivers/input/misc/pcspkr.c in pcspkr)
  ----

Reported-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20150930164132.3733.24643.stgit@localhost.localdomain
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf probe: Begin and end libdwfl report session correctly

Fix a trival bug about libdwfl usage of the report session, it should
explicitly begin and end a report session around dwfl_report_offline().

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20150930164128.3733.59876.stgit@localhost.localdomain
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf probe: Fix to remove dot suffix from second or latter events

Fix to remove dot suffix (e.g. .const, .isra) from the second or latter
events which has suffix numbers.

Since the previous commit 35a23ff928b0 ("perf probe: Cut off the gcc
optimization postfixes from function name") didn't care about the suffix
numbered events, therefore we'll have an error when we add additional
events on the same dot suffix functions.

e.g.
  ----
  # ./perf probe -f -a get_sigframe.isra.2.constprop.3 \
   -a get_sigframe.isra.2.constprop.3
  Failed to write event: Invalid argument
    Error: Failed to add events.
  ----

This fixes above issue as below:
  ----
  # ./perf probe -f -a get_sigframe.isra.2.constprop.3 \
   -a get_sigframe.isra.2.constprop.3
  Added new events:
    probe:get_sigframe   (on get_sigframe.isra.2.constprop.3)
    probe:get_sigframe_1 (on get_sigframe.isra.2.constprop.3)

  You can now use it in all perf tools, such as:

          perf record -e probe:get_sigframe_1 -aR sleep 1

  ----

Signed-off-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20150930164130.3733.26573.stgit@localhost.localdomain
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib symbol: Introduce kallsyms2elf_type

Map 't', 'T' (text, local, global), 'w' and 'W' (weak text, local,
global) as STT_FUNC, and the rest as STT_OBJECT

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-sbwcixulpc5v1xuxn3xvm0nn@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib symbol: Rename kallsyms2elf_type to kallsyms2elf_binding

It is about binding, not type, we have just a letter in kallsyms that
should map both for the ELF type (STT_FUNC, etc) and to the ELF
symbol binding (STB_WEAK, STB_GLOBAL, etc), so rename it now before
introducing kallsyms2_elf_type()

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-uu5vj343ms1q2wm55690on6v@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf machine: Add method for common kernel_map(FUNCTION) operation

And it is also a step in the direction of killing the separation of data
and text maps in map_groups.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-rrds86kb3wx5wk8v38v56gw8@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf machine: Use machine__kernel_map() thoroughly

In places where we were using its open coded equivalent.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-khkdugcdoqy3tkszm3jdxgbe@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix build break on powerpc due to sample_reg_masks

The perf_regs.c file does not get built on Powerpc as CONFIG_PERF_REGS
is false.  So the weak definition for 'sample_regs_masks' doesn't get
picked up.

Adding perf_regs.o to util/Build unconditionally, exposes a redefinition
error for 'perf_reg_value()' function (due to the static inline version
in util/perf_regs.h). So use #ifdef HAVE_PERF_REGS_SUPPORT' around that
function.

Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Dominik Dingel <dingel@linux.vnet.ibm.com>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Stephane Eranian <eranian@google.com>
Cc: linuxppc-dev@ozlabs.org
Link: http://lkml.kernel.org/r/20150930182836.GA27858@us.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Amend documentation about max_stack and synthesized callchains

The --max_stack option was added as an optimization to reduce processing time,
so people specifying --max-stack might get a increased processing time if
combined with synthesized callchains, but otherwise no real harm.

A warning about setting both --max_stack and the synthesized callchains max
depth seems like overkill.  Amend the documentation.

Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/560A5155.4060105@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf maps: Introduce maps__find_symbol_by_name()

Out of map_groups__find_symbol_by_name(), so that we can turn this later
one first into a call to maps__find_symbol_by_name(MAP__FUNCTION) +
MAP__VARIABLE, and then to just one call, we'll merge MAP__FUNCTION with
MAP__VARIABLE maps, to simplify the code.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-pvkar0jacqn92g148u9sqttt@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix shadowed declaration in parse-events.c

The error variable breaks build on CentOS 6.7, due to a collision with a
global error symbol:

    CC       util/parse-events.o
  cc1: warnings being treated as errors
  util/parse-events.c:419: error: declaration of ‘error’ shadows a global
  declaration
  util/util.h:135: error: shadowed declaration is here
  util/parse-events.c: In function ‘add_tracepoint_multi_event’:
  ...

Using different argument names instead to fix it.

Reported-by: Vinson Lee <vlee@twopensource.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: linux-tip-commits@vger.kernel.org
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Raphael Beamonte <raphael.beamonte@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150929150531.GI27383@krava.redhat.com
[ Fix one more case, at line 770 ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools: Fix shadowed declaration in err.h

The error variable breaks build on CentOS 6.7, due to collision with
global error symbol:

    CC       util/evlist.o
  cc1: warnings being treated as errors
  In file included from util/evlist.c:28:
  tools/include/linux/err.h: In function ‘ERR_PTR’:
  tools/include/linux/err.h:34: error: declaration of ‘error’ shadows a global declaration
  util/util.h:135: error: shadowed declaration is here

Using 'error_' name instead to fix it.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/n/tip-i9mdgdbrgauy3fe76s9rd125@git.kernel.org
Reported-by: Vinson Lee <vlee@twopensource.com>
[ Use 'error_' instead of 'err' to, visually, not diverge too much from include/linux/err.h ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

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:

  - Accept a zero --itrace period, meaning "as often as possible".  In the case
    of Intel PT that is the same as a period of 1 and a unit of 'instructions'
    (i.e.  --itrace=i1i). (Adrian Hunter)

  - Harmonize itrace's synthesized callchains with the existing --max-stack
    tool option. (Adrian Hunter)

  - Allow time to be displayed in nanoseconds in 'perf script'. (Adrian Hunter)

  - Fix potential infinite loop when handling Intel PT timestamps. (Adrian Hunter)

  - Slighly improve Intel PT debug logging. (Adrian Hunter)

  - Warn when AUX data has been lost, just like when processing PERF_RECORD_LOST.
    (Adrian Hunter)

  - Further document export-to-postgresql.py script. (Adrian Hunter)

  - Add option to synthesize branch stack from auxtrace data. (Adrian Hunter)

  - Use equivalent logic to avoid using dso->kernel. (Arnaldo Carvalho de Melo)

  - Show proper error messages when parsing bad terms for hw/sw events. (He Kuang)

  - Tracepoint event parsing improvements. (He Kuang)

  - Store tracing mountpoint for better error message. (Jiri Olsa)

  - Add fixdep to tools/build, bringing it closer to the kernel counterpart, from
    where it is being lifted. (Jiri Olsa)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf tools: Enable event_config terms to tracepoint events

This patch enables config terms for tracepoint perf events. Valid terms
for tracepoint events are 'call-graph' and 'stack-size', so we can use
different callgraph settings for each event and eliminate unnecessary
overhead.

Here is an example for using different call-graph config for each
tracepoint.

  $ perf record -e syscalls:sys_enter_write/call-graph=fp/
                -e syscalls:sys_exit_write/call-graph=no/
                dd if=/dev/zero of=test bs=4k count=10

  $ perf report --stdio

  #
  # Total Lost Samples: 0
  #
  # Samples: 13  of event 'syscalls:sys_enter_write'
  # Event count (approx.): 13
  #
  # Children      Self  Command  Shared Object       Symbol
  # ........  ........  .......  ..................  ......................
  #
      76.92%    76.92%  dd       libpthread-2.20.so  [.] __write_nocancel
                   |
                   ---__write_nocancel

      23.08%    23.08%  dd       libc-2.20.so        [.] write
                   |
                   ---write
                      |
                      |--33.33%-- 0x2031342820736574
                      |
                      |--33.33%-- 0xa6e69207364726f
                      |
                       --33.33%-- 0x34202c7320393039
  ...

  # Samples: 13  of event 'syscalls:sys_exit_write'
  # Event count (approx.): 13
  #
  # Children      Self  Command  Shared Object       Symbol
  # ........  ........  .......  ..................  ......................
  #
      76.92%    76.92%  dd       libpthread-2.20.so  [.] __write_nocancel
      23.08%    23.08%  dd       libc-2.20.so        [.] write
       7.69%     0.00%  dd       [unknown]           [.] 0x0a6e69207364726f
       7.69%     0.00%  dd       [unknown]           [.] 0x2031342820736574
       7.69%     0.00%  dd       [unknown]           [.] 0x34202c7320393039

Signed-off-by: He Kuang <hekuang@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1443412336-120050-4-git-send-email-hekuang@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Adds the tracepoint name parsing support

Adds rules for parsing tracepoint names. Change rules of tracepoint which
derives from PE_NAMEs into tracepoint names directly, so adding more rules
based on tracepoint names will be easier.

Changes v2-v3:
   - Change __event_legacy_tracepoint label in bison file to tracepoint_name
   - Fix formats error.

Signed-off-by: He Kuang <hekuang@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1443412336-120050-3-git-send-email-hekuang@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Show proper error message for wrong terms of hw/sw events

Show proper error message and show valid terms when wrong config terms
is specified for hw/sw type perf events.

This patch makes the original error format function formats_error_string()
more generic, which only outputs the static config terms for hw/sw perf
events, and prepends pmu formats for pmu events.

Before this patch:

  $ perf record -e 'cpu-clock/freqx=200/' -a sleep 1
  invalid or unsupported event: 'cpu-clock/freqx=200/'
  Run 'perf list' for a list of valid events

   usage: perf record [<options>] [<command>]
      or: perf record [<options>] -- <command> [<options>]

      -e, --event <event>   event selector. use 'perf list' to list available events

After this patch:

  $ perf record -e 'cpu-clock/freqx=200/' -a sleep 1
  event syntax error: 'cpu-clock/freqx=200/'
                                 \___ unknown term

  valid terms: config,config1,config2,name,period,freq,branch_type,time,call-graph,stack-size

  Run 'perf list' for a list of valid events

   usage: perf record [<options>] [<command>]
      or: perf record [<options>] -- <command> [<options>]

      -e, --event <event>   event selector. use 'perf list' to list available events

Signed-off-by: He Kuang <hekuang@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1443412336-120050-2-git-send-email-hekuang@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Adds the config_term callback for different type events

Currently, function config_term() is used for checking config terms of
all types of events, while unknown terms is not reported as an error
because pmu events have valid terms in sysfs.

But this is wrong when unknown terms are specificed to hw/sw events.
This patch Adds the config_term callback so we can use separate check
routines for each type of events.

Signed-off-by: He Kuang <hekuang@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1443412336-120050-1-git-send-email-hekuang@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Add mispred-all config option to aid use with autofdo

autofdo incorrectly expects branch flags to include either mispred or
predicted.  In fact mispred = predicted = 0 is valid and means the flags
are not supported, which they aren't by Intel PT.

To make autofdo work, add a config option which will cause Intel PT
decoder to set the mispred flag on all branches.

Below is an example of using Intel PT with autofdo.  The example is
also added to the Intel PT documentation.  It requires autofdo
(https://github.com/google/autofdo) and gcc version 5.  The bubble
sort example is from the AutoFDO tutorial (https://gcc.gnu.org/wiki/AutoFDO/Tutorial)
amended to take the number of elements as a parameter.

$ gcc-5 -O3 sort.c -o sort_optimized
$ ./sort_optimized 30000
Bubble sorting array of 30000 elements
2254 ms

$ cat ~/.perfconfig
[intel-pt]
mispred-all

$ perf record -e intel_pt//u ./sort 3000
Bubble sorting array of 3000 elements
58 ms
[ perf record: Woken up 2 times to write data ]
[ perf record: Captured and wrote 3.939 MB perf.data ]
$ perf inject -i perf.data -o inj --itrace=i100usle --strip
$ ./create_gcov --binary=./sort --profile=inj --gcov=sort.gcov -gcov_version=1
$ gcc-5 -O3 -fauto-profile=sort.gcov sort.c -o sort_autofdo
$ ./sort_autofdo 30000
Bubble sorting array of 30000 elements
2155 ms

Note there is currently no advantage to using Intel PT instead of LBR,
but that may change in the future if greater use is made of the data.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-26-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf inject: Add --strip option to strip out non-synthesized events

Add a new option --strip which is used with --itrace to strip out
non-synthesized events.  This results in a perf.data file that is
simpler for external tools to parse.  In particular, this can be used to
prepare a perf.data file for consumption by autofdo.

A subsequent patch makes a change to Intel PT also to enable use with
autofdo and gives an example of that use.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-25-git-send-email-adrian.hunter@intel.com
[ Made it use perf_evlist__remove() + perf_evsel__delete() ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf inject: Remove more aux-related stuff when processing instruction traces

perf inject can process instruction traces (using the --itrace option)
which removes aux-related events and replaces them with the requested
synthesized events.

However there are still some leftovers, namely PERF_RECORD_ITRACE_START
events and the original evsel (selected event) e.g. intel_pt//

For the sake of completeness, remove them too.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-24-git-send-email-adrian.hunter@intel.com
[ Made it use perf_evlist__remove() + perf_evsel__delete() ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Add perf_evlist__remove()

Add a counterpart to perf_evlist__add() that does the opposite and
deletes the evsel.

This will be used by perf inject to remove unwanted evsels.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-23-git-send-email-adrian.hunter@intel.com
[ Renamed it from perf_evlist__del() to perf_evlist__remove() and removed the perf_evsel__delete() call ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Add perf_evlist__id2evsel_strict()

perf_evlist__id2evsel_strict() is the same as perf_evlist__id2evsel()
except that it ensures that the id must match.

This will be used by perf inject to find a specific evsel that is to be
deleted, hence the need to match exactly.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-22-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf script: Make scripting_max_stack value allow for synthesized callchains

perf script has a setting to set the maximum stack depth when processing
callchains.  The setting defaults to the hard-coded maximum definition
PERF_MAX_STACK_DEPTH which is 127.

It is possible, when processing instruction traces, to synthesize
callchains.  Synthesized callchains do not have the kernel size
limitation and are whatever size the user requests, although validation
presently prevents the user requested a value greater that 1024.  The
default value is 16.

To allow for synthesized callchains, make the scripting_max_stack value
at least the same size as the synthesized callchain size.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-21-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf scripting python: Allow for max_stack greater than PERF_MAX_STACK_DEPTH

Use the scripting_max_stack value to allow for values greater than
PERF_MAX_STACK_DEPTH.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-20-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf script: Add a setting for maximum stack depth

Add a setting for maximum stack depth in preparation for allowing for
synthesized callchains.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-19-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists: Allow for max_stack greater than PERF_MAX_STACK_DEPTH

Use the max_stack value instead of PERF_MAX_STACK_DEPTH so that
arbitrary-sized callchains can be supported.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-17-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Make max_stack value allow for synthesized callchains

perf report has an option (--max-stack) to set the maximum stack depth
when processing callchains.  The option defaults to the hard-coded
maximum definition PERF_MAX_STACK_DEPTH which is 127.  The intention of
the option is to allow the user to reduce the processing time by
reducing the amount of the callchain that is processed.

It is also possible, when processing instruction traces, to synthesize
callchains.  Synthesized callchains do not have the kernel size
limitation and are whatever size the user requests, although validation
presently prevents the user requested a value greater that 1024.  The
default value is 16.

To allow for synthesized callchains, make the max_stack value at least
the same size as the synthesized callchain size.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-16-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Support generating branch stack

Add support for generating branch stack context for PT samples.  The
decoder reports a configurable number of branches as branch context for
each sample. Internally it keeps track of them by using a simple sliding
window.  We also flush the last branch buffer on each sample to avoid
overlapping intervals.

This is useful for:

- Reporting accurate basic block edge frequencies through the perf
  report branch view
- Using with --branch-history to get the wider context of samples
- Other users of LBRs

Also the Documentation is updated.

Examples:

Record with Intel PT:

perf record -e intel_pt//u ls

Branch stacks are used by default if synthesized so:

perf report --itrace=ile

is the same as:

perf report --itrace=ile -b

Branch history can be requested also:

perf report --itrace=igle --branch-history

Based-on-patch-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-15-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Move branch filter logic

intel_pt_synth_branch_sample() skips synthesizing if the branch does not
match the branch filter.  That logic was sitting in the middle of the
function but is more efficiently placed at the start of the function, so
move it.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-14-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf inject: Set branch stack feature flag when synthesizing branch stacks

The branch stack feature flag is set by 'perf record' when recording
data that contains branch stacks.  Consequently, when 'perf inject'
synthesizes branch stacks, the feature flag should be set also.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-13-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Skip events with null branch stacks

A non-synthesized event might not have a branch stack if branch stacks
have been synthesized (using itrace options).

An example of that is when Intel PT records sched_switch events for
decoding purposes.  Those sched_switch events do not have branch stacks
even though the Intel PT decoder may be synthesizing other events that
do due to the itrace options.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-12-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Also do default setup for synthesized branch stacks

The 'perf report' tool will default to displaying branch stacks (-b
option) if they are present.  Make that also happen for synthesized
branch stacks.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-11-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Adjust sample type validation for synthesized branch stacks

perf report looks at event sample types to determine if branch stacks
have been sampled.  Adjust the validation to know about instruction
tracing options.

This change allows the use of the -b option which otherwise would
complain with an error like:

Error:
Selected -b but no branch data. Did you call perf record without -b?
# To display the perf.data header info,
# please use --header/--header-only options.
#

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-10-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf auxtrace: Add option to synthesize branch stacks on samples

Add AUX area tracing option 'l' to synthesize branch stacks on samples
just like sample type PERF_SAMPLE_BRANCH_STACK.  This is taken into use
by Intel PT in a subsequent patch.

Based-on-patch-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-9-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add more documentation to export-to-postgresql.py script

Add some comments to the script and some 'views' to the created database
that better illustrate the database structure and how it can be used.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-8-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf session: Warn when AUX data has been lost

By default 'perf record' will postprocess the perf.data file to
determine build-ids.  When that happens, the number of lost perf events
is displayed.

Make that also happen for AUX events.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-7-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf script: Allow time to be displayed in nanoseconds

Add option --ns to display time to 9 decimal places.  That is useful in
some cases, for example when using Intel PT cycle accurate mode.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-6-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Make logging slightly more efficient

Logging is only used for debugging. Use macros to save calling into the
functions only to return immediately when logging is not enabled.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-5-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Fix potential loop forever

TSC packets contain only 7 bytes of TSC.  The 8th byte is assumed to
change so infrequently that its value can be inferred.  However the
logic must cater for a 7 byte wraparound, which it does by adding 1 to
the top byte.

The existing code was doing that with a while loop even though the
addition should only need to be done once.  That logic won't work (will
loop forever) if TSC wraps around at the 8th byte.  Theoretically that
would take at least 10 years, unless something else went wrong.

And what else could go wrong.  Well, if the chunks of trace data are
processed out of order, it will make it look like the 7-byte TSC has
gone backwards (i.e. wrapped).  If that happens 256 times then stuck in
the while loop it will be.

Fix that by getting rid of the unnecessary while loop.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-4-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Fix sample type validation for synthesized callchains

Processing instruction tracing data (e.g. Intel PT) can synthesize
callchains e.g.

$ perf record -e intel_pt//u uname
$ perf report --stdio --itrace=ige

However perf report's callgraph option gets extra validation, so:

$ perf report --stdio --itrace=ige -gflat
Error:
Selected -g or --branch-history but no callchain data. Did
you call 'perf record' without -g?
# To display the perf.data header info,
# please use --header/--header-only options.
#

Fix the validation to know about instruction tracing options so
above command works.

A side-effect of the change is that the default option to
accumulate the callchain of child functions comes into force.
To get the previous behaviour the --no-children option can be
used e.g.

       $ perf report --stdio --itrace=ige -gflat --no-children

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-3-git-send-email-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf auxtrace: Fix 'instructions' period of zero

Instruction tracing options (i.e. --itrace) include an option for
sampling instructions at an arbitrary period. e.g.

--itrace=i10us

means make an 'instructions' sample for every 10us of trace.

Currently the logic does not distinguish between a period of
zero and no period being specified at all, so it gets treated
as the default period which is 100000.  That doesn't really
make sense.

Fix it so that zero period is accepted and treated as meaning
"as often as possible".

In the case of Intel PT that is the same as a period of 1 and
a unit of 'instructions' (i.e. --itrace=i1i).

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/1443186956-18718-2-git-send-email-adrian.hunter@intel.com
[ Add a few lines describing this in the Documentation/intel-pt.txt file ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build: Build fixdep helper from perf and basic libs

Adding the fixdep target into the Makefile.include to ease up building of
fixdep helper, that needs to be built before we dive in to the build itself.
The user can invoke the fixdep target to build the helper.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443004442-32660-8-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Rename the 'single_dep' target to 'prepare'

And use the new 'prepare' target for the $(PERF_IN) target.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443004442-32660-7-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build: Make the fixdep helper part of the build process

Making the fixdep helper to be invoked within dep-cmd.

Each user of the build framework needs to make sure fixdep exists before
executing the build itself.

If the build doesn't find fixdep, it falls back to the old style
dependency tracking.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443004442-32660-6-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build: Move dependency copy into function

So it's easier to add more functionality in the following commit.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443004442-32660-5-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build: Add fixdep dependency helper

For dependency tracking we currently use targets that fall out of the
gcc -MD command. We store this info in the .cmd file and include as
makefile during the build.

This format put object as target and all the c and header files as
dependencies, like:

  util/abspath.o: util/abspath.c /usr/include/stdc-predef.h util/cache.h \
   /usr/include/bits/wordsize.h /usr/include/gnu/stubs.h \
   ...

If any of those dependency header files (krava.h below) is removed the
build fails on:

  make[1]: *** No rule to make target 'krava.h', needed by 'inc.o'.  Stop.

This patch adds fixdep helper, that is used by kbuild to alter the shape
of the object dependencies like:

  source_util/abspath.o := util/abspath.c

  deps_util/abspath.o := \
    /usr/include/stdc-predef.h \
    util/cache.h \
    ...

  util/abspath.o: $(deps_util/abspath.o)

  $(deps_util/abspath.o):

With this format the header removal won't make the build fail, because
it'll be picked up by the last empty target defined for each header.

As previously mentioned the fixdep tool is taken from kbuild. It's not
complete backport, only the part that alters the standard dependency
info was taken, the part that adds the CONFIG_* dependency logic will be
probably taken later on.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Kai Germaschewski <kai.germaschewski@gmx.de>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443004442-32660-4-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build: Add test for missing include

The current build framework fails to cope with header file removal. The
reason is that the removed header file stays in the .cmd file target
rule and forces the build to fail.

This issue is fixed and explained in the following patches.

Adding a new build test that simulates header removal.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443004442-32660-3-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build: Add Makefile.include

To ease up build framework code setup for users.

More shared code will be added in the following patches.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1443004442-32660-2-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib api fs: Store tracing mountpoint for better error message

Storing the actual tracing path mountpoint to display correct
error message hint ('Hint:' line). The error hint rediscovers
mountpoints, but it could be different from what we actually
used in tracing path.

Before we'd display debugfs mount even though tracefs was used:
  $ perf record -e sched:sched_krava ls
  event syntax error: 'sched:sched_krava'
                       \___ can't access trace events

  Error:  No permissions to read /sys/kernel/debug/tracing/events/sched/sched_krava
  Hint:   Try 'sudo mount -o remount,mode=755 /sys/kernel/debug'
  ...

After this change, correct mountpoint is displayed:
  $ perf record -e sched:sched_krava ls
  event syntax error: 'sched:sched_krava'
                       \___ can't access trace events

  Error:  No permissions to read /sys/kernel/debug/tracing/events/sched/sched_krava
  Hint:   Try 'sudo mount -o remount,mode=755 /sys/kernel/debug/tracing'
  ...

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Raphael Beamonte <raphael.beamonte@gmail.com>
Link: http://lkml.kernel.org/r/1442674027-19427-1-git-send-email-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Use __map__is_kernel() when synthesizing kernel module mmap records

Equivalent and removes one more case of using dso->kernel.

  # perf record -a usleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.768 MB perf.data (30 samples) ]

Before:

  [root@zoo ~]# perf script --show-task --show-mmap | head -3
   swapper 0 [0] 0.0: PERF_RECORD_MMAP -1/0: [0xffffffff81000000(0x1f000000) @ 0xffffffff81000000]: x [kernel.kallsyms]_text
   swapper 0 [0] 0.0: PERF_RECORD_MMAP -1/0: [0xffffffffa0000000(0xa000) @ 0]: x /lib/modules/4.3.0-rc1+/kernel/drivers/acpi/video.ko
   swapper 0 [0] 0.0: PERF_RECORD_MMAP -1/0: [0xffffffffa000a000(0x5000) @ 0]: x /lib/modules/4.3.0-rc1+/kernel/drivers/i2c/algos/i2c-algo-bit.ko
  #

  # perf script --show-task --show-mmap | head -3
   swapper 0 [0] 0.0: PERF_RECORD_MMAP -1/0: [0xffffffff81000000(0x1f000000) @ 0xffffffff81000000]: x [kernel.kallsyms]_text
   swapper 0 [0] 0.0: PERF_RECORD_MMAP -1/0: [0xffffffffa0000000(0xa000) @ 0]: x /lib/modules/4.3.0-rc1+/kernel/drivers/acpi/video.ko
   swapper 0 [0] 0.0: PERF_RECORD_MMAP -1/0: [0xffffffffa000a000(0x5000) @ 0]: x /lib/modules/4.3.0-rc1+/kernel/drivers/i2c/algos/i2c-algo-bit.ko
  #

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-b65xe578dwq22mzmmj5y94wr@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf hists browser: Use the map to determine if a DSO is being used as a kernel

The map is what should say if an ELF (or some other format) image is
being used for some particular purpose, as a kernel, host or guest.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-zufousvfar0710p4qj71c32d@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Filter symbols based on __map__is_kernel(map)

Instead of using dso->kernel, this is equivalent at the moment,
and helps in reducing the accesses to dso->kernel.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-1pc2v63iphtifovw3bv0bo1v@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf/core, perf/x86: Change needlessly global functions and a variable to static

Fixes various sparse warnings.

Signed-off-by: Geliang Tang <geliangtang@163.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/70c14234da1bed6e3e67b9c419e2d5e376ab4f32.1443367286.git.geliangtang@163.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge branch 'linus' into perf/core, to pick up fixes before applying new changes

Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge branch 'upstream' of git://git.linux-mips.org/ralf/upstream-linus

Pull MIPS fixes from Ralf Baechle:
 - Properly setup irq handling for ATH79 platforms
 - Fix bootmem mapstart calculation for contiguous maps
 - Handle little endian and older CPUs correct in BPF
 - Fix console for Fulong 2E systems
 - Handle FTLB correctly on R6 CPUs
 - Fixes for CM, GIC and MAAR support code

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus:
  MIPS: Initialise MAARs on secondary CPUs
  MIPS: print MAAR configuration during boot
  MIPS: mm: compile maar_init unconditionally
  irqchip: mips-gic: Fix pending & mask reads for MIPS64 with 32b GIC.
  irqchip: mips-gic: Convert CPU numbers to VP IDs.
  MIPS: CM: Provide a function to map from CPU to VP ID.
  MIPS: Fix FTLB detection for R6
  MIPS: cpu-features: Add cpu_has_ftlb
  MIPS: ATH79: Add irq chip ar7240-misc-intc
  MIPS: ATH79: Set missing irq ack handler for ar7100-misc-intc irq chip
  MIPS: BPF: Fix build on pre-R2 little endian CPUs
  MIPS: BPF: Avoid unreachable code on little endian
  MIPS: bootmem: Fix mapstart calculation for contiguous maps
  MIPS: Fix console output for Fulong2e system

Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Thomas Gleixner:
 "Another pile of fixes for perf:

   - Plug overflows and races in the core code

   - Sanitize the flow of the perf syscall so we error out before
     handling the more complex and hard to undo setups

   - Improve and fix Broadwell and Skylake hardware support

   - Revert a fix which broke what it tried to fix in perf tools

   - A couple of smaller fixes in various places of perf tools"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf tools: Fix copying of /proc/kcore
  perf intel-pt: Remove no_force_psb from documentation
  perf probe: Use existing routine to look for a kernel module by dso->short_name
  perf/x86: Change test_aperfmperf() and test_intel() to static
  tools lib traceevent: Fix string handling in heterogeneous arch environments
  perf record: Avoid infinite loop at buildid processing with no samples
  perf: Fix races in computing the header sizes
  perf: Fix u16 overflows
  perf: Restructure perf syscall point of no return
  perf/x86/intel: Fix Skylake FRONTEND MSR extrareg mask
  perf/x86/intel/pebs: Add PEBS frontend profiling for Skylake
  perf/x86/intel: Make the CYCLE_ACTIVITY.* constraint on Broadwell more specific
  perf tools: Bool functions shouldn't return -1
  tools build: Add test for presence of __get_cpuid() gcc builtin
  tools build: Add test for presence of numa_num_possible_cpus() in libnuma
  Revert "perf symbols: Fix mismatched declarations for elf_getphdrnum"
  perf stat: Fix per-pkg event reporting bug

Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull scheduler fix from Thomas Gleixner:
 "A single bug fix for the scheduler to prevent dequeueing of the idle
  task when setting the cpus allowed mask"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched: Fix crash trying to dequeue/enqueue the idle thread

Merge branch 'locking-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull locking fix from Thomas Gleixner:
 "A single bugfix for lockdep to preserve the pinning counter when
  rebuilding the lock stack"

* 'locking-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/lockdep: Fix hlock->pin_count reset on lock stack rebuilds

MIPS: Initialise MAARs on secondary CPUs

MAARs should be initialised on each CPU (or rather, core) in the system
in order to achieve consistent behaviour & performance. Previously they
have only been initialised on the boot CPU which leads to performance
problems if tasks are later scheduled on a secondary CPU, particularly
if those tasks make use of unaligned vector accesses where some CPUs
don't handle any cases in hardware for non-speculative memory regions.
Fix this by recording the MAAR configuration from the boot CPU and
applying it to secondary CPUs as part of their bringup.

Reported-by: Doug Gilmore <doug.gilmore@imgtec.com>
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Steven J. Hill <Steven.Hill@imgtec.com>
Cc: Andrew Bresticker <abrestic@chromium.org>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: David Hildenbrand <dahi@linux.vnet.ibm.com>
Cc: linux-kernel@vger.kernel.org
Cc: Aaro Koskinen <aaro.koskinen@iki.fi>
Cc: James Hogan <james.hogan@imgtec.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Markos Chandras <markos.chandras@imgtec.com>
Cc: Hemmo Nieminen <hemmo.nieminen@iki.fi>
Cc: Alex Smith <alex.smith@imgtec.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Patchwork: https://patchwork.linux-mips.org/patch/11239/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>