perf vendor events intel: Fix Load_Miss_Real_Latency on SKL/SKX

Fix incorrect event names for the Load_Miss_Real_Latency metric for
Skylake and Skylake Server.

Fixes https://github.com/andikleen/pmu-tools/issues/158

Before:

  % perf stat -M Load_Miss_Real_Latency true
  event syntax error: '..ss.pending,mem_load_retired.l1_miss_ps,mem_load_retired.fb_hit_ps}:W'
                                    \___ parser error

   Usage: perf stat [<options>] [<command>]

      -M, --metrics <metric/metric group list>
                            monitor specified metrics or metric groups (separated by ,)

After:

  % perf stat -M Load_Miss_Real_Latency true

   Performance counter stats for 'true':

             279,204      l1d_pend_miss.pending     #     14.0 Load_Miss_Real_Latency
               4,784      mem_load_uops_retired.l1_miss
              15,188      mem_load_uops_retired.hit_lfb

         0.000899640 seconds time elapsed

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Link: http://lkml.kernel.org/r/20181120050635.4215-1-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf parse-events: Fix unchecked usage of strncpy()

The strncpy() function may leave the destination string buffer
unterminated, better use strlcpy() that we have a __weak fallback
implementation for systems without it.

This fixes this warning on an Alpine Linux Edge system with gcc 8.2:

  util/parse-events.c: In function 'print_symbol_events':
  util/parse-events.c:2465:4: error: 'strncpy' specified bound 100 equals destination size [-Werror=stringop-truncation]
      strncpy(name, syms->symbol, MAX_NAME_LEN);
      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  In function 'print_symbol_events.constprop',
      inlined from 'print_events' at util/parse-events.c:2508:2:
  util/parse-events.c:2465:4: error: 'strncpy' specified bound 100 equals destination size [-Werror=stringop-truncation]
      strncpy(name, syms->symbol, MAX_NAME_LEN);
      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  In function 'print_symbol_events.constprop',
      inlined from 'print_events' at util/parse-events.c:2511:2:
  util/parse-events.c:2465:4: error: 'strncpy' specified bound 100 equals destination size [-Werror=stringop-truncation]
      strncpy(name, syms->symbol, MAX_NAME_LEN);
      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  cc1: all warnings being treated as errors

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: 947b4ad1d198 ("perf list: Fix max event string size")
Link: https://lkml.kernel.org/n/tip-b663e33bm6x8hrkie4uxh7u2@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf probe: Fix unchecked usage of strncpy()

The strncpy() function may leave the destination string buffer
unterminated, better use strlcpy() that we have a __weak fallback
implementation for systems without it.

In this case the 'target' buffer is coming from a list of build-ids that
are expected to have a len of at most (SBUILD_ID_SIZE - 1) chars, so
probably we're safe, but since we're using strncpy() here, use strlcpy()
instead to provide the intended safety checking without the using the
problematic strncpy() function.

This fixes this warning on an Alpine Linux Edge system with gcc 8.2:

  util/probe-file.c: In function 'probe_cache__open.isra.5':
  util/probe-file.c:427:3: error: 'strncpy' specified bound 41 equals destination size [-Werror=stringop-truncation]
     strncpy(sbuildid, target, SBUILD_ID_SIZE);
     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  cc1: all warnings being treated as errors

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: 1f3736c9c833 ("perf probe: Show all cached probes")
Link: https://lkml.kernel.org/n/tip-l7n8ggc9kl38qtdlouke5yp5@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf ui helpline: Use strlcpy() as a shorter form of strncpy() + explicit set nul

The strncpy() function may leave the destination string buffer
unterminated, better use strlcpy() that we have a __weak fallback
implementation for systems without it.

In this case we are actually setting the null byte at the right place,
but since we pass the buffer size as the limit to strncpy() and not
it minus one, gcc ends up warning us about that, see below. So, lets
just switch to the shorter form provided by strlcpy().

This fixes this warning on an Alpine Linux Edge system with gcc 8.2:

  ui/tui/helpline.c: In function 'tui_helpline__push':
  ui/tui/helpline.c:27:2: error: 'strncpy' specified bound 512 equals destination size [-Werror=stringop-truncation]
    strncpy(ui_helpline__current, msg, sz)[sz - 1] = '\0';
    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  cc1: all warnings being treated as errors

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: e6e904687949 ("perf ui: Introduce struct ui_helpline")
Link: https://lkml.kernel.org/n/tip-d1wz0hjjsh19xbalw69qpytj@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf svghelper: Fix unchecked usage of strncpy()

The strncpy() function may leave the destination string buffer
unterminated, better use strlcpy() that we have a __weak fallback
implementation for systems without it.

In this specific case this would only happen if fgets() was buggy, as
its man page states that it should read one less byte than the size of
the destination buffer, so that it can put the nul byte at the end of
it, so it would never copy 255 non-nul chars, as fgets reads into the
orig buffer at most 254 non-nul chars and terminates it. But lets just
switch to strlcpy to keep the original intent and silence the gcc 8.2
warning.

This fixes this warning on an Alpine Linux Edge system with gcc 8.2:

  In function 'cpu_model',
      inlined from 'svg_cpu_box' at util/svghelper.c:378:2:
  util/svghelper.c:337:5: error: 'strncpy' output may be truncated copying 255 bytes from a string of length 255 [-Werror=stringop-truncation]
       strncpy(cpu_m, &buf[13], 255);
       ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Fixes: f48d55ce7871 ("perf: Add a SVG helper library file")
Link: https://lkml.kernel.org/n/tip-xzkoo0gyr56gej39ltivuh9g@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf help: Remove needless use of strncpy()

Since we make sure the destination buffer has at least strlen(orig) + 1,
no need to do a strncpy(dest, orig, strlen(orig)), just use strcpy(dest,
orig).

This silences this gcc 8.2 warning on Alpine Linux:

  In function 'add_man_viewer',
      inlined from 'perf_help_config' at builtin-help.c:284:3:
  builtin-help.c:192:2: error: 'strncpy' output truncated before terminating nul copying as many bytes from a string as its length [-Werror=stringop-truncation]
    strncpy((*p)->name, name, len);
    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  builtin-help.c: In function 'perf_help_config':
  builtin-help.c:187:15: note: length computed here
    size_t len = strlen(name);
                 ^~~~~~~~~~~~

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: 078006012401 ("perf_counter tools: add in basic glue from Git")
Link: https://lkml.kernel.org/n/tip-2f69l7drca427ob4km8i7kvo@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf header: Fix unchecked usage of strncpy()

The strncpy() function may leave the destination string buffer
unterminated, better use strlcpy() that we have a __weak fallback
implementation for systems without it.

This fixes this warning on an Alpine Linux Edge system with gcc 8.2:

  util/header.c: In function 'perf_event__synthesize_event_update_name':
  util/header.c:3625:2: error: 'strncpy' output truncated before terminating nul copying as many bytes from a string as its length [-Werror=stringop-truncation]
    strncpy(ev->data, evsel->name, len);
    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  util/header.c:3618:15: note: length computed here
    size_t len = strlen(evsel->name);
                 ^~~~~~~~~~~~~~~~~~~

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: a6e5281780d1 ("perf tools: Add event_update event unit type")
Link: https://lkml.kernel.org/n/tip-wycz66iy8dl2z3yifgqf894p@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf header: Fix unchecked usage of strncpy()

The strncpy() function may leave the destination string buffer
unterminated, better use strlcpy() that we have a __weak fallback
implementation for systems without it.

This fixes this warning on an Alpine Linux Edge system with gcc 8.2:

  util/header.c: In function 'perf_event__synthesize_event_update_unit':
  util/header.c:3586:2: error: 'strncpy' output truncated before terminating nul copying as many bytes from a string as its length [-Werror=stringop-truncation]
    strncpy(ev->data, evsel->unit, size);
    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  util/header.c:3579:16: note: length computed here
    size_t size = strlen(evsel->unit);
                  ^~~~~~~~~~~~~~~~~~~

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: a6e5281780d1 ("perf tools: Add event_update event unit type")
Link: https://lkml.kernel.org/n/tip-fiikh5nay70bv4zskw2aa858@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf dso: Fix unchecked usage of strncpy()

The strncpy() function may leave the destination string buffer
unterminated, better use strlcpy() that we have a __weak fallback
implementation for systems without it.

This fixes this warning on an Alpine Linux Edge system with gcc 8.2:

  In function 'decompress_kmodule',
      inlined from 'dso__decompress_kmodule_fd' at util/dso.c:305:9:
  util/dso.c:298:3: error: 'strncpy' destination unchanged after copying no bytes [-Werror=stringop-truncation]
     strncpy(pathname, tmpbuf, len);
     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    CC       /tmp/build/perf/util/values.o
    CC       /tmp/build/perf/util/debug.o
  cc1: all warnings being treated as errors

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: c9a8a6131fb6 ("perf tools: Move the temp file processing into decompress_kmodule")
Link: https://lkml.kernel.org/n/tip-tl2hdxj64tt4k8btbi6a0ugw@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cs-etm: Add support for PTMv1.1 decoding

This patch is re-using the mechanic set forth by ETMv3 to add support
for PTM decoding.  Configuration for both encoding protocol is similar
but the generated stream itself is very different, hence requiring
special handling.

Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: coresight@lists.linaro.org
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1543955944-10042-4-git-send-email-mathieu.poirier@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cs-etm: Add support for ETMv3 trace decoding

Add support for the creation of packet printer and decoder for the ETMv3
trace architecture.  That way traces generated by tracers adhering to
that trace protocol can be handled properly by the perf infrastructure.

Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: coresight@lists.linaro.org
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1543955944-10042-3-git-send-email-mathieu.poirier@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cs-etm: Add configuration for ETMv3 trace protocol

This patch deals with the proper initialisation of configuration
parameters for the ETMv3 trace protocol in order to properly handle
packets generated by tracers following this specification.

Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: coresight@lists.linaro.org
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1543955944-10042-2-git-send-email-mathieu.poirier@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Move perf_top__reset_sample_counters() to after counts display

Move the perf_top__reset_sample_counters() call to right after we
display the counters so we can see the updated numbers for longer.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/n/tip-o72pyiwt05f3p2juprwmz2jo@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Display slow reader warning when droping samples

Currently we display the "Too slow to read ring buffer.." helpline only
in the slow reader thread. This patch triggers it also when the
processing thread drops samples, because it has the same reason, which
is too many data on input.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/n/tip-bnev2mloavyurmgchcr3o24o@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Save and display the drop count stats

Add drop count to 'perf top' headers:

  # perf top --stdio
   PerfTop:    3549 irqs/sec  kernel:51.8%  exact: 100.0% lost: 0/0 drop: 0/0 [4000Hz cycles:ppp],  (all, 8 CPUs)

  # perf top
  Samples: 0  of event 'cycles:ppp', 4000 Hz, Event count (approx.): 0 lost: 0/0 drop: 0/0

The format is: <current period drop>/<total drop>

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-2lj87zz8tq9ye1ntax3ulw0n@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Drop samples which are behind the refresh rate

Drop samples from processing thread if they get behind the latest event
read from the kernel maps. If it gets behind more than the refresh rate
(-d option), drop the sample.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/n/tip-x533ra5c1pgofvbtsizzuydd@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Set the 'session_done' volatile variable when exiting

So we can get out of hist processing ASAP on user request.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/n/tip-r8aufbgbixr2f85s3wcoaw9v@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Use cond variable instead of a lock

Use conditional variable logic to synchronize between the reading and
processing threads. Currently it's done by having mutex around rotation
code.

Using a POSIX cond variable to sync both threads after queues rotation:

  Process thread:

    - Detects data
    - Switches queues
    - Sets rotate variable
    - Waits in pthread_cond_wait()

  Read thread:

    - Detects rotate is set
    - Kicks the process thread with a pthread_cond_signal()

After this rotation is safely completed and both threads can continue
with the new queue.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-3rdeg23rv3brvy1pwt3igvyw@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Add processing thread

Add a new thread that takes care of the hist creating to alleviate the
main reader thread so it can keep perf mmaps served in time so that we
reduce the possibility of losing events.

The 'perf top' command now spawns 2 extra threads, the data processing
is the following:

  1) The main thread reads the data from mmaps and queues them to
     ordered events object;

  2) The processing threads takes the data from the ordered events
     object and create initial histogram;

  3) The GUI thread periodically sorts the initial histogram and
     presents it.

Passing the data between threads 1 and 2 is done by having 2 ordered
events queues. One is always being stored by thread 1 while the other is
flushed out in thread 2.

Passing the data between threads 2 and 3 stays the same as was initially
for threads 1 and 3.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-hhf4hllgkmle9wl1aly1jli0@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Move lost events warning to helpline

We can't display the UI box saying that we are slow in the reader
thread.  That will make 'perf top' even slower and the user even more
angry ;-)

Move the UI box message from the reader thread to the UI thread and
change it to a helpline, so there's no need to 'press any key'.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/n/tip-x4k0iuw7tt6mywsaguq6jfwu@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Save and display the lost count stats

Add a 'lost count' to 'perf top' headers:

  # perf top --stdio
   PerfTop:    3850 irqs/sec  kernel:49.0%  exact: 100.0% lost: 0/0 [4000Hz cycles:ppp],  (all, 8 CPUs)

  # perf top
  Samples: 0  of event 'cycles:ppp', 4000 Hz, Event count (approx.): 0 lost: 0/0

The format is: <current period lost>/<total lost>

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lkml.kernel.org/n/tip-zo11rn270gij5jtp8fknpf8u@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf ordered_events: Add private data member

We will need it in following patch, where we can't use the
container_of() trick to get the higher level object.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-vgs9aoek21v14o3obza586yy@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf ordered_events: Rework show_progress for __ordered_events__flush

Decide to use the progress bar one level higher, we will need this in
following patch.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/n/tip-ocjdukp2a8ujikkmafd0j5zv@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Support 'srccode' output

When looking at PT or brstackinsn traces with 'perf script' it can be
very useful to see the source code. This adds a simple facility to print
them with 'perf script', if the information is available through dwarf

  % perf record ...
  % perf script -F insn,ip,sym,srccode
  ...

            4004c6 main
  5               for (i = 0; i < 10000000; i++)
             4004cd main
  5               for (i = 0; i < 10000000; i++)
             4004c6 main
  5               for (i = 0; i < 10000000; i++)
             4004cd main
  5               for (i = 0; i < 10000000; i++)
             4004cd main
  5               for (i = 0; i < 10000000; i++)
             4004cd main
  5               for (i = 0; i < 10000000; i++)
             4004cd main
  5               for (i = 0; i < 10000000; i++)
             4004cd main
  5               for (i = 0; i < 10000000; i++)
             4004b3 main
  6                       v++;

  % perf record -b ...
  % perf script -F insn,ip,sym,srccode,brstackinsn

  ...
         main+22:
          0000000000400543        insn: e8 ca ff ff ff            # PRED
  |18                     f1();
          f1:
          0000000000400512        insn: 55
  |10       {
          0000000000400513        insn: 48 89 e5
          0000000000400516        insn: b8 00 00 00 00
  |11             f2();
          000000000040051b        insn: e8 d6 ff ff ff            # PRED
          f2:
          00000000004004f6        insn: 55
  |5        {
          00000000004004f7        insn: 48 89 e5
          00000000004004fa        insn: 8b 05 2c 0b 20 00
  |6              c = a / b;
          0000000000400500        insn: 8b 0d 2a 0b 20 00
          0000000000400506        insn: 99
          0000000000400507        insn: f7 f9
          0000000000400509        insn: 89 05 29 0b 20 00
          000000000040050f        insn: 90
  |7        }
          0000000000400510        insn: 5d
          0000000000400511        insn: c3                        # PRED
          f1+14:
          0000000000400520        insn: b8 00 00 00 00
  |12             f2();
          0000000000400525        insn: e8 cc ff ff ff            # PRED
          f2:
          00000000004004f6        insn: 55
  |5        {
          00000000004004f7        insn: 48 89 e5
          00000000004004fa        insn: 8b 05 2c 0b 20 00
  |6              c = a / b;

Not supported for callchains currently, would need some layout changes
there.

Committer notes:

Fixed the build on Alpine Linux (3.4 .. 3.8) by addressing this
warning:

  In file included from util/srccode.c:19:0:
  /usr/include/sys/fcntl.h:1:2: error: #warning redirecting incorrect #include <sys/fcntl.h> to <fcntl.h> [-Werror=cpp]
   #warning redirecting incorrect #include <sys/fcntl.h> to <fcntl.h>
    ^~~~~~~
  cc1: all warnings being treated as errors

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Tested-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20181204001848.24769-1-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: We need to consider "nr" if "__syscall_nr" is not there

To cope with older kernels that don't have this patch backported:

  026842d148b9 ("tracing/syscalls: Rename "/format" tracepoint field name "nr" to "__syscall_nr:")

This makes 'perf trace' work again in RHEL7 kernels.

Cc: Taeung Song <treeze.taeung@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-6h1syw2isegnhb1bjmtr9x9k@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Allow specifying proc-map-timeout in config file

The default timeout of 500ms for parsing /proc/<pid>/maps files is too
short for profiling many of our services.

This can be overridden by passing --proc-map-timeout to the relevant
command but it'd be nice to globally increase our default value.

This patch permits setting a different default with the
core.proc-map-timeout config file parameter.

Signed-off-by: Mark Drayton <mbd@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20181204203420.1683114-1-mbd@fb.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib subcmd: Fix a few source code comment typos

Go over the tools/ files that are maintained in Arnaldo's tree and
fix common typos: half of them were in comments, the other half
in JSON files.

No change in functionality intended.

Committer notes:

This was split from a larger patch as there are code that is,
additionally, maintained outside the kernel tree, so to ease
cherry-picking and/or backporting, split this into multiple patches.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20181203102200.GA104797@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Fix diverse comment typos

Go over the tools/ files that are maintained in Arnaldo's tree and
fix common typos: half of them were in comments, the other half
in JSON files.

No change in functionality intended.

Committer notes:

This was split from a larger patch as there are code that is,
additionally, maintained outside the kernel tree, so to ease
cherry-picking and/or backporting, split this into multiple patches.

Just typos in comments, no need to backport, reducing the possibility of
possible backporting artifacts.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20181203102200.GA104797@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf-loader: Fix debugging message typo

Go over the tools/ files that are maintained in Arnaldo's tree and
fix common typos: half of them were in comments, the other half
in JSON files.

No change in functionality intended.

Committer notes:

This was split from a larger patch as there are code that is,
additionally, maintained outside the kernel tree, so to ease cherry
picking and/or backporting, split this into multiple patches.

This one has information that is presented to the user, albeit in debug
mode.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/20181203102200.GA104797@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools Documentation: Fix diverse typos

Go over the tools/ files that are maintained in Arnaldo's tree and
fix common typos: half of them were in comments, the other half
in JSON files.

No change in functionality intended.

Committer notes:

This was split from a larger patch as there are code that is,
additionally, maintained outside the kernel tree, so to ease cherry
picking and/or backporting, split this into multiple patches.

In this particular case, it affects documentation, so may be interesting
to cherry pick as it is information that is presented to the user.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20181203102200.GA104797@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Fix diverse typos in comments

Go over the tools/ files that are maintained in Arnaldo's tree and
fix common typos: half of them were in comments, the other half
in JSON files.

No change in functionality intended.

Committer notes:

This was split from a larger patch as there are code that is,
additionally, maintained outside the kernel tree, so to ease cherry
picking and/or backporting, split this into multiple patches.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt (VMware) <rostedt@goodmis.org>
Cc: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Link: http://lkml.kernel.org/r/20181203102200.GA104797@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf vendor events intel: Fix diverse typos

Go over the tools/ files that are maintained in Arnaldo's tree and
fix common typos: half of them were in comments, the other half
in JSON files.

( Care should be taken not to re-import these typos in the future,
  if the JSON files get updated by the vendor without fixing the typos. )

No change in functionality intended.

Committer notes:

This was split from a larger patch as there are code that is,
additionally, maintained outside the kernel tree, so to ease cherry
picking and/or backporting, split this into multiple patches.

Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20181203102200.GA104797@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests ARM: Disable breakpoint tests 32-bit

The breakpoint tests on the ARM 32-bit kernel are broken in several
ways.

The breakpoint length requested does not necessarily match whether the
function address has the Thumb bit (bit 0) set or not, and this does
matter to the ARM kernel hw_breakpoint infrastructure. See [1] for
background.

[1]: https://lkml.org/lkml/2018/11/15/205

As Will indicated, the overflow handling would require single-stepping
which is not supported at the moment. Just disable those tests for the
ARM 32-bit platforms and update the comment above to explain these
limitations.

Co-developed-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Acked-by: Jiri Olsa <jolsa@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20181203191138.2419-1-f.fainelli@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf cs-etm: Support for ARM A32/T32 instruction sets in CoreSight trace

This patch adds support for generating instruction samples from trace of
AArch32 programs using the A32 and T32 instruction sets.

T32 has variable 2 or 4 byte instruction size, so the conversion between
addresses and instruction counts requires extra information from the
trace decoder, requiring version 0.10.0 of OpenCSD.  A check for the
OpenCSD library version has been added to the feature check for OpenCSD.

Signed-off-by: Robert Walker <robert.walker@arm.com>
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Tested-by: Leo Yan <leo.yan@linaro.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: coresight@lists.linaro.org
Cc: linux-arm-kernel@lists.infradead.org
Link: http://lkml.kernel.org/r/1543839526-30348-1-git-send-email-robert.walker@arm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf beauty mmap_flags: Fixed syntax error Fixed missing ']' error

Committer testing:

Before:

  # tools/perf/trace/beauty/mmap_flags.sh
  static const char *mmap_flags[] = {
[ilog2(0x40) + 1] = "32BIT",
  tools/perf/trace/beauty/mmap_flags.sh: line 23: [: missing `]'
[ilog2(0x01) + 1] = "SHARED",
[ilog2(0x02) + 1] = "PRIVATE",
[ilog2(0x10) + 1] = "FIXED",
[ilog2(0x20) + 1] = "ANONYMOUS",
[ilog2(0x100000) + 1] = "FIXED_NOREPLACE",
  tools/perf/trace/beauty/mmap_flags.sh: line 28: [: missing `]'
[ilog2(0x0100) + 1] = "GROWSDOWN",
[ilog2(0x0800) + 1] = "DENYWRITE",
[ilog2(0x1000) + 1] = "EXECUTABLE",
[ilog2(0x2000) + 1] = "LOCKED",
[ilog2(0x4000) + 1] = "NORESERVE",
[ilog2(0x8000) + 1] = "POPULATE",
[ilog2(0x10000) + 1] = "NONBLOCK",
[ilog2(0x20000) + 1] = "STACK",
[ilog2(0x40000) + 1] = "HUGETLB",
[ilog2(0x80000) + 1] = "SYNC",
  };
  #

After:

  $ tools/perf/trace/beauty/mmap_flags.sh
  static const char *mmap_flags[] = {
[ilog2(0x40) + 1] = "32BIT",
[ilog2(0x01) + 1] = "SHARED",
[ilog2(0x02) + 1] = "PRIVATE",
[ilog2(0x10) + 1] = "FIXED",
[ilog2(0x20) + 1] = "ANONYMOUS",
[ilog2(0x100000) + 1] = "FIXED_NOREPLACE",
[ilog2(0x0100) + 1] = "GROWSDOWN",
[ilog2(0x0800) + 1] = "DENYWRITE",
[ilog2(0x1000) + 1] = "EXECUTABLE",
[ilog2(0x2000) + 1] = "LOCKED",
[ilog2(0x4000) + 1] = "NORESERVE",
[ilog2(0x8000) + 1] = "POPULATE",
[ilog2(0x10000) + 1] = "NONBLOCK",
[ilog2(0x20000) + 1] = "STACK",
[ilog2(0x40000) + 1] = "HUGETLB",
[ilog2(0x80000) + 1] = "SYNC",
  };
  $

Signed-off-by: Sihyeon Jang <uneedsihyeon@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Fixes: aca70cc40a0b ("perf beauty mmap_flags: Check if the arch has a mmap.h file")
Link: http://lkml.kernel.org/r/20181202080651.4685-1-uneedsihyeon@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: traceevent API cleanup

In order to make libtraceevent into a proper library, its API should be
straightforward. This patch hides few API functions, intended for
internal usage only:

tep_free_event(), tep_free_format_field(), __tep_data2host2(),
__tep_data2host4() and __tep_data2host8().
The patch also alignes the libtraceevent summary man page with
these API changes.

Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20181130154647.891651290@goodmis.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: traceevent API cleanup, remove __tep_data2host*()

In order to make libtraceevent into a proper library, its API should be
straightforward. The __tep_data2host*() functions are going to no longer
be available as a libtraceevent API, tep_read_number() should be used
instead. This patch replaces __tep_data2host*() usage with
tep_read_number() in perf.

Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20181130154647.743979275@goodmis.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Rename tep_free_format() to tep_free_event()

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts. This renames tep_free_format() to tep_free_event(), which
describes more closely the purpose of the function.

Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20181130154647.591673556@goodmis.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent, perf tools: Rename 'struct tep_event_format' to 'struct tep_event'

In order to make libtraceevent into a proper library, variables, data
structures and functions require a unique prefix to prevent name space
conflicts.

This renames 'struct tep_event_format' to 'struct tep_event', which
describes more closely the purpose of the struct.

Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20181130154647.436403995@goodmis.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
[ Fixup conflict with 6e33c250a88f ("tools lib traceevent: Fix compile warnings in tools/lib/traceevent/event-parse.c") ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Install trace-seq.h API header file

This patch installs trace-seq.h header file on "make install".

Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20181130154647.176265533@goodmis.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Added support for pkg-config

This patch implements integration with pkg-config framework.  pkg-config
can be used by the library users to determine required CFLAGS and
LDFLAGS in order to use the library

Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20181130154647.022471992@goodmis.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Implement new API tep_get_ref()

This patch implements a new API of the tracevent library:

  int tep_get_ref(struct tep_handle *tep);

The API returns the reference counter "ref_count" of the tep handler.
As "struct tep_handle" is internal only, its members cannot be accessed
by the library users, the API is used to get the reference counter.

Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: http://lkml.kernel.org/r/20181130154646.890615385@goodmis.org
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Documentation average IPC and IPC coverage

Add explanations for new columns "IPC" and "IPC coverage" in perf
documentation.

 v5:
 ---
 Update the description according to Ingo's comments.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1543586097-27632-5-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf report: Display average IPC and IPC coverage per symbol

Support displaying the average IPC and IPC coverage per symbol in 'perf
report' --tui and --stdio modes.

For example,

 $ perf record -b ...
 $ perf report -s symbol

 Overhead  Symbol                           IPC   [IPC Coverage]
   39.60%  [.] __random                     2.30  [ 54.8%]
   18.02%  [.] main                         0.43  [ 54.3%]
   14.21%  [.] compute_flag                 2.29  [100.0%]
   14.16%  [.] rand                         0.36  [100.0%]
    7.06%  [.] __random_r                   2.57  [ 70.5%]
    6.85%  [.] rand@plt                     0.00  [  0.0%]

Jiri Olsa <jolsa@redhat.com> provided the patch to support the --stdio
mode. I merged Jiri's code in this patch.

  $ perf report -s symbol --stdio

    # Overhead  Symbol                       IPC   [IPC Coverage]
    # ........  ...........................  ....................
    #
      39.60%  [.] __random                   2.30  [ 54.8%]
      18.02%  [.] main                       0.43  [ 54.3%]
      14.21%  [.] compute_flag               2.29  [100.0%]
      14.16%  [.] rand                       0.36  [100.0%]
       7.06%  [.] __random_r                 2.57  [ 70.5%]
       6.85%  [.] rand@plt                   0.00  [  0.0%]
       0.02%  [k] run_timer_softirq          1.60  [ 57.2%]

The columns "IPC" and "[IPC Coverage]" are automatically enabled when
the sort-key "symbol" is specified. If the perf.data file doesn't
contain timed LBR information, columns are filled with "-".

For example,

  # Overhead  Symbol                       IPC   [IPC Coverage]
  # ........  ...........................  ....................
  #
      46.57%  [.] main                     -      -
      17.60%  [.] rand                     -      -
      15.84%  [.] __random_r               -      -
      11.90%  [.] __random                 -      -
       6.50%  [.] compute_flag             -      -
       1.59%  [.] rand@plt                 -      -
       0.00%  [.] _dl_relocate_object      -      -
       0.00%  [k] tlb_flush_mmu            -      -
       0.00%  [k] perf_event_mmap          -      -
       0.00%  [k] native_sched_clock       -      -
       0.00%  [k] intel_pmu_handle_irq_v4  -      -
       0.00%  [k] native_write_msr         -      -

 v3:
 ---
 Removed the sortkey 'ipc' from command-line. The columns "IPC"
 and "[IPC Coverage]" are automatically enabled when "symbol"
 is specified.

 v2:
 ---
 Merge in Jiri's patch to support stdio mode

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1543586097-27632-4-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Create a annotate2 flag in struct symbol

We often use the symbol__annotate2() to annotate a specified symbol.
While annotating may take some time, so in order to avoid annotating the
same symbol repeatedly, the patch creates a new flag to indicate the
symbol has been annotated.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1543586097-27632-3-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Compute average IPC and IPC coverage per symbol

Add support to 'perf report' annotate view or 'perf annotate --stdio2'
to aggregate the IPC derived from timed LBRs per symbol. We compute the
average IPC and the IPC coverage percentage.

For example:

  $ perf annotate --stdio2

  Percent  IPC Cycle (Average IPC: 2.30, IPC Coverage: 54.8%)

                          Disassembly of section .text:

                          000000000003aac0 <random@@GLIBC_2.2.5>:
    8.32  3.28              sub    $0x18,%rsp
          3.28              mov    $0x1,%esi
          3.28              xor    %eax,%eax
          3.28              cmpl   $0x0,argp_program_version_hook@@GLIBC_2.2.5+0x1e0
   11.57  3.28     1      ↓ je     20
                            lock   cmpxchg %esi,__abort_msg@@GLIBC_PRIVATE+0x8a0
                          ↓ jne    29
                          ↓ jmp    43
   11.57  1.10        20:   cmpxchg %esi,__abort_msg@@GLIBC_PRIVATE+0x8a0
    0.00  1.10     1      ↓ je     43
                      29:   lea    __abort_msg@@GLIBC_PRIVATE+0x8a0,%rdi
                            sub    $0x80,%rsp
                          → callq  __lll_lock_wait_private
                            add    $0x80,%rsp
    0.00  3.00        43:   lea    __ctype_b@GLIBC_2.2.5+0x38,%rdi
          3.00              lea    0xc(%rsp),%rsi
    8.49  3.00     1      → callq  __random_r
    7.91  1.94              cmpl   $0x0,argp_program_version_hook@@GLIBC_2.2.5+0x1e0
    0.00  1.94     1      ↓ je     68
                            lock   decl   __abort_msg@@GLIBC_PRIVATE+0x8a0
                          ↓ jne    70
                          ↓ jmp    8a
    0.00  2.00        68:   decl   __abort_msg@@GLIBC_PRIVATE+0x8a0
   21.56  2.00     1      ↓ je     8a
                      70:   lea    __abort_msg@@GLIBC_PRIVATE+0x8a0,%rdi
                            sub    $0x80,%rsp
                          → callq  __lll_unlock_wake_private
                            add    $0x80,%rsp
   21.56  2.90        8a:   movslq 0xc(%rsp),%rax
          2.90              add    $0x18,%rsp
    9.03  2.90     1      ← retq

It shows for this symbol the average IPC is 2.30 and the IPC coverage is
54.8%.

Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1543586097-27632-2-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Add sanity check to is_timestamp_in_us()

This patch adds a sanity check to is_timestamp_in_us() input parameter
trace_clock. It avoids a potential segfault in this function for the
case trace_clock is NULL.

Reported-by: Slavomir Kaslev <kaslevs@vmware.com>
Signed-off-by: Tzvetomir Stoyanov <tstoyanov@vmware.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/20181128145552.68c4f87b@gandalf.local.home
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf beauty mmap_flags: Check if the arch has a mmap.h file

If not, then just use what is in asm-generic. This fixes the build for
my sh4, m68k and riscv64 perf test build containers that were failing
due to 80ee5668b8a7 ("perf beauty: Add a generator for MAP_ mmap's flag
constants"), that were not covered in the cset introducing those
tools/arch/*/include/uapi/asm/mman.h files.

  f3539c12d819 ("tools include: Add uapi mman.h for each architecture")

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: 80ee5668b8a7 ("perf beauty: Add a generator for MAP_ mmap's flag constants")
Link: https://lkml.kernel.org/n/tip-rpy9t2e0wxpnum1yvxhreafe@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Extend trace writing to multi AIO

Multi AIO trace writing allows caching more kernel data into userspace
memory postponing trace writing for the sake of overall profiling data
thruput increase. It could be seen as kernel data buffer extension into
userspace memory.

With an --aio option value different from 0 (default value is 1) the
tool has capability to cache more and more data into user space along
with delegating spill to AIO.

That allows avoiding to suspend at record__aio_sync() between calls of
record__mmap_read_evlist() and increases profiling data thruput at the
cost of userspace memory.

Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/050bb053-e7f3-aa83-fde7-f27ff90be7f6@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf record: Enable asynchronous trace writing

The trace file offset is read once before mmaps iterating loop and
written back after all performance data is enqueued for aio writing.

The trace file offset is incremented linearly after every successful aio
write operation.

record__aio_sync() blocks till completion of the started AIO operation
and then proceeds.

record__aio_mmap_read_sync() implements a barrier for all incomplete
aio write requests.

Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/ce2d45e9-d236-871c-7c8f-1bed2d37e8ac@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf mmap: Map data buffer for preserving collected data

The map->data buffer is used to preserve map->base profiling data for
writing to disk. AIO map->cblock is used to queue corresponding
map->data buffer for asynchronous writing.

Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/5fcda10c-6c63-68df-383a-c6d9e5d1f918@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build feature: Check if libaio is available

This will be used by 'perf record' to speed up reading the perf ring
buffer.

Committer testing:

  $ make -C tools/perf O=/tmp/build/perf
  make: Entering directory '/home/acme/git/perf/tools/perf'
    BUILD:   Doing 'make -j8' parallel build

  Auto-detecting system features:
  ...                         dwarf: [ on  ]
  ...            dwarf_getlocations: [ on  ]
  ...                         glibc: [ on  ]
  ...                          gtk2: [ OFF ]
  ...                      libaudit: [ OFF ]
  ...                        libbfd: [ OFF ]
  ...                        libelf: [ on  ]
  ...                       libnuma: [ OFF ]
  ...        numa_num_possible_cpus: [ OFF ]
  ...                       libperl: [ OFF ]
  ...                     libpython: [ OFF ]
  ...                      libslang: [ on  ]
  ...                     libcrypto: [ on  ]
  ...                     libunwind: [ on  ]
  ...            libdw-dwarf-unwind: [ on  ]
  ...                          zlib: [ on  ]
  ...                          lzma: [ on  ]
  ...                     get_cpuid: [ on  ]
  ...                           bpf: [ on  ]
  ...                        libaio: [ on  ]

  $ ls -la /tmp/build/perf/feature/test-libaio.*
  -rwxrwxr-x. 1 acme acme 18296 Nov 26 08:49 /tmp/build/perf/feature/test-libaio.bin
  -rw-rw-r--. 1 acme acme  1165 Nov 26 08:49 /tmp/build/perf/feature/test-libaio.d
  -rw-rw-r--. 1 acme acme     0 Nov 26 08:49 /tmp/build/perf/feature/test-libaio.make.output
  $
  $ grep -i aio /tmp/build/perf/FEATURE-DUMP
  feature-libaio=1
  $

Signed-off-by: Alexey Budankov <alexey.budankov@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reviewed-by: Jiri Olsa <jolsa@redhat.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/5fcda10c-6c63-68df-383a-c6d9e5d1f918@linux.intel.com
[ split from a larger patch ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf intel-pt: Fix error with config term "pt=0"

Users should never use 'pt=0', but if they do it may give a meaningless
error:

$ perf record -e intel_pt/pt=0/u uname
Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for
event (intel_pt/pt=0/u).

Fix that by forcing 'pt=1'.

Committer testing:

  # perf record -e intel_pt/pt=0/u uname
  Error:
  The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (intel_pt/pt=0/u).
  /bin/dmesg | grep -i perf may provide additional information.

  # perf record -e intel_pt/pt=0/u uname
  pt=0 doesn't make sense, forcing pt=1
  Linux
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.020 MB perf.data ]
  #

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/b7c5b4e5-9497-10e5-fd43-5f3e4a0fe51d@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf top: Allow passing a kallsyms file

This basically replicates what was done for 'perf report' in:

   b226a5a72901 ("perf report: Allow user to specify path to kallsyms file")

This should help with resolving eBPF symbols, that are in kallsyms but,
of course, not in vmlinux.

Reported-by: Ivan Babrou <ibobrik@gmail.com>
Tested-by: Ivan Babrou <ibobrik@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-x52mx1ybq8128rtg9hjrj5qk@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Use ERR_CAST instead of ERR_PTR(PTR_ERR())

Use ERR_CAST inlined function instead of ERR_PTR(PTR_ERR(...)).  This
makes it more readable and also fix this warning detected by
err_cast.cocci:

  tools/perf/util/bpf-loader.c:1606:11-18: WARNING: ERR_CAST can be used with op

Signed-off-by: Wen Yang <wen.yang99@zte.com.cn>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Julia Lawall <julia.lawall@lip6.fr>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Wen Yang <yellowriver2010@hotmail.com>
Cc: zhong.weidong@zte.com.cn
Link: http://lkml.kernel.org/r/20181127090610.28488-1-wen.yang99@zte.com.cn
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools include: Adopt ERR_CAST() from the kernel err.h header

Add ERR_CAST(), so that tools can use it, just like the kernel.

This addresses coccinelle checks that are being performed to tools/ in
addition to kernel sources, so lets add this to cover that and to get
tools code closer to kernel coding standards.

This originally was introduced in the kernel headers in this cset:

  d1bc8e954452 ("Add an ERR_CAST() function to complement ERR_PTR and co.")

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Julia Lawall <julia.lawall@lip6.fr>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: Wen Yang <yellowriver2010@hotmail.com>
Cc: zhong.weidong@zte.com.cn
Link: https://lkml.kernel.org/n/tip-tlt97p066zyhzqhl5jt86og7@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf test: Fix perf_event_attr test failure

Fix inconsistent use of tabs and spaces error:

  # perf test 16 -v
  16: Setup struct perf_event_attr                          :
  --- start ---
  test child forked, pid 20224
    File "/usr/libexec/perf-core/tests/attr.py", line 119
      log.warning("expected %s=%s, got %s" % (t, self[t], other[t]))
                                                                 ^
  TabError: inconsistent use of tabs and spaces in indentation
  test child finished with -1
  ---- end ----
  Setup struct perf_event_attr: FAILED!

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/20181122140456.16817-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tests record: Allow for 'sleep' being 'coreutils'

If the 'sleep' command is provided by coreutils, then the "PERF_RECORD_*
events & perf_sample fields" test will fail because the MMAP name is
'coreutils' not 'sleep', and there is an extra COMM event. Fix the test
to detect that case.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Link: http://lkml.kernel.org/r/20181122135545.16295-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools lib traceevent: Fix compile warnings in tools/lib/traceevent/event-parse.c

Fix following warnings:

  event-parse.c: In function ‘tep_find_event_by_name’:
  event-parse.c:3521:21: warning: ‘event’ may be used uninitialized in this function [-Wmaybe-uninitialized]
    pevent->last_event = event;
    ~~~~~~~~~~~~~~~~~~~^~~~~~~
    CC       ui/gtk/hists.o
    LINK     plugin_mac80211.so
    CC       nlattr.o
  event-parse.c: In function ‘tep_data_lat_fmt’:
  event-parse.c:5200:4: warning: ‘migrate_disable’ may be used uninitialized in this function [-Wmaybe-uninitialized]
      trace_seq_printf(s, "%d", migrate_disable);
      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  event-parse.c:5207:4: warning: ‘lock_depth’ may be used uninitialized in this function [-Wmaybe-uninitialized]
      trace_seq_printf(s, "%d", lock_depth);
      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    LINK     plugin_sched_switch.so
    LINK     plugin_function.so
    LINK     plugin_xen.so
  event-parse.c: In function ‘tep_event_info’:
  event-parse.c:5047:7: warning: ‘len_arg’ may be used uninitialized in this function [-Wmaybe-uninitialized]
         trace_seq_printf(s, format, len_arg, (char)val);
         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  event-parse.c:4884:6: note: ‘len_arg’ was declared here
    int len_arg;
        ^~~~~~~
  event-parse.c:4338:11: warning: ‘vsize’ may be used uninitialized in this function [-Wmaybe-uninitialized]
       val = tep_read_number(pevent, bptr, vsize);
             ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  event-parse.c:4224:6: note: ‘vsize’ was declared here
    int vsize;
        ^~~~~

$ gcc --version
  gcc (Clear Linux OS for Intel Architecture) 8.2.1 20180502

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Steven Rostedt (VMware) <rostedt@goodmis.org>
Cc: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
Link: http://lkml.kernel.org/r/20181122112937.10582-1-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf script: Use fallbacks for branch stacks

Branch stacks do not necessarily have the same cpumode as the 'ip'. Use
the fallback functions in those cases.

This patch depends on patch "perf tools: Add fallback functions for cases
where cpumode is insufficient".

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: stable@vger.kernel.org # 4.19
Link: http://lkml.kernel.org/r/20181106210712.12098-4-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Use fallback for sample_addr_correlates_sym() cases

thread__resolve() is used in the sample_addr_correlates_sym() cases
where 'addr' is a destination of a branch which does not necessarily
have the same cpumode as the 'ip'. Use the fallback function in that
case.

This patch depends on patch "perf tools: Add fallback functions for
cases where cpumode is insufficient".

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: stable@vger.kernel.org # 4.19
Link: http://lkml.kernel.org/r/20181106210712.12098-3-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf thread: Add fallback functions for cases where cpumode is insufficient

For branch stacks or branch samples, the sample cpumode might not be
correct because it applies only to the sample 'ip' and not necessary to
'addr' or branch stack addresses. Add fallback functions that can be
used to deal with those cases

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: stable@vger.kernel.org # 4.19
Link: http://lkml.kernel.org/r/20181106210712.12098-2-adrian.hunter@intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf machine: Record if a arch has a single user/kernel address space

Some architectures have a single address space for kernel and user
addresses, which makes it possible to determine if an address is in
kernel space or user space. Some don't, e.g.: sparc.

Cache that info in perf_env so that, for instance, code needing to
fallback failed symbol lookups at the kernel space in single address
space arches can lookup at userspace.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: stable@vger.kernel.org # 4.19
Link: http://lkml.kernel.org/r/20181106210712.12098-2-adrian.hunter@intel.com
[ split from a larger patch ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf env: Also consider env->arch == NULL as local operation

We'll set a new machine field based on env->arch, which for live mode,
like with 'perf top' means we need to use uname() to figure the name of
the arch, fix perf_env__arch() to consider both (env == NULL) and
(env->arch == NULL) as local operation.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: stable@vger.kernel.org # 4.19
Link: https://lkml.kernel.org/n/tip-vcz4ufzdon7cwy8dm2ua53xk@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf map: Remove extra indirection from map__find()

A double pointer is used in map__find() where a single pointer is enough
because the function doesn't affect the rbtree and the rbtree is locked.

Signed-off-by: Eric Saint-Etienne <eric.saint.etienne@oracle.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Eric Saint-Etienne <eric.saintetienne@gmail.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1542969759-24346-1-git-send-email-eric.saint.etienne@oracle.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Fix CSV mode column output for non-cgroup events

When using the -x option, perf stat prints CSV-style output with one
event per line.  For each event, it prints the count, the unit, the
event name, the cgroup, and a bunch of other event specific fields (such
as insn per cycles).

When you use CSV-style mode, you expect a normalized output where each
event is printed with the same number of fields regardless of what it is
so it can easily be imported into a spreadsheet or parsed.

For instance, if an event does not have a unit, then print an empty
field for it.

Although this approach was implemented for the unit, it was not for the
cgroup.

When mixing cgroup and non-cgroup events, then non-cgroup events would
not show an empty field, instead the next field was printed, make
columns not line up correctly.

This patch fixes the cgroup output issues by forcing an empty field
for non-cgroup events as soon as one event has cgroup.

Before:

  <not counted> @ @cycles @foo    @ 0    @100.00@@
  2531614       @ @cycles @6420922@100.00@    @

foo cgroup lines up with time_running!

After:

  <not counted> @ @cycles @foo @0       @100.00@@
  2594834       @ @cycles @    @5287372 @100.00@@

Fields line up.

Signed-off-by: Stephane Eranian <eranian@google.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1541587845-9150-1-git-send-email-eranian@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Fix shadow stats for clock events

Commit 0aa802a79469 ("perf stat: Get rid of extra clock display
function") introduced scale and unit for clock events. Thus,
perf_stat__update_shadow_stats() now saves scaled values of clock events
in msecs, instead of original nsecs. But while calculating values of
shadow stats we still consider clock event values in nsecs. This results
in a wrong shadow stat values. Ex,

  # ./perf stat -e task-clock,cycles ls
    <SNIP>
              2.60 msec task-clock:u    #    0.877 CPUs utilized
         2,430,564      cycles:u        # 1215282.000 GHz

Fix this by saving original nsec values for clock events in
perf_stat__update_shadow_stats(). After patch:

  # ./perf stat -e task-clock,cycles ls
    <SNIP>
              3.14 msec task-clock:u    #    0.839 CPUs utilized
         3,094,528      cycles:u        #    0.985 GHz

Suggested-by: Jiri Olsa <jolsa@redhat.com>
Reported-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Thomas Richter <tmricht@linux.vnet.ibm.com>
Cc: yuzhoujian@didichuxing.com
Fixes: 0aa802a79469 ("perf stat: Get rid of extra clock display function")
Link: http://lkml.kernel.org/r/20181116042843.24067-1-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf build: Give better hint about devel package for libssl

In debian/ubuntu its libssl-dev, but for fedora/RHEL/Centos/etc its
openssl-devel, fix it.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: 8ee4646038e4 ("perf build: Add libcrypto feature detection")
Link: https://lkml.kernel.org/n/tip-lnxqszts6aq2c9jy4b7mlnym@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge tag 'perf-core-for-mingo-4.21-20181122' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/core

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

- Start using BPF maps in 'perf trace' for filters in the augmented syscalls
  code, keeping the existing code for tracepoint filters so that we can switch
  back and forth while getting everything BPFied (Arnaldo Carvalho de Melo)

- Suppress potential format-truncation warning in the PMU code (Ben Hutchings)

- Introduce 'perf bench epoll', with "wait" and "ctl" benchmarks (Davidlohr Bueso)

- Fix slowness due to -ffunction-section, do it by sorting the maps by name, so
  avoiding the using rb_first/next to traverse all entries looking for a map name,
  that with --ffunction-section gets to thousands of maps (Eric Saint-Etienne)

- Separate jvmti cmlr check (Jiri Olsa)

- Allow using the stepping when figuring out which JSON files to use for a x86
  processor, so that Cascadelake server can be support, which has the same
  cpuid as some other processor, being different only in the stepping (Kan Liang)

- Share code and output format for uregs and iregs 'perf script' output (Milian Wolff)

- Use perf_evsel__is_clocki() for clock events in 'perf stat' (Ravi Bangoria)

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

perf pmu: Move *_cpuid_str() weak functions to header.c

The weak functions, strcmp_cpuid_str() and get_cpuid_str(), are defined
in pmu.c.

Most of the cpuid related functions, including *_cpuid_str()'s
declaration and platform specific definition, are in header.c/h.

To make the declaration and definition of all cpuid related functions in
a consistent place, move the weak functions to header.c.

There is no functional change.

Suggested-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Link: http://lkml.kernel.org/r/20181121164939.13482-1-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf symbols: Fix slowness due to -ffunction-section

Perf can take minutes to parse an image when -ffunction-section is used.
This is especially true with the kernel image when it is compiled this
way, which is the arm64 default since the patcheset "Enable deadcode
elimination at link time".

Perf organize maps using a rbtree. Whenever perf finds a new symbols, it
first searches this rbtree for the map it belongs to, by strcmp()'aring
section names.  When it finds the map with the right name, it uses it to
add the symbol. With a usual image there aren't so many maps but when
using -ffunction-section there's basically one map per function.  With
the kernel image that's north of 40,000 maps. For most symbols perf has
to parses the entire rbtree to eventually create a new map and add it.
Consequently perf spends most of the time browsing a rbtree that keeps
getting larger.

This performance fix introduces a secondary rbtree that indexes maps
based on the section name.

Signed-off-by: Eric Saint-Etienne <eric.saint.etienne@oracle.com>
Reviewed-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Reviewed-by: David Aldridge <david.aldridge@oracle.com>
Reviewed-by: Rob Gardner <rob.gardner@oracle.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1542822679-25591-1-git-send-email-eric.saint.etienne@oracle.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf jvmti: Separate jvmti cmlr check

The Compiled Method Load Record (cmlr) is JDK specific interface to
access JVM stack info. This makes the jvmti agent code not compile under
another jdk, which does not support that.

Separating jvmti cmlr check into special feature check, and adding
HAVE_JVMTI_CMLR macro to indicate that.

Mark cmlr code in jvmti/libjvmti.c with HAVE_JVMTI_CMLR, so we can
compile it on system without cmlr support.

This change makes the jvmti compile with java-1.8.0-ibm package. It's
without the line numbers support, but the rest works.

Adding NO_JVMTI_CMLR compile variable for testing.

Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ben Gainey <ben.gainey@arm.com>
Cc: Gustavo Luiz Duarte <gduarte@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/20181121154341.21521-1-jolsa@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf vendor events: Add JSON metrics for Cascadelake server

Add JSON metrics (based on event list v1) for Cascadelake server

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/3ab97c73-c197-8555-1a35-b54636e667e6@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf vendor events: Add stepping in CPUID string for x86

The perf tools cannot find the proper event list for the Cascadelake
server.  Because the Cascadelake server and the Skylake server have the
same CPU model number, which are used by the perf tools to find the
event list.

The stepping for Skylake server is up to 4.

The stepping for Cascadelake server starts from 5.

The stepping can be used to distinguish between them.

The stepping is added in get_cpuid_str().

The stepping information for Skylake server is updated in mapfile.csv.

A x86 specific strcmp_cpuid_cmp() function is added to handle two CPUID
formats in mapfile.csv, "vendor-family-model-stepping" and
"vendor-family-model":

- If a cpuid-regular-expression from the mapfile.csv using the new
  stepping format, a cpuid-string generated on the machine must include
  stepping. Otherwise, it is a mismatch.

- If the cpuid-regular-expression using the old non-stepping format,
  the stepping in the cpuid-string will be ignored.

The script, using environment string "PERF_CPUID" without stepping on
Skylake server, will be broken. If so, users must fix their scripts.

Committer notes:

Fixed this build error on centos:6 and debian:7:

  arch/x86/util/header.c: In function 'is_full_cpuid':
  arch/x86/util/header.c:82:39: error: declaration of 'cpuid' shadows a global declaration [-Werror=shadow]
  arch/x86/util/header.c:12:1: error: shadowed declaration is here [-Werror=shadow]
  arch/x86/util/header.c: In function 'strcmp_cpuid_str':
  arch/x86/util/header.c:98:56: error: declaration of 'cpuid' shadows a global declaration [-Werror=shadow]
  arch/x86/util/header.c:12:1: error: shadowed declaration is here [-Werror=shadow]
  cc1: all warnings being treated as errors

Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20181114212416.15665-1-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf stat: Use perf_evsel__is_clocki() for clock events

We already have function to check if a given event is either
SW_CPU_CLOCK or SW_TASK_CLOCK. Utilize it.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: Jin Yao <yao.jin@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Thomas Richter <tmricht@linux.vnet.ibm.com>
Cc: yuzhoujian@didichuxing.com
Link: http://lkml.kernel.org/r/20181115095533.16930-1-ravi.bangoria@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf pmu: Suppress potential format-truncation warning

Depending on which functions are inlined in util/pmu.c, the snprintf()
calls in perf_pmu__parse_{scale,unit,per_pkg,snapshot}() might trigger a
warning:

  util/pmu.c: In function 'pmu_aliases':
  util/pmu.c:178:31: error: '%s' directive output may be truncated writing up to 255 bytes into a region of size between 0 and 4095 [-Werror=format-truncation=]
    snprintf(path, PATH_MAX, "%s/%s.unit", dir, name);
                               ^~

I found this when trying to build perf from Linux 3.16 with gcc 8.
However I can reproduce the problem in mainline if I force
__perf_pmu__new_alias() to be inlined.

Suppress this by using scnprintf() as has been done elsewhere in perf.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: http://lkml.kernel.org/r/20181111184524.fux4taownc6ndbx6@decadent.org.uk
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf tools: Add Hygon Dhyana support

The tool perf is useful for the performance analysis on the Hygon Dhyana
platform. But right now there is no Hygon support for it to analyze the
KVM guest os data. So add Hygon Dhyana support to it by checking vendor
string to share the code path of AMD.

Signed-off-by: Pu Wen <puwen@hygon.cn>
Acked-by: Borislav Petkov <bp@suse.de>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1542008451-31735-1-git-send-email-puwen@hygon.cn
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bench: Add epoll_ctl(2) benchmark

Benchmark the various operations allowed for epoll_ctl(2).  The idea is
to concurrently stress a single epoll instance doing add/mod/del
operations.

Committer testing:

  # perf bench epoll ctl
  # Running 'epoll/ctl' benchmark:
  Run summary [PID 20344]: 4 threads doing epoll_ctl ops 64 file-descriptors for 8 secs.

  [thread  0] fdmap: 0x21a46b0 ... 0x21a47ac [ add: 1680960 ops; mod: 1680960 ops; del: 1680960 ops ]
  [thread  1] fdmap: 0x21a4960 ... 0x21a4a5c [ add: 1685440 ops; mod: 1685440 ops; del: 1685440 ops ]
  [thread  2] fdmap: 0x21a4c10 ... 0x21a4d0c [ add: 1674368 ops; mod: 1674368 ops; del: 1674368 ops ]
  [thread  3] fdmap: 0x21a4ec0 ... 0x21a4fbc [ add: 1677568 ops; mod: 1677568 ops; del: 1677568 ops ]

  Averaged 1679584 ADD operations (+- 0.14%)
  Averaged 1679584 MOD operations (+- 0.14%)
  Averaged 1679584 DEL operations (+- 0.14%)
  #

Lets measure those calls with 'perf trace' to get a glympse at what this
benchmark is doing in terms of syscalls:

  # perf trace -m32768 -s perf bench epoll ctl
  # Running 'epoll/ctl' benchmark:
  Run summary [PID 20405]: 4 threads doing epoll_ctl ops 64 file-descriptors for 8 secs.

  [thread  0] fdmap: 0x21764e0 ... 0x21765dc [ add: 1100480 ops; mod: 1100480 ops; del: 1100480 ops ]
  [thread  1] fdmap: 0x2176790 ... 0x217688c [ add: 1250176 ops; mod: 1250176 ops; del: 1250176 ops ]
  [thread  2] fdmap: 0x2176a40 ... 0x2176b3c [ add: 1022464 ops; mod: 1022464 ops; del: 1022464 ops ]
  [thread  3] fdmap: 0x2176cf0 ... 0x2176dec [ add: 705472 ops; mod: 705472 ops; del: 705472 ops ]

  Averaged 1019648 ADD operations (+- 11.27%)
  Averaged 1019648 MOD operations (+- 11.27%)
  Averaged 1019648 DEL operations (+- 11.27%)

  Summary of events:

  epoll-ctl (20405), 1264 events, 0.0%

   syscall            calls    total       min       avg       max      stddev
                               (msec)    (msec)    (msec)    (msec)        (%)
   --------------- -------- --------- --------- --------- ---------     ------
   eventfd2             256     9.514     0.001     0.037     5.243     68.00%
   clone                  4     1.245     0.204     0.311     0.531     24.13%
   mprotect              66     0.345     0.002     0.005     0.021      7.43%
   openat                45     0.313     0.004     0.007     0.073     21.93%
   mmap                  88     0.302     0.002     0.003     0.013      5.02%
   futex                  4     0.160     0.002     0.040     0.140     83.43%
   sched_setaffinity      4     0.124     0.005     0.031     0.070     49.39%
   read                  44     0.103     0.001     0.002     0.013     15.54%
   fstat                 40     0.052     0.001     0.001     0.003      5.43%
   close                 39     0.039     0.001     0.001     0.001      1.48%
   stat                   9     0.034     0.003     0.004     0.006      7.30%
   access                 3     0.023     0.007     0.008     0.008      4.25%
   open                   2     0.021     0.008     0.011     0.013     22.60%
   getdents               4     0.019     0.001     0.005     0.009     37.15%
   write                  2     0.013     0.004     0.007     0.009     38.48%
   munmap                 1     0.010     0.010     0.010     0.010      0.00%
   brk                    3     0.006     0.001     0.002     0.003     26.34%
   rt_sigprocmask         2     0.004     0.001     0.002     0.003     43.95%
   rt_sigaction           3     0.004     0.001     0.001     0.002     16.07%
   prlimit64              3     0.004     0.001     0.001     0.001      5.39%
   prctl                  1     0.003     0.003     0.003     0.003      0.00%
   epoll_create           1     0.003     0.003     0.003     0.003      0.00%
   lseek                  2     0.002     0.001     0.001     0.001     11.42%
   sched_getaffinity        1     0.002     0.002     0.002     0.002      0.00%
   arch_prctl             1     0.002     0.002     0.002     0.002      0.00%
   set_tid_address        1     0.001     0.001     0.001     0.001      0.00%
   getpid                 1     0.001     0.001     0.001     0.001      0.00%
   set_robust_list        1     0.001     0.001     0.001     0.001      0.00%
   execve                 1     0.000     0.000     0.000     0.000      0.00%

 epoll-ctl (20406), 1245480 events, 14.6%

   syscall            calls    total       min       avg       max      stddev
                               (msec)    (msec)    (msec)    (msec)        (%)
   --------------- -------- --------- --------- --------- ---------     ------
   epoll_ctl         619511  1034.927     0.001     0.002     6.691      0.67%
   nanosleep           3226   616.114     0.006     0.191    10.376      7.57%
   futex                  2    11.336     0.002     5.668    11.334     99.97%
   set_robust_list        1     0.001     0.001     0.001     0.001      0.00%
   clone                  1     0.000     0.000     0.000     0.000      0.00%

 epoll-ctl (20407), 1243151 events, 14.5%

   syscall            calls    total       min       avg       max      stddev
                               (msec)    (msec)    (msec)    (msec)        (%)
   --------------- -------- --------- --------- --------- ---------     ------
   epoll_ctl         618350  1042.181     0.001     0.002     2.512      0.40%
   nanosleep           3220   366.261     0.012     0.114    18.162      9.59%
   futex                  4     5.463     0.001     1.366     5.427     99.12%
   set_robust_list        1     0.002     0.002     0.002     0.002      0.00%

 epoll-ctl (20408), 1801690 events, 21.1%

   syscall            calls    total       min       avg       max      stddev
                               (msec)    (msec)    (msec)    (msec)        (%)
   --------------- -------- --------- --------- --------- ---------     ------
   epoll_ctl         896174  1540.581     0.001     0.002     6.987      0.74%
   nanosleep           4667   783.393     0.006     0.168    10.419      7.10%
   futex                  2     4.682     0.002     2.341     4.681     99.93%
   set_robust_list        1     0.002     0.002     0.002     0.002      0.00%
   clone                  1     0.000     0.000     0.000     0.000      0.00%

 epoll-ctl (20409), 4254890 events, 49.8%

   syscall            calls    total       min       avg       max      stddev
                               (msec)    (msec)    (msec)    (msec)        (%)
   --------------- -------- --------- --------- --------- ---------     ------
   epoll_ctl        2116416  3768.097     0.001     0.002     9.956      0.41%
   nanosleep          11023  1141.778     0.006     0.104     9.447      4.95%
   futex                  3     0.037     0.002     0.012     0.029     70.50%
   set_robust_list        1     0.008     0.008     0.008     0.008      0.00%
   madvise                1     0.005     0.005     0.005     0.005      0.00%
   clone                  1     0.000     0.000     0.000     0.000      0.00%
  #

Committer notes:

Fix build on fedora:24-x-ARC-uClibc, debian:experimental-x-mips,
debian:experimental-x-mipsel, ubuntu:16.04-x-arm and ubuntu:16.04-x-powerpc

    CC       /tmp/build/perf/bench/epoll-ctl.o
  bench/epoll-ctl.c: In function 'init_fdmaps':
  bench/epoll-ctl.c:214:16: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
    for (i = 0; i < nfds; i+=inc) {
                  ^
  bench/epoll-ctl.c: In function 'bench_epoll_ctl':
  bench/epoll-ctl.c:377:16: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
    for (i = 0; i < nthreads; i++) {
                  ^
  bench/epoll-ctl.c:388:16: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
    for (i = 0; i < nthreads; i++) {
                  ^
  cc1: all warnings being treated as errors

Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Jason Baron <jbaron@akamai.com>
Link: http://lkml.kernel.org/r/20181106152226.20883-3-dave@stgolabs.net
[ Use inttypes.h to print rlim_t fields, fixing the build on Alpine Linux / musl libc ]
[ Check if eventfd() is available, i.e. if HAVE_EVENTFD is defined ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bench: Add epoll parallel epoll_wait benchmark

This program benchmarks concurrent epoll_wait(2) for file descriptors
that are monitored with with EPOLLIN along various semantics, by a
single epoll instance. Such conditions can be found when using
single/combined or multiple queuing when load balancing.

Each thread has a number of private, nonblocking file descriptors,
referred to as fdmap. A writer thread will constantly be writing to the
fdmaps of all threads, minimizing each threads's chances of epoll_wait
not finding any ready read events and blocking as this is not what we
want to stress. Full details in the start of the C file.

Committer testing:

  # perf bench
  Usage:
perf bench [<common options>] <collection> <benchmark> [<options>]

        # List of all available benchmark collections:

         sched: Scheduler and IPC benchmarks
           mem: Memory access benchmarks
          numa: NUMA scheduling and MM benchmarks
         futex: Futex stressing benchmarks
         epoll: Epoll stressing benchmarks
           all: All benchmarks

  # perf bench epoll

        # List of available benchmarks for collection 'epoll':

          wait: Benchmark epoll concurrent epoll_waits
           all: Run all futex benchmarks

  # perf bench epoll wait
  # Running 'epoll/wait' benchmark:
  Run summary [PID 19295]: 3 threads monitoring on 64 file-descriptors for 8 secs.

  [thread  0] fdmap: 0xdaa650 ... 0xdaa74c [ 328241 ops/sec ]
  [thread  1] fdmap: 0xdaa900 ... 0xdaa9fc [ 351695 ops/sec ]
  [thread  2] fdmap: 0xdaabb0 ... 0xdaacac [ 381423 ops/sec ]

  Averaged 353786 operations/sec (+- 4.35%), total secs = 8
  #

Committer notes:

Fix the build on debian:experimental-x-mips, debian:experimental-x-mipsel
and others:

    CC       /tmp/build/perf/bench/epoll-wait.o
  bench/epoll-wait.c: In function 'writerfn':
  bench/epoll-wait.c:399:12: error: format '%ld' expects argument of type 'long int', but argument 2 has type 'size_t' {aka 'unsigned int'} [-Werror=format=]
    printinfo("exiting writer-thread (total full-loops: %ld)\n", iter);
              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  ~~~~
  bench/epoll-wait.c:86:31: note: in definition of macro 'printinfo'
    do { if (__verbose) { printf(fmt, ## arg); fflush(stdout); } } while (0)
                                 ^~~
  cc1: all warnings being treated as errors

Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Jason Baron <jbaron@akamai.com> <jbaron@akamai.com>
Link: http://lkml.kernel.org/r/20181106152226.20883-2-dave@stgolabs.net
Link: http://lkml.kernel.org/r/20181106182349.thdkpvshkna5vd7o@linux-r8p5>
[ Applied above fixup as per Davidlohr's request ]
[ Use inttypes.h to print rlim_t fields, fixing the build on Alpine Linux / musl libc ]
[ Check if eventfd() is available, i.e. if HAVE_EVENTFD is defined ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools build feature: Check if eventfd() is available

A new 'perf bench epoll' will use this, and to disable it for older
systems, add a feature test for this API.

This is just a simple program that if successfully compiled, means that
the feature is present, at least at the library level, in a build that
sets the output directory to /tmp/build/perf (using O=/tmp/build/perf),
we end up with:

  $ ls -la /tmp/build/perf/feature/test-eventfd*
  -rwxrwxr-x. 1 acme acme 8176 Nov 21 15:58 /tmp/build/perf/feature/test-eventfd.bin
  -rw-rw-r--. 1 acme acme  588 Nov 21 15:58 /tmp/build/perf/feature/test-eventfd.d
  -rw-rw-r--. 1 acme acme    0 Nov 21 15:58 /tmp/build/perf/feature/test-eventfd.make.output
  $ ldd /tmp/build/perf/feature/test-eventfd.bin
  linux-vdso.so.1 (0x00007fff3bf3f000)
  libc.so.6 => /lib64/libc.so.6 (0x00007fa984061000)
  /lib64/ld-linux-x86-64.so.2 (0x00007fa984417000)
  $ grep eventfd -A 2 -B 2 /tmp/build/perf/FEATURE-DUMP
  feature-dwarf=1
  feature-dwarf_getlocations=1
  feature-eventfd=1
  feature-fortify-source=1
  feature-sync-compare-and-swap=1
  $

The main thing here is that in the end we'll have -DHAVE_EVENTFD in
CFLAGS, and then the 'perf bench' entry needing that API can be
selectively pruned.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Jason Baron <jbaron@akamai.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-wkeldwob7dpx6jvtuzl8164k@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bench: Move HAVE_PTHREAD_ATTR_SETAFFINITY_NP into bench.h

Both futex and epoll need this call, and can cause build failure on
systems that don't have it pthread_attr_setaffinity_np().

Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Reported-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dbueso@suse.de>
Cc: Jason Baron <jbaron@akamai.com>
Link: http://lkml.kernel.org/r/20181109210719.pr7ohayuwqmfp2wl@linux-r8p5
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf script: Share code and output format for uregs and iregs output

The iregs output was missing the newline at end as well as the leading
ABI output. This made it hard to compare the iregs and uregs values.
Instead, use a single function to output the register values and use it
for both, iregs and uregs, to ensure the output is consistent.

Before:

  perf  7049 [-01]  1343.354347:          1 cycles:ppp:
        ffffffffa7bc21ce perf_event_exec+0x18e (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7ead3 setup_new_exec+0xf3 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7cd7be5 load_elf_binary+0x395 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7e540 search_binary_handler+0x80 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7f1aa __do_execve_file.isra.13+0x58a (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7f561 do_execve+0x21 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7f596 __x64_sys_execve+0x26 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7a041cb do_syscall_64+0x5b (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa840008c entry_SYSCALL_64+0x7c (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
    AX:0x80000000    BX:0x0    CX:0x0    DX:0x7    SI:0xf    DI:0x286    BP:0xffff95bc8213a460    SP:0xffffacbf0ba97d18    IP:0xffffffffa7bc21cd FLAGS:0x28e    CS:0x10    SS:0x18    R8:0x2    R9:0x21440   R10:0x33816fb3b8c   R11:0x1   R12:0xffff95bc8213a460   R13:0xffff95bc8213a400   R14:0xffff95bc8213a400   R15:0x1  ABI:2    AX:0xffffffffffffffda    BX:0xffffffffffffffff    CX:0x7f84ad85798b    DX:0x560209699d50    SI:0x7ffe2c7a6820    DI:0x7ffe2c7a8c9b    BP:0x7ffe2c7a20d0    SP:0x7ffe2c7a2058    IP:0x7f84ad85798b FLAGS:0x206    CS:0x33    SS:0x2b    R8:0x7ffe2c7a2030    R9:0x7f84ae55f010   R10:0x8   R11:0x206   R12:0xffffffffffffffff   R13:0xffffffffffffffff   R14:0xffffffffffffffff   R15:0xffffffffffffffff

  perf  7049 [-01]  1343.354363:          1 cycles:ppp:
        ...

After:

  perf  7049 [-01]  1343.354347:          1 cycles:ppp:
        ffffffffa7bc21ce perf_event_exec+0x18e (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7ead3 setup_new_exec+0xf3 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7cd7be5 load_elf_binary+0x395 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7e540 search_binary_handler+0x80 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7f1aa __do_execve_file.isra.13+0x58a (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7f561 do_execve+0x21 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7c7f596 __x64_sys_execve+0x26 (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa7a041cb do_syscall_64+0x5b (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
        ffffffffa840008c entry_SYSCALL_64+0x7c (/lib/modules/4.20.0-rc1perf-devel-05115-gc0bc98f76e39-dirty/build/vmlinux)
    ABI:2    AX:0x80000000    BX:0x0    CX:0x0    DX:0x7    SI:0xf    DI:0x286    BP:0xffff95bc8213a460    SP:0xffffacbf0ba97d18    IP:0xffffffffa7bc21cd FLAGS:0x28e    CS:0x10    SS:0x18    R8:0x2    R9:0x21440   R10:0x33816fb3b8c   R11:0x1   R12:0xffff95bc8213a460   R13:0xffff95bc8213a400   R14:0xffff95bc8213a400   R15:0x1
    ABI:2    AX:0xffffffffffffffda    BX:0xffffffffffffffff    CX:0x7f84ad85798b    DX:0x560209699d50    SI:0x7ffe2c7a6820    DI:0x7ffe2c7a8c9b    BP:0x7ffe2c7a20d0    SP:0x7ffe2c7a2058    IP:0x7f84ad85798b FLAGS:0x206    CS:0x33    SS:0x2b    R8:0x7ffe2c7a2030    R9:0x7f84ae55f010   R10:0x8   R11:0x206   R12:0xffffffffffffffff   R13:0xffffffffffffffff   R14:0xffffffffffffffff   R15:0xffffffffffffffff

  perf  7049 [-01]  1343.354363:          1 cycles:ppp:
        ...

Signed-off-by: Milian Wolff <milian.wolff@kdab.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20181107223437.9071-1-milian.wolff@kdab.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Reduce the hardcoded .max_entries for pid_maps

While working on augmented syscalls I got into this error:

  # trace -vv --filter-pids 2469,1663 -e tools/perf/examples/bpf/augmented_raw_syscalls.c sleep 1
  <SNIP>
  libbpf: map 0 is "__augmented_syscalls__"
  libbpf: map 1 is "__bpf_stdout__"
  libbpf: map 2 is "pids_filtered"
  libbpf: map 3 is "syscalls"
  libbpf: collecting relocating info for: '.text'
  libbpf: relo for 13 value 84 name 133
  libbpf: relocation: insn_idx=3
  libbpf: relocation: find map 3 (pids_filtered) for insn 3
  libbpf: collecting relocating info for: 'raw_syscalls:sys_enter'
  libbpf: relo for 8 value 0 name 0
  libbpf: relocation: insn_idx=1
  libbpf: relo for 8 value 0 name 0
  libbpf: relocation: insn_idx=3
  libbpf: relo for 9 value 28 name 178
  libbpf: relocation: insn_idx=36
  libbpf: relocation: find map 1 (__augmented_syscalls__) for insn 36
  libbpf: collecting relocating info for: 'raw_syscalls:sys_exit'
  libbpf: relo for 8 value 0 name 0
  libbpf: relocation: insn_idx=0
  libbpf: relo for 8 value 0 name 0
  libbpf: relocation: insn_idx=2
  bpf: config program 'raw_syscalls:sys_enter'
  bpf: config program 'raw_syscalls:sys_exit'
  libbpf: create map __bpf_stdout__: fd=3
  libbpf: create map __augmented_syscalls__: fd=4
  libbpf: create map syscalls: fd=5
  libbpf: create map pids_filtered: fd=6
  libbpf: added 13 insn from .text to prog raw_syscalls:sys_enter
  libbpf: added 13 insn from .text to prog raw_syscalls:sys_exit
  libbpf: load bpf program failed: Operation not permitted
  libbpf: failed to load program 'raw_syscalls:sys_exit'
  libbpf: failed to load object 'tools/perf/examples/bpf/augmented_raw_syscalls.c'
  bpf: load objects failed: err=-4009: (Incorrect kernel version)
  event syntax error: 'tools/perf/examples/bpf/augmented_raw_syscalls.c'
                       \___ Failed to load program for unknown reason

  (add -v to see detail)
  Run 'perf list' for a list of valid events

   Usage: perf trace [<options>] [<command>]
      or: perf trace [<options>] -- <command> [<options>]
      or: perf trace record [<options>] [<command>]
      or: perf trace record [<options>] -- <command> [<options>]

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

If I then try to use strace (perf trace'ing 'perf trace' needs some more work
before its possible) to get a bit more info I get:

  # strace -e bpf trace --filter-pids 2469,1663 -e tools/perf/examples/bpf/augmented_raw_syscalls.c sleep 1
  bpf(BPF_MAP_CREATE, {map_type=BPF_MAP_TYPE_PERF_EVENT_ARRAY, key_size=4, value_size=4, max_entries=4, map_flags=0, inner_map_fd=0, map_name="__bpf_stdout__", map_ifindex=0}, 72) = 3
  bpf(BPF_MAP_CREATE, {map_type=BPF_MAP_TYPE_PERF_EVENT_ARRAY, key_size=4, value_size=4, max_entries=4, map_flags=0, inner_map_fd=0, map_name="__augmented_sys", map_ifindex=0}, 72) = 4
  bpf(BPF_MAP_CREATE, {map_type=BPF_MAP_TYPE_ARRAY, key_size=4, value_size=1, max_entries=500, map_flags=0, inner_map_fd=0, map_name="syscalls", map_ifindex=0}, 72) = 5
  bpf(BPF_MAP_CREATE, {map_type=BPF_MAP_TYPE_HASH, key_size=4, value_size=1, max_entries=512, map_flags=0, inner_map_fd=0, map_name="pids_filtered", map_ifindex=0}, 72) = 6
  bpf(BPF_PROG_LOAD, {prog_type=BPF_PROG_TYPE_TRACEPOINT, insn_cnt=57, insns=0x1223f50, license="GPL", log_level=0, log_size=0, log_buf=NULL, kern_version=KERNEL_VERSION(4, 18, 10), prog_flags=0, prog_name="sys_enter", prog_ifindex=0, expected_attach_type=BPF_CGROUP_INET_INGRESS}, 72) = 7
  bpf(BPF_PROG_LOAD, {prog_type=BPF_PROG_TYPE_TRACEPOINT, insn_cnt=18, insns=0x1224120, license="GPL", log_level=0, log_size=0, log_buf=NULL, kern_version=KERNEL_VERSION(4, 18, 10), prog_flags=0, prog_name="sys_exit", prog_ifindex=0, expected_attach_type=BPF_CGROUP_INET_INGRESS}, 72) = -1 EPERM (Operation not permitted)
  bpf(BPF_PROG_LOAD, {prog_type=BPF_PROG_TYPE_TRACEPOINT, insn_cnt=18, insns=0x1224120, license="GPL", log_level=1, log_size=262144, log_buf="", kern_version=KERNEL_VERSION(4, 18, 10), prog_flags=0, prog_name="sys_exit", prog_ifindex=0, expected_attach_type=BPF_CGROUP_INET_INGRESS}, 72) = -1 EPERM (Operation not permitted)
  bpf(BPF_PROG_LOAD, {prog_type=BPF_PROG_TYPE_KPROBE, insn_cnt=18, insns=0x1224120, license="GPL", log_level=0, log_size=0, log_buf=NULL, kern_version=KERNEL_VERSION(4, 18, 10), prog_flags=0, prog_name="sys_exit", prog_ifindex=0, expected_attach_type=BPF_CGROUP_INET_INGRESS}, 72) = -1 EPERM (Operation not permitted)
  event syntax error: 'tools/perf/examples/bpf/augmented_raw_syscalls.c'
                       \___ Failed to load program for unknown reason
  <SNIP similar output as without 'strace'>
  #

I managed to create the maps, etc, but then installing the "sys_exit" hook into
the "raw_syscalls:sys_exit" tracepoint somehow gets -EPERMed...

I then go and try reducing the size of this new table:

  +++ b/tools/perf/examples/bpf/augmented_raw_syscalls.c
  @@ -47,6 +47,17 @@ struct augmented_filename {
   #define SYS_OPEN 2
   #define SYS_OPENAT 257

  +struct syscall {
  +       bool    filtered;
  +};
  +
  +struct bpf_map SEC("maps") syscalls = {
  +       .type        = BPF_MAP_TYPE_ARRAY,
  +       .key_size    = sizeof(int),
  +       .value_size  = sizeof(struct syscall),
  +       .max_entries = 500,
  +};

And after reducing that .max_entries a tad, it works. So yeah, the "unknown
reason" should be related to the number of bytes all this is taking, reduce the
default for pid_map()s so that we can have a "syscalls" map with enough slots
for all syscalls in most arches. And take notes about this error message,
improve it :-)

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: Edward Cree <ecree@solarflare.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Martin KaFai Lau <kafai@fb.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: Yonghong Song <yhs@fb.com>
Link: https://lkml.kernel.org/n/tip-yjzhak8asumz9e9hts2dgplp@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf script: Add newline after uregs output

This change makes it much easier to easily distinguish between
consecutive samples by keeping the empty line between them, like we see
when we do not enable uregs output.

Before:

  cpp-inlining 28298 [-01] 54837.342780:    3068085 cycles:pp:
              7ffff7c96709 __hypot_finite+0xa9 (/usr/lib/libm-2.28.so)
              ...
   ABI:2    AX:0x0    BX:0x40f56cf6    CX:0x294a3ae7    ...
  cpp-inlining 28298 [-01] 54837.344493:    2881929 cycles:pp:
              7ffff7c96696 __hypot_finite+0x36 (/usr/lib/libm-2.28.so)
              ...
   ABI:2    AX:0x40d440c7    BX:0x40d440c7    CX:0x4d45e5da    ...

After:

  cpp-inlining 28298 [-01] 54837.342780:    3068085 cycles:pp:
              7ffff7c96709 __hypot_finite+0xa9 (/usr/lib/libm-2.28.so)
              ...
   ABI:2    AX:0x0    BX:0x40f56cf6    CX:0x294a3ae7    ...

  cpp-inlining 28298 [-01] 54837.344493:    2881929 cycles:pp:
              7ffff7c96696 __hypot_finite+0x36 (/usr/lib/libm-2.28.so)
              ...
   ABI:2    AX:0x40d440c7    BX:0x40d440c7    CX:0x4d45e5da    ...

Signed-off-by: Milian Wolff <milian.wolff@kdab.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20181107093705.16346-1-milian.wolff@kdab.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Revert "perf augmented_syscalls: Drop 'write', 'poll' for testing without self pid filter"

Now that we have the "filtered_pids" logic in place, no need to do this
rough filter to avoid the feedback loop from 'perf trace's own syscalls,
revert it.

This reverts commit 7ed71f124284359676b6496ae7db724fee9da753.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-88vh02cnkam0vv5f9vp02o3h@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf augmented_syscalls: Remove example hardcoded set of filtered pids

Now that 'perf trace' fills in that "filtered_pids" BPF map, remove the
set of filtered pids used as an example to test that feature.

