linux-toradex.git/tools/perf, branch v6.0-rc4

perf stat: Capitalize topdown metrics' names

2022-08-27T14:55:17+00:00

Capitalize topdown metrics' names to follow the intel SDM.

Before:

 # ./perf stat -a  sleep 1

 Performance counter stats for 'system wide':

        228,094.05 msec cpu-clock                        #  225.026 CPUs utilized
               842      context-switches                 #    3.691 /sec
               224      cpu-migrations                   #    0.982 /sec
                70      page-faults                      #    0.307 /sec
        23,164,105      cycles                           #    0.000 GHz
        29,403,446      instructions                     #    1.27  insn per cycle
         5,268,185      branches                         #   23.097 K/sec
            33,239      branch-misses                    #    0.63% of all branches
       136,248,990      slots                            #  597.337 K/sec
        32,976,450      topdown-retiring                 #     24.2% retiring
         4,651,918      topdown-bad-spec                 #      3.4% bad speculation
        26,148,695      topdown-fe-bound                 #     19.2% frontend bound
        72,515,776      topdown-be-bound                 #     53.2% backend bound
         6,008,540      topdown-heavy-ops                #      4.4% heavy operations       #     19.8% light operations
         3,934,049      topdown-br-mispredict            #      2.9% branch mispredict      #      0.5% machine clears
        16,655,439      topdown-fetch-lat                #     12.2% fetch latency          #      7.0% fetch bandwidth
        41,635,972      topdown-mem-bound                #     30.5% memory bound           #     22.7% Core bound

       1.013634593 seconds time elapsed

After:

 # ./perf stat -a  sleep 1

 Performance counter stats for 'system wide':

        228,081.94 msec cpu-clock                        #  225.003 CPUs utilized
               824      context-switches                 #    3.613 /sec
               224      cpu-migrations                   #    0.982 /sec
                67      page-faults                      #    0.294 /sec
        22,647,423      cycles                           #    0.000 GHz
        28,870,551      instructions                     #    1.27  insn per cycle
         5,167,099      branches                         #   22.655 K/sec
            32,383      branch-misses                    #    0.63% of all branches
       133,411,074      slots                            #  584.926 K/sec
        32,352,607      topdown-retiring                 #     24.3% Retiring
         4,456,977      topdown-bad-spec                 #      3.3% Bad Speculation
        25,626,487      topdown-fe-bound                 #     19.2% Frontend Bound
        70,955,316      topdown-be-bound                 #     53.2% Backend Bound
         5,834,844      topdown-heavy-ops                #      4.4% Heavy Operations       #     19.9% Light Operations
         3,738,781      topdown-br-mispredict            #      2.8% Branch Mispredict      #      0.5% Machine Clears
        16,286,803      topdown-fetch-lat                #     12.2% Fetch Latency          #      7.0% Fetch Bandwidth
        40,802,069      topdown-mem-bound                #     30.6% Memory Bound           #     22.6% Core Bound

       1.013683125 seconds time elapsed

Reviewed-by: Kan Liang 
Signed-off-by: Xing Zhengjun 
Acked-by: Ian Rogers 
Cc: Alexander Shishkin 
Cc: Andi Kleen 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Link: https://lore.kernel.org/r/20220825015458.3252239-1-zhengjun.xing@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo

perf docs: Update the documentation for the save_type filter

2022-08-27T14:55:17+00:00

Update the documentation to reflect the kernel changes.

Signed-off-by: Kan Liang 
Cc: Andi Kleen 
Cc: Ian Rogers 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20220816125612.2042397-2-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo

perf sched: Fix memory leaks in __cmd_record detected with -fsanitize=address

2022-08-27T14:55:17+00:00

An array of strings is passed to cmd_record but not freed. As
cmd_record modifies the array, add another array as a copy that can be
mutated allowing the original array contents to all be freed.

Detected with -fsanitize=address.

Signed-off-by: Ian Rogers 
Cc: Alexander Shishkin 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20220824145733.409005-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf record: Fix manpage formatting of description of support to hybrid systems

2022-08-27T14:55:17+00:00

The Intel hybrid description is written in a different style than the
rest of the perf record man page. There were some new command line
options added after it which resulted in very strange section ordering.
Move the hybrid include last.

Also the sub sections in the hybrid document don't fit the record
manpage well (especially since it talks about all kinds of unrelated
commands). I left this for now, but would be better to separate this
properly in the different man pages.

It would be better to use sub sections for the other sections, but these
don't seem to be supported in AsciiDoc?

Some of the examples are still misrendered in the manpage with an
indented troff command, but I don't know how to fix that.

In any case it's now better than before.

Signed-off-by: Andi Kleen 
Cc: zhengjun.xing@intel.com
Link: https://lore.kernel.org/r/20220818100127.249401-1-ak@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo

perf test: Stat test for repeat with a weak group

2022-08-27T14:55:17+00:00

Breaking a weak group requires multiple passes of an evlist, with
multiple runs this can introduce bugs ultimately leading to
segfaults. Add a test to cover this.

Signed-off-by: Ian Rogers 
Cc: Alexander Shishkin 
Cc: Andi Kleen 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20220822213352.75721-2-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf stat: Clear evsel->reset_group for each stat run

2022-08-27T14:55:17+00:00

If a weak group is broken then the reset_group flag remains set for
the next run. Having reset_group set means the counter isn't created
and ultimately a segfault.