That feature works like this:

Starting a system wide 'strace' like 'perf trace' augmented session we
noticed that lots of events take place for a pid, which ends up being
the feedback loop of perf trace's syscalls being processed by the
'gnome-terminal' process:

  # perf trace -e tools/perf/examples/bpf/augmented_raw_syscalls.c
     0.391 ( 0.002 ms): gnome-terminal/2469 read(fd: 17</dev/ptmx>, buf: 0x564b79f750bc, count: 8176) = 453
     0.394 ( 0.001 ms): gnome-terminal/2469 read(fd: 17</dev/ptmx>, buf: 0x564b79f75280, count: 7724) = -1 EAGAIN Resource temporarily unavailable
     0.438 ( 0.001 ms): gnome-terminal/2469 read(fd: 4<anon_inode:[eventfd]>, buf: 0x7fffc696aeb0, count: 16) = 8
     0.519 ( 0.001 ms): gnome-terminal/2469 read(fd: 17</dev/ptmx>, buf: 0x564b79f75280, count: 7724) = 114
     0.522 ( 0.001 ms): gnome-terminal/2469 read(fd: 17</dev/ptmx>, buf: 0x564b79f752f1, count: 7611) = -1 EAGAIN Resource temporarily unavailable
  ^C

So we can use --filter-pids to get rid of that one, and in this case what is
being used to implement that functionality is that "filtered_pids" BPF map that
the tools/perf/examples/bpf/augmented_raw_syscalls.c created and that 'perf trace'
bpf loader noticed and created a "struct bpf_map" associated that then got populated
by 'perf trace':

  # perf trace --filter-pids 2469 -e tools/perf/examples/bpf/augmented_raw_syscalls.c
     0.020 ( 0.002 ms): gnome-shell/1663 epoll_pwait(epfd: 12<anon_inode:[eventpoll]>, events: 0x7ffd8f3ef960, maxevents: 32, sigsetsize: 8) = 1
     0.025 ( 0.002 ms): gnome-shell/1663 read(fd: 24</dev/input/event4>, buf: 0x560c01bb8240, count: 8112) = 48
     0.029 ( 0.001 ms): gnome-shell/1663 read(fd: 24</dev/input/event4>, buf: 0x560c01bb8258, count: 8088) = -1 EAGAIN Resource temporarily unavailable
     0.032 ( 0.001 ms): gnome-shell/1663 read(fd: 24</dev/input/event4>, buf: 0x560c01bb8240, count: 8112) = -1 EAGAIN Resource temporarily unavailable
     0.040 ( 0.003 ms): gnome-shell/1663 recvmsg(fd: 46<socket:[35893]>, msg: 0x7ffd8f3ef950) = -1 EAGAIN Resource temporarily unavailable
    21.529 ( 0.002 ms): gnome-shell/1663 epoll_pwait(epfd: 5<anon_inode:[eventpoll]>, events: 0x7ffd8f3ef960, maxevents: 32, sigsetsize: 8) = 1
    21.533 ( 0.004 ms): gnome-shell/1663 recvmsg(fd: 82<socket:[42826]>, msg: 0x7ffd8f3ef7b0, flags: DONTWAIT|CMSG_CLOEXEC) = 236
    21.581 ( 0.006 ms): gnome-shell/1663 ioctl(fd: 8</dev/dri/card0>, cmd: DRM_I915_GEM_BUSY, arg: 0x7ffd8f3ef060) = 0
    21.605 ( 0.020 ms): gnome-shell/1663 ioctl(fd: 8</dev/dri/card0>, cmd: DRM_I915_GEM_CREATE, arg: 0x7ffd8f3eeea0) = 0
    21.626 ( 0.119 ms): gnome-shell/1663 ioctl(fd: 8</dev/dri/card0>, cmd: DRM_I915_GEM_SET_DOMAIN, arg: 0x7ffd8f3eee94) = 0
    21.746 ( 0.081 ms): gnome-shell/1663 ioctl(fd: 8</dev/dri/card0>, cmd: DRM_I915_GEM_PWRITE, arg: 0x7ffd8f3eeea0) = 0
  ^C

Oops, yet another gnome process that is involved with the output that
'perf trace' generates, lets filter that out too:

  # perf trace --filter-pids 2469,1663 -e tools/perf/examples/bpf/augmented_raw_syscalls.c
         ? (         ): wpa_supplicant/1366  ... [continued]: select()) = 0 Timeout
     0.006 ( 0.002 ms): wpa_supplicant/1366 clock_gettime(which_clock: BOOTTIME, tp: 0x7fffe5b1e430) = 0
     0.011 ( 0.001 ms): wpa_supplicant/1366 clock_gettime(which_clock: BOOTTIME, tp: 0x7fffe5b1e3e0) = 0
     0.014 ( 0.001 ms): wpa_supplicant/1366 clock_gettime(which_clock: BOOTTIME, tp: 0x7fffe5b1e430) = 0
         ? (         ): gmain/1791  ... [continued]: poll()) = 0 Timeout
     0.017 (         ): wpa_supplicant/1366 select(n: 6, inp: 0x55646fed3ad0, outp: 0x55646fed3b60, exp: 0x55646fed3bf0, tvp: 0x7fffe5b1e4a0) ...
   157.879 ( 0.019 ms): gmain/1791 inotify_add_watch(fd: 8<anon_inode:inotify>, pathname: , mask: 16789454) = -1 ENOENT No such file or directory
         ? (         ): cupsd/1001  ... [continued]: epoll_pwait()) = 0
         ? (         ): gsd-color/1908  ... [continued]: poll()) = 0 Timeout
   499.615 (         ): cupsd/1001 epoll_pwait(epfd: 4<anon_inode:[eventpoll]>, events: 0x557a21166500, maxevents: 4096, timeout: 1000, sigsetsize: 8) ...
   586.593 ( 0.004 ms): gsd-color/1908 recvmsg(fd: 3<socket:[38074]>, msg: 0x7ffdef34e800) = -1 EAGAIN Resource temporarily unavailable
         ? (         ): fwupd/2230  ... [continued]: poll()) = 0 Timeout
         ? (         ): rtkit-daemon/906  ... [continued]: poll()) = 0 Timeout
         ? (         ): rtkit-daemon/907  ... [continued]: poll()) = 1
   724.603 ( 0.007 ms): rtkit-daemon/907 read(fd: 6<anon_inode:[eventfd]>, buf: 0x7f05ff768d08, count: 8) = 8
         ? (         ): ssh/5461  ... [continued]: select()) = 1
   810.431 ( 0.002 ms): ssh/5461 clock_gettime(which_clock: BOOTTIME, tp: 0x7ffd7f39f870) = 0
   ^C

Several syscall exit events for syscalls in flight when 'perf trace' started, etc. Saner :-)

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-c3tu5yg204p5mvr9kvwew07n@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Fill in BPF "filtered_pids" map when present

This makes the augmented_syscalls support the --filter-pids and
auto-filtered feedback loop pids just like when working without BPF,
i.e. with just raw_syscalls:sys_{enter,exit} and tracepoint filters.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-zc5n453sxxm0tz1zfwwelyti@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: See if there is a map named "filtered_pids"

Lookup for the first map named "filtered_pids" and, if augmenting
syscalls, i.e. if a BPF event is present and the
"__augmented_syscalls__" is present, then fill in that map with the pids
to filter, be it feedback loop ones (perf trace's pid, its father if it
is "sshd", more auto-filtered in the future) or the ones explicitely
stated in the tool command line via --filter-pids.

The code to actually fill in the map comes next.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-rhzytmw7qpe6lqyjxi1ded9t@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Add "_from_option" suffix to trace__set_filter()

As we'll need that name for a new function to set filters for both
tracepoints and BPF maps for filtering pids.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-mdkck6hf3fnd21rz2766280q@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Rename perf_evlist__set_filter* to perf_evlist__set_tp_filter*

To better reflect that this is a tracepoint filter, as opposed, for
instance to map based BPF filters.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-9138svli6ddcphrr3ymy9oy3@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf augmented_syscalls: Use pid_filter

Just to test filtering a bunch of pids, now its time to go and get that
hooked up in 'perf trace', right after we load the bpf program, if we
find a "pids_filtered" map defined, we'll populate it with the filtered
pids.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-1i9s27wqqdhafk3fappow84x@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf augmented_syscalls: Drop 'write', 'poll' for testing without self pid filter

When testing system wide tracing without filtering the syscalls called
by 'perf trace' itself we get into a feedback loop, drop for now those
two syscalls, that are the ones that 'perf trace' does in its loop for
writing the syscalls it intercepts, to help with testing till we get
that filtering in place.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-rkbu536af66dbsfx51sr8yof@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Add simple pid_filter class accessible to BPF proggies

Will be used in the augmented_raw_syscalls.c to implement 'perf trace
--filter-pids'.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-9sybmz4vchlbpqwx2am13h9e@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Add defines for map insertion/lookup

Starting with a helper for a basic pid_map(), a hash using a pid as a
key.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-gdwvq53wltvq6b3g5tdmh0cw@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf augmented_syscalls: Remove needless linux/socket.h include

Leftover from when we started augmented_raw_syscalls.c from
tools/perf/examples/bpf/augmented_syscalls.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: e58a0322dbac ("perf examples bpf: Start augmenting raw_syscalls:sys_{start,exit}")
Link: https://lkml.kernel.org/n/tip-pmts9ls2skh8n3zisb4txudd@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf augmented_syscalls: Filter on a hard coded pid

Just to show where we'll hook pid based filters, and what we use to
obtain the current pid, using a BPF getpid() equivalent.

Now we need to remove that hardcoded PID with a BPF hash map, so that we
start by filtering 'perf trace's own PID, implement the --filter-pid
functionality, etc.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-oshrcgcekiyhd0whwisxfvtv@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf bpf: Add unistd.h to the headers accessible to bpf proggies

Start with a getpid() function wrapping BPF_FUNC_get_current_pid_tgid,
idea is to mimic the system headers.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-zo8hv22onidep7tm785dzxfk@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

Merge tag 'perf-urgent-for-mingo-4.20-20181121' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux into perf/urgent

Pull perf/urgent fixes:

- Update kernel ABI headers, one of them lead to a small change in
  the ioctl 'cmd' beautifier in 'perf trace' to support the new ISO7816
  commands. (Arnaldo Carvalho de Melo)

- Restore proper cwd on return from mnt namespace (Jiri Olsa)

- Add feature check for the get_current_dir_name() function used in the
  namespace fix from Jiri, that is not available in systems such as
  Alpine Linux, which uses the  musl libc (Arnaldo Carvalho de Melo)

- Fix crash in 'perf record' when synthesizing the unit for events such
  as 'cpu-clock' (Jiri Olsa)

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

perf/x86/intel: Fix regression by default disabling perfmon v4 interrupt handling

Kyle Huey reported that 'rr', a replay debugger, broke due to the following commit:

  af3bdb991a5c ("perf/x86/intel: Add a separate Arch Perfmon v4 PMI handler")

Rework the 'disable_counter_freezing' __setup() parameter such that we
can explicitly enable/disable it and switch to default disabled.

To this purpose, rename the parameter to "perf_v4_pmi=" which is a much
better description and allows requiring a bool argument.

[ mingo: Improved the changelog some more. ]

Reported-by: Kyle Huey <me@kylehuey.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert O'Callahan <robert@ocallahan.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: acme@kernel.org
Link: http://lkml.kernel.org/r/20181120170842.GZ2131@hirez.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>

perf tools beauty ioctl: Support new ISO7816 commands

Introduced in:

  ad8c0eaa0a41 ("tty/serial_core: add ISO7816 infrastructure")

Now 'perf trace' will be able to pretty-print the 'cmd' ioctl arg when
used in capable systems with software emitting those commands.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: https://lkml.kernel.org/n/tip-7bds48dhckfnleie08mit314@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>