A simple reproduction of this is:

  # perf stat -r2 -e '{cycles,cycles,cycles,cycles,cycles,cycles,cycles,cycles,cycles,cycles}:W

which will be added as a test in the next patch.

Fixes: 4804e0111662d7d8 ("perf stat: Use affinity for opening events")
Reviewed-by: Andi Kleen 
Signed-off-by: Ian Rogers 
Tested-by: Arnaldo Carvalho de Melo 
Tested-by: Xing Zhengjun 
Cc: Alexander Shishkin 
Cc: Andi Kleen 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20220822213352.75721-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf python: Fix build when PYTHON_CONFIG is user supplied

2022-08-27T14:55:16+00:00

The previous change to Python autodetection had a small mistake where
the auto value was used to determine the Python binary, rather than the
user supplied value. The Python binary is only used for one part of the
build process, rather than the final linking, so it was producing
correct builds in most scenarios, especially when the auto detected
value matched what the user wanted, or the system only had a valid set
of Pythons.

Change it so that the Python binary path is derived from either the
PYTHON_CONFIG value or PYTHON value, depending on what is specified by
the user. This was the original intention.

This error was spotted in a build failure an odd cross compilation
environment after commit 4c41cb46a732fe82 ("perf python: Prefer
python3") was merged.

Fixes: 630af16eee495f58 ("perf tools: Use Python devtools for version autodetection rather than runtime")
Signed-off-by: James Clark 
Acked-by: Ian Rogers 
Cc: Alexander Shishkin 
Cc: Ingo Molnar 
Cc: James Clark 
Cc: Jiri Olsa 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Link: https://lore.kernel.org/r/20220728093946.1337642-1-james.clark@arm.com
Signed-off-by: Arnaldo Carvalho de Melo

perf tools: Support reading PERF_FORMAT_LOST

2022-08-19T18:56:56+00:00

The recent kernel added lost count can be read from either read(2) or
ring buffer data with PERF_SAMPLE_READ.  As it's a variable length data
we need to access it according to the format info.

But for perf tools use cases, PERF_FORMAT_ID is always set.  So we can
only check PERF_FORMAT_LOST bit to determine the data format.

Add sample_read_value_size() and next_sample_read_value() helpers to
make it a bit easier to access.  Use them in all places where it reads
the struct sample_read_value.

Signed-off-by: Namhyung Kim 
Acked-by: Jiri Olsa 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: Peter Zijlstra 
Link: https://lore.kernel.org/r/20220819003644.508916-5-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf beauty: Update copy of linux/socket.h with the kernel sources

2022-08-19T18:30:33+00:00

To pick the changes in:

  7fa875b8e53c288d ("net: copy from user before calling __copy_msghdr")
  ebe73a284f4de8c5 ("net: Allow custom iter handler in msghdr")
  7c701d92b2b5e517 ("skbuff: carry external ubuf_info in msghdr")
  c04245328dd7e915 ("net: make __sys_accept4_file() static")

That don't result in any changes in the tables generated from that
header.

This silences this perf build warning:

  Warning: Kernel ABI header at 'tools/perf/trace/beauty/include/linux/socket.h' differs from latest version at 'include/linux/socket.h'
  diff -u tools/perf/trace/beauty/include/linux/socket.h include/linux/socket.h

Cc: David Ahern 
Cc: David S. Miller 
Cc: Dylan Yudaken 
Cc: Jakub Kicinski 
Cc: Jens Axboe 
Cc: Pavel Begunkov 
Cc: Yajun Deng 
Link: https://lore.kernel.org/lkml/YvzYs+F+Xzq8Hvvp@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf cpumap: Fix alignment for masks in event encoding

2022-08-19T18:30:28+00:00

A mask encoding of a cpu map is laid out as:

  u16 nr
  u16 long_size
  unsigned long mask[];

However, the mask may be 8-byte aligned meaning there is a 4-byte pad
after long_size. This means 32-bit and 64-bit builds see the mask as
being at different offsets. On top of this the structure is in the byte
data[] encoded as:

  u16 type
  char data[]

This means the mask's struct isn't the required 4 or 8 byte aligned, but
is offset by 2. Consequently the long reads and writes are causing
undefined behavior as the alignment is broken.

Fix the mask struct by creating explicit 32 and 64-bit variants, use a
union to avoid data[] and casts; the struct must be packed so the
layout matches the existing perf.data layout. Taking an address of a
member of a packed struct breaks alignment so pass the packed
perf_record_cpu_map_data to functions, so they can access variables with
the right alignment.

As the 64-bit version has 4 bytes of padding, optimizing writing to only
write the 32-bit version.

Committer notes:

Disable warnings about 'packed' that break the build in some arches like
riscv64, but just around that specific struct.

Signed-off-by: Ian Rogers 
Acked-by: Jiri Olsa 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Alexey Bayduraev 
Cc: Athira Jajeev 
Cc: Colin Ian King 
Cc: Dave Marchevsky 
Cc: German Gomez 
Cc: Gustavo A. R. Silva 
Cc: Ingo Molnar 
Cc: James Clark 
Cc: Kees Kook 
Cc: Leo Yan 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Riccardo Mancini 
Cc: Song Liu 
Cc: Stephane Eranian 
Link: https://lore.kernel.org/r/20220614143353.1559597-5-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo