tools/perf/tests: Fix compilation error with strncpy in tests/tool_pmu

perf fails to compile on systems with GCC version11
as below:

In file included from /usr/include/string.h:519,
                 from /home/athir/perf-tools-next/tools/include/linux/bitmap.h:5,
                 from /home/athir/perf-tools-next/tools/perf/util/pmu.h:5,
                 from /home/athir/perf-tools-next/tools/perf/util/evsel.h:14,
                 from /home/athir/perf-tools-next/tools/perf/util/evlist.h:14,
                 from tests/tool_pmu.c:3:
In function ‘strncpy’,
    inlined from ‘do_test’ at tests/tool_pmu.c:25:3:
/usr/include/bits/string_fortified.h:95:10: error: ‘__builtin_strncpy’ specified bound 128 equals destination size [-Werror=stringop-truncation]
   95 |   return __builtin___strncpy_chk (__dest, __src, __len,
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                                   __glibc_objsize (__dest));
      |                                   ~~~~~~~~~~~~~~~~~~~~~~~~~

The compile error is from strncpy refernce in do_test:
strncpy(str, tool_pmu__event_to_str(ev), sizeof(str));

This behaviour is not observed with GCC version 8, but observed
with GCC version 11 . This is message from gcc for detecting
truncation while using strncpu. Use snprintf instead of strncpy
here to be safe.

Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: akanksha@linux.ibm.com
Cc: hbathini@linux.ibm.com
Cc: kjain@linux.ibm.com
Cc: maddy@linux.ibm.com
Cc: disgoel@linux.vnet.ibm.com
Cc: linuxppc-dev@lists.ozlabs.org
Link: https://lore.kernel.org/r/20241013173742.71882-1-atrajeev@linux.vnet.ibm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf report: Display columns Predicted/Abort/Cycles in --branch-history

The original commit message:

"
Use current sort mechanism but the real .se_cmp() just returns 0 so
that new columns "Predicted", "Abort" and "Cycles" are created in display
but actually these keys are not the sort keys.

For example:

Overhead  Source:Line   Symbol    Shared Object  Predicted  Abort  Cycles
........  ............  ........  .............  .........  .....  ......

  38.25%  div.c:45      [.] main  div            97.6%      0      3
"

Update missed commit from series "perf report: Show branch flags/cycles
in --branch-history callgraph view" to apply to current repository so that
new columns described above are visible.

Link to original series:
https://lore.kernel.org/lkml/1477876794-30749-1-git-send-email-yao.jin@linux.intel.com/

Reported-by: Dr. David Alan Gilbert <linux@treblig.org>
Suggested-by: Kan Liang <kan.liang@linux.intel.com>
Co-developed-by: Jin Yao <yao.jin@linux.intel.com>
Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Tested-by: Thomas Falcon <thomas.falcon@intel.com>
Signed-off-by: Thomas Falcon <thomas.falcon@intel.com>
Link: https://lore.kernel.org/r/20241010184046.203822-1-thomas.falcon@intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tests: Add tool PMU test

Ensure parsing with and without PMU creates events with the expected
config values. This ensures the tool.json doesn't get out of sync with
tool_pmu_event enum.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-11-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tool_pmu: Switch to standard pmu functions and json descriptions

Use the regular PMU approaches with tool json events to reduce the
amount of special tool_pmu code - tool_pmu__config_terms and
tool_pmu__for_each_event_cb are removed. Some functions remain, like
tool_pmu__str_to_event, as conveniences to metricgroups. Add
tool_pmu__skip_event/tool_pmu__num_skip_events to handle the case that
tool json events shouldn't appear on certain architectures. This isn't
done in jevents.py due to complexity in the empty-pmu-events.c and
when all vendor json is built into the tool.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-10-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf jevents: Add tool event json under a common architecture

Introduce the notion of a common architecture/model that can be used
to find event tables for common PMUs like the tool PMU. By having tool
events be json standard PMU attribute configuration, descriptions,
etc. can be used and these routines are already optimized for things
like binary searching.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-9-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tool_pmu: Move expr literals to tool_pmu

Add the expr literals like "#smt_on" as tool events, this allows stat
events to give the values. On my laptop with hyperthreading enabled:

```
$ perf stat -e "has_pmem,num_cores,num_cpus,num_cpus_online,num_dies,num_packages,smt_on,system_tsc_freq" true

 Performance counter stats for 'true':

                 0      has_pmem
                 8      num_cores
                16      num_cpus
                16      num_cpus_online
                 1      num_dies
                 1      num_packages
                 1      smt_on
     2,496,000,000      system_tsc_freq

       0.001113637 seconds time elapsed

       0.001218000 seconds user
       0.000000000 seconds sys
```

And with hyperthreading disabled:
```
$ perf stat -e "has_pmem,num_cores,num_cpus,num_cpus_online,num_dies,num_packages,smt_on,system_tsc_freq" true

 Performance counter stats for 'true':

                 0      has_pmem
                 8      num_cores
                16      num_cpus
                 8      num_cpus_online
                 1      num_dies
                 1      num_packages
                 0      smt_on
     2,496,000,000      system_tsc_freq

       0.000802115 seconds time elapsed

       0.000000000 seconds user
       0.000806000 seconds sys
```

As zero matters for these values, in stat-display
should_skip_zero_counter only skip the zero value if it is not the
first aggregation index.

The tool event implementations are used in expr but not evaluated as
events for simplicity. Also core_wide isn't made a tool event as it
requires command line parameters.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-8-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tool_pmu: Rename perf_tool_event__* to tool_pmu__*

Now the events are associated with the tool PMU, rename the functions
to reflect this.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-7-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tool_pmu: Rename enum perf_tool_event to tool_pmu_event

To better reflect the events listed are from the tool PMU. Rename the
enum values from PERF_TOOL_* to TOOL_PMU__EVENT_*.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-6-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tool_pmu: Factor tool events into their own PMU

Rather than treat tool events as a special kind of event, create a
tool only PMU where the events/aliases match the existing
duration_time, user_time and system_time events. Remove special
parsing and printing support for the tool events, but add function
calls for when PMU functions are called on a tool_pmu.

Move the tool PMU code in evsel into tool_pmu.c to better encapsulate
the tool event behavior in that file.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-5-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf parse-events: Expose/rename config_term_name

Expose config_term_name as parse_events__term_type_str so that PMUs not
in pmu.c may access it.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-4-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf pmu: Allow hardcoded terms to be applied to attributes

Hard coded terms like "config=10" are skipped by perf_pmu__config
assuming they were already applied to a perf_event_attr by parse
event's config_attr function. When doing a reverse number to name
lookup in perf_pmu__name_from_config, as the hardcoded terms aren't
applied the config value is incorrect leading to misses or false
matches. Fix this by adding a parameter to have perf_pmu__config apply
hardcoded terms too (not just in parse event's config_term_common).

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-3-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf pmu: Simplify an asprintf error message

Use ifs rather than ?: to avoid a large compound statement.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20241002032016.333748-2-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tools: Remove unused color_fwrite_lines

color_fwrite_lines() was added by 2009's commit
8fc0321f1ad0 ("perf_counter tools: Add color terminal output support")

but has never been used.

Remove it.

Signed-off-by: Dr. David Alan Gilbert <linux@treblig.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20241009003938.254936-1-linux@treblig.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test x86: Fix typo in intel-pt-test

Change function name "is_hydrid" to "is_hybrid".

Signed-off-by: Thomas Falcon <thomas.falcon@intel.com>
Reviewed-by: Adrian Hunter <adrian.hunter@intel.com>
Link: https://lore.kernel.org/r/20241007194758.78659-1-thomas.falcon@intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf probe: Remove unused add_perf_probe_events

add_perf_probe_events has been unused since 2015's commit
b02137cc6550 ("perf probe: Move print logic into cmd_probe()")
which confusingly now uses perf_add_probe_events.

Remove it.

Signed-off-by: Dr. David Alan Gilbert <linux@treblig.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Link: https://lore.kernel.org/r/20240929010659.430208-1-linux@treblig.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test attr: Add back missing topdown events

With the patch 0b6c5371c03c "Add missing topdown metrics events" eight
topdown metric events with numbers ranging from 0x8000 to 0x8700 were
added to the test since they were added as 'perf stat' default events.
Later the patch 951efb9976ce "Update no event/metric expectations" kept
only 4 of those events(0x8000-0x8300).

Currently, the topdown events with numbers 0x8400 to 0x8700 are missing
from the list of expected events resulting in a failure. Add back the
missing topdown events.

Fixes: 951efb9976ce ("perf test attr: Update no event/metric expectations")
Signed-off-by: Veronika Molnarova <vmolnaro@redhat.com>
Tested-by: Ian Rogers <irogers@google.com>
Cc: mpetlan@redhat.com
Link: https://lore.kernel.org/r/20240311081611.7835-1-vmolnaro@redhat.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf arm-spe: Dump metadata with version 2

This commit dumps metadata with version 2. It dumps metadata for header
and per CPU data respectively in the arm_spe_print_info() function to
support metadata version 2 format.

After:

  0 0 0x3c0 [0x1b0]: PERF_RECORD_AUXTRACE_INFO type: 4
  Header version     :2
  Header size        :4
  PMU type v2        :13
  CPU number         :8
    Magic            :0x1010101010101010
    CPU #            :0
    Num of params    :3
    MIDR             :0x410fd801
    PMU Type         :-1
    Min Interval     :0
    Magic            :0x1010101010101010
    CPU #            :1
    Num of params    :3
    MIDR             :0x410fd801
    PMU Type         :-1
    Min Interval     :0
    Magic            :0x1010101010101010
    CPU #            :2
    Num of params    :3
    MIDR             :0x410fd870
    PMU Type         :13
    Min Interval     :1024
    Magic            :0x1010101010101010
    CPU #            :3
    Num of params    :3
    MIDR             :0x410fd870
    PMU Type         :13
    Min Interval     :1024
    Magic            :0x1010101010101010
    CPU #            :4
    Num of params    :3
    MIDR             :0x410fd870
    PMU Type         :13
    Min Interval     :1024
    Magic            :0x1010101010101010
    CPU #            :5
    Num of params    :3
    MIDR             :0x410fd870
    PMU Type         :13
    Min Interval     :1024
    Magic            :0x1010101010101010
    CPU #            :6
    Num of params    :3
    MIDR             :0x410fd850
    PMU Type         :-1
    Min Interval     :0
    Magic            :0x1010101010101010
    CPU #            :7
    Num of params    :3
    MIDR             :0x410fd850
    PMU Type         :-1
    Min Interval     :0

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Besar Wicaksono <bwicaksono@nvidia.com>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20241003184302.190806-6-leo.yan@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf arm-spe: Support metadata version 2

This commit is to support metadata version 2 and at the meantime it is
backward compatible for version 1's format.

The metadata version 1 doesn't include the ARM_SPE_HEADER_VERSION field.
As version 1 is fixed with two u64 fields, by checking the metadata
size, it distinguishes the metadata is version 1 or version 2 (and any
new versions if later will have). For version 2, it reads out CPU number
and retrieves the metadata info for every CPU.

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Besar Wicaksono <bwicaksono@nvidia.com>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20241003184302.190806-5-leo.yan@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf arm-spe: Save per CPU information in metadata

Save the Arm SPE information on a per-CPU basis. This approach is easier
in the decoding phase for retrieving metadata based on the CPU number of
every Arm SPE record.

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Besar Wicaksono <bwicaksono@nvidia.com>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20241003184302.190806-4-leo.yan@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf arm-spe: Calculate meta data size

The metadata is designed to contain a header and per CPU information.

The arm_spe_find_cpus() function is introduced to identify how many CPUs
support ARM SPE. Based on the CPU number, calculates the metadata size.

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Besar Wicaksono <bwicaksono@nvidia.com>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20241003184302.190806-3-leo.yan@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf arm-spe: Define metadata header version 2

The first version's metadata header structure doesn't include a field to
indicate a header version, which is not friendly for extension.

Define the metadata version 2 format with a new header structure and
extend per CPU's metadata. In the meantime, the old metadata header will
still be supported for backward compatibility.

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Besar Wicaksono <bwicaksono@nvidia.com>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20241003184302.190806-2-leo.yan@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf list: update option desc in man page

There is a difference between the SYNOPSIS section of the help message
and the man page (tools/perf/Documentation/perf-list.txt) for the perf
list command. After checking, we found that the help message reflected
the latest specifications. Therefore, revised the SYNOPSIS section of
the man page to match the help message.

Signed-off-by: Yoshihiro Furudera <fj5100bi@fujitsu.com>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Liang
Link: https://lore.kernel.org/r/20241003002404.2592094-1-fj5100bi@fujitsu.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Restore sample rate for perf_event_attr

Test "Setup struct perf_event_attr" consists of multiple test cases that
can affect the max sample rate value for perf events. Some test cases
check this value as it should not be lowered under the set minimum for
the given test. Currently, it is possible for the test cases to affect
each other as the previous tests can lower the sample rate, leading to
a possible failure of some of the future test cases as the value is not
restored at any point.

   # 10: Setup struct perf_event_attr:
    --- start ---
    test child forked, pid 104220
    Using CPUID 0x00000000413fd0c1
    running './tests/attr/test-record-C0'
    Current sample rate: 10000
    running './tests/attr/test-record-basic'
    Current sample rate: 900
    running './tests/attr/test-record-branch-any'
    Current sample rate: 600
    running './tests/attr/test-record-dummy-C0'
    Current sample rate: 600
    expected sample_period=4000, got 600
    FAILED './tests/attr/test-record-dummy-C0' - match failure

Restore the max sample rate value for perf events to a reasonable value
before each test case if its value was lowered too much to ensure the
same conditions for each test case.

   # 10: Setup struct perf_event_attr:
    --- start ---
    test child forked, pid 107222
    Using CPUID 0x00000000413fd0c1
    running './tests/attr/test-record-C0'
    Current sample rate: 10000
    running './tests/attr/test-record-basic'
    Current sample rate: 800
    running './tests/attr/test-record-branch-any'
    Current sample rate: 700
    unsupp  './tests/attr/test-record-branch-any'
    running './tests/attr/test-record-branch-filter-any'
    Current sample rate: 10000
    running './tests/attr/test-record-count'
    Current sample rate: 10000
    running './tests/attr/test-record-data'
    Current sample rate: 600
    running './tests/attr/test-record-dummy-C0'
    Current sample rate: 800
    running './tests/attr/test-record-freq'
    Current sample rate: 10000
    ...

Cc: Michael Petlan <mpetlan@redhat.com>
Cc: Radostin Stoyanov <rstoyano@redhat.com>
Signed-off-by: Veronika Molnarova <vmolnaro@redhat.com>
Link: https://lore.kernel.org/r/20241003125136.15918-1-vmolnaro@redhat.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf trace: Keep exited threads for summary

Since 9ffa6c7512ca ("perf machine thread: Remove exited threads by
default") perf cleans exited threads up, but as said, sometimes they
are necessary to be kept. The mentioned commit does not cover all the
cases, we also need the information to construct the summary table in
perf-trace.

Before:
    # perf trace -s true

     Summary of events:

After:
    # perf trace -s -- true

     Summary of events:

     true (383382), 64 events, 91.4%

       syscall            calls  errors  total       min       avg       max       stddev
                                         (msec)    (msec)    (msec)    (msec)        (%)
       --------------- --------  ------ -------- --------- --------- ---------     ------
       mmap                   8      0     0.150     0.013     0.019     0.031     11.90%
       mprotect               3      0     0.045     0.014     0.015     0.017      6.47%
       openat                 2      0     0.014     0.006     0.007     0.007      9.73%
       munmap                 1      0     0.009     0.009     0.009     0.009      0.00%
       access                 1      1     0.009     0.009     0.009     0.009      0.00%
       pread64                4      0     0.006     0.001     0.001     0.002      4.53%
       fstat                  2      0     0.005     0.001     0.002     0.003     37.59%
       arch_prctl             2      1     0.003     0.001     0.002     0.002     25.91%
       read                   1      0     0.003     0.003     0.003     0.003      0.00%
       close                  2      0     0.003     0.001     0.001     0.001      3.86%
       brk                    1      0     0.002     0.002     0.002     0.002      0.00%
       rseq                   1      0     0.001     0.001     0.001     0.001      0.00%
       prlimit64              1      0     0.001     0.001     0.001     0.001      0.00%
       set_robust_list        1      0     0.001     0.001     0.001     0.001      0.00%
       set_tid_address        1      0     0.001     0.001     0.001     0.001      0.00%
       execve                 1      0     0.000     0.000     0.000     0.000      0.00%

[namhyung: simplified the condition]

Fixes: 9ffa6c7512ca ("perf machine thread: Remove exited threads by default")
Reported-by: Veronika Molnarova <vmolnaro@redhat.com>
Signed-off-by: Michael Petlan <mpetlan@redhat.com>
Link: https://lore.kernel.org/r/20240927151926.399474-1-mpetlan@redhat.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf/test: perf test 86 fails on s390

Command perf test 86 fails on s390:
 # perf test -F 86
 ping 868299 [007] 28248.013596: probe_libc:inet_pton_1: (3ff95948020)
 3ff95948020 inet_pton+0x0 (inlined)
 3ff9595e6e7 text_to_binary_address+0x1007 (inlined)
 3ff9595e6e7 gaih_inet+0x1007 (inlined)
 FAIL: expected backtrace entry \
 "main\+0x[[:xdigit:]]+[[:space:]]\(.*/bin/ping.*\)$"
 got "3ff9595e6e7 gaih_inet+0x1007 (inlined)"
 86: probe libc's inet_pton & backtrace it with ping  : FAILED!
 #

The root cause is a new stack layout, two functions have been added
as seen below.

 # perf script | tac | grep -m1 '^ping' -B9 | tac
 ping  866856 [007] 25979.494921: probe_libc:inet_pton: (3ff8ec48020)
     3ff8ec48020 inet_pton+0x0 (inlined)
 new -->     3ff8ec5e6e7 text_to_binary_address+0x1007 (inlined)
 new -->     3ff8ec5e6e7 gaih_inet+0x1007 (inlined)
     3ff8ec5e6e7 getaddrinfo+0x1007 (/usr/lib64/libc.so.6)
     2aa3fe04bf5 main+0xff5 (/usr/bin/ping)
     3ff8eb34a5b __libc_start_call_main+0x8b (/usr/lib64/libc.so.6)
     3ff8eb34b5d __libc_start_main@GLIBC_2.2+0xad (inlined)
     2aa3fe06a1f [unknown] (/usr/bin/ping)

 #

The new functions in the call chain are:
 - text_to_binary_address()
 - gaih_inet().

Both functions are inlined and do not show up in the output
of the nm command:

  # nm -a /usr/lib64/libc.so.6 | \
grep -E '(text_to_binary_address|gaih_inet)$'
  #

There is no possibility to add these 2 functions depending on their
existance in the C library.

Add text_to_binary_address() and gaih_inet() to the list of
expected functions in an compatible way and extend the regular
expression. On s390 the backtrace can now be

   Before                   After
   probe_libc:inet_pton     probe_libc:inet_pton
   inet_pton     inet_pton
   getaddrinfo     getaddrinfo | text_to_binary_address
   main     main | gaih_inet

Output after:
 # perf test -F 86
 86: probe libc's inet_pton & backtrace it with ping  : Ok
 #

Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Cc: agordeev@linux.ibm.com
Cc: gor@linux.ibm.com
Cc: hca@linux.ibm.com
Cc: sumanthk@linux.ibm.com
Link: https://lore.kernel.org/r/20241001124224.3370306-1-tmricht@linux.ibm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

tools/perf: Allow inherit + PERF_SAMPLE_READ when opening events

The "perf record" tool will now default to this new mode if the user
specifies a sampling group when not in system-wide mode, and when
"--no-inherit" is not specified.

This change updates evsel to allow the combination of inherit
and PERF_SAMPLE_READ.

A fallback is implemented for kernel versions where this feature is not
supported.

Signed-off-by: Ben Gainey <ben.gainey@arm.com>
Cc: james.clark@arm.com
Link: https://lore.kernel.org/r/20241001121505.1009685-3-ben.gainey@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

tools/perf: Correctly calculate sample period for inherited SAMPLE_READ values

Sample period calculation in deliver_sample_value is updated to
calculate the per-thread period delta for events that are inherit +
PERF_SAMPLE_READ. When the sampling event has this configuration, the
read_format.id is used with the tid from the sample to lookup the
storage of the previously accumulated counter total before calculating
the delta. All existing valid configurations where read_format.value
represents some global value continue to use just the read_format.id to
locate the storage of the previously accumulated total.

perf_sample_id is modified to support tracking per-thread
values, along with the existing global per-id values. In the
per-thread case, values are stored in a hash by tid within the
perf_sample_id, and are dynamically allocated as the number is not known
ahead of time.

Signed-off-by: Ben Gainey <ben.gainey@arm.com>
Cc: james.clark@arm.com
Link: https://lore.kernel.org/r/20241001121505.1009685-2-ben.gainey@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Skip not fail syscall tp fields test when insufficient permissions

Clean up return value to be TEST_* rather than unspecific integer. Add
test case skip reason. Skip test if EACCES comes back from
evsel__newtp.

Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20241001052327.7052-5-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Skip not fail tp fields test when insufficient permissions

Clean up return value to be TEST_* rather than unspecific integer. Add
test case skip reason. Skip test if EACCES comes back from
evsel__newtp.

Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20241001052327.7052-4-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Fix memory leaks on event-times error paths

These error paths occur without sufficient permissions. Fix the memory
leaks to make leak sanitizer happier.

Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20241001052327.7052-3-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf stat: Fix affinity memory leaks on error path

Missed cleanup when an error occurs.

Fixes: 49de179577e7 ("perf stat: No need to setup affinities when starting a workload")
Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20241001052327.7052-2-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf jevents: Don't stop at the first matched pmu when searching a events table

The "perf all PMU test" fails on a Coffee Lake machine.

The failure is caused by the below change in the commit e2641db83f18
("perf vendor events: Add/update skylake events/metrics").

+    {
+        "BriefDescription": "This 48-bit fixed counter counts the UCLK cycles",
+        "Counter": "FIXED",
+        "EventCode": "0xff",
+        "EventName": "UNC_CLOCK.SOCKET",
+        "PerPkg": "1",
+        "PublicDescription": "This 48-bit fixed counter counts the UCLK cycles.",
+        "Unit": "cbox_0"
     }

The other cbox events have the unit name "CBOX", while the fixed counter
has a unit name "cbox_0". So the events_table will maintain separate
entries for cbox and cbox_0.

The perf_pmus__print_pmu_events() calculates the total number of events,
allocate an aliases buffer, store all the events into the buffer, sort,
and print all the aliases one by one.

The problem is that the calculated total number of events doesn't match
the stored events in the aliases buffer.

The perf_pmu__num_events() is used to calculate the number of events. It
invokes the pmu_events_table__num_events() to go through the entire
events_table to find all events. Because of the
pmu_uncore_alias_match(), the suffix of uncore PMU will be ignored. So
the events for cbox and cbox_0 are all counted.

When storing events into the aliases buffer, the
perf_pmu__for_each_event() only process the events for cbox.

Since a bigger buffer was allocated, the last entry are all 0.
When printing all the aliases, null will be outputted, and trigger the
failure.

The mismatch was introduced from the commit e3edd6cf6399 ("perf
pmu-events: Reduce processed events by passing PMU"). The
pmu_events_table__for_each_event() stops immediately once a pmu is set.
But for uncore, especially this case, the method is wrong and mismatch
what perf does in the perf_pmu__num_events().

With the patch,
$ perf list pmu | grep -A 1 clock.socket
   unc_clock.socket
        [This 48-bit fixed counter counts the UCLK cycles. Unit: uncore_cbox_0
$ perf test "perf all PMU test"
  107: perf all PMU test                                               : Ok

Reported-by: kernel test robot <oliver.sang@intel.com>
Closes: https://lore.kernel.org/all/202407101021.2c8baddb-oliver.sang@intel.com/
Fixes: e3edd6cf6399 ("perf pmu-events: Reduce processed events by passing PMU")
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Sandipan Das <sandipan.das@amd.com>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: Xu Yang <xu.yang_2@nxp.com>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20241001021431.814811-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tests: Add more topdown events regroup tests

Add more test cases to cover all supported topdown events regroup cases.

Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Yongwei Ma <yongwei.ma@intel.com>
Link: https://lore.kernel.org/r/20240913084712.13861-7-dapeng1.mi@linux.intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tests: Add topdown events counting and sampling tests

Add counting and leader sampling tests to verify topdown events including
raw format can be reordered correctly.

Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Yongwei Ma <yongwei.ma@intel.com>
Link: https://lore.kernel.org/r/20240913084712.13861-6-dapeng1.mi@linux.intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf tests: Add leader sampling test in record tests

Add leader sampling test to validate event counts are captured into
record and the count value is consistent.

Suggested-by: Kan Liang <kan.liang@linux.intel.com>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Yongwei Ma <yongwei.ma@intel.com>
Link: https://lore.kernel.org/r/20240913084712.13861-5-dapeng1.mi@linux.intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf x86/topdown: Don't move topdown metric events in group

when running below perf command, we say error is reported.

perf record -e "{slots,instructions,topdown-retiring}:S" -vv -C0 sleep 1

------------------------------------------------------------
perf_event_attr:
  type                             4 (cpu)
  size                             168
  config                           0x400 (slots)
  sample_type                      IP|TID|TIME|READ|CPU|PERIOD|IDENTIFIER
  read_format                      ID|GROUP|LOST
  disabled                         1
  sample_id_all                    1
  exclude_guest                    1
------------------------------------------------------------
sys_perf_event_open: pid -1  cpu 0  group_fd -1  flags 0x8 = 5
------------------------------------------------------------
perf_event_attr:
  type                             4 (cpu)
  size                             168
  config                           0x8000 (topdown-retiring)
  { sample_period, sample_freq }   4000
  sample_type                      IP|TID|TIME|READ|CPU|PERIOD|IDENTIFIER
  read_format                      ID|GROUP|LOST
  freq                             1
  sample_id_all                    1
  exclude_guest                    1
------------------------------------------------------------
sys_perf_event_open: pid -1  cpu 0  group_fd 5  flags 0x8
sys_perf_event_open failed, error -22

Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for
event (topdown-retiring).

The reason of error is that the events are regrouped and
topdown-retiring event is moved to closely after the slots event and
topdown-retiring event needs to do the sampling, but Intel PMU driver
doesn't support to sample topdown metrics events.

For topdown metrics events, it just requires to be in a group which has
slots event as leader. It doesn't require topdown metrics event must be
closely after slots event. Thus it's a overkill to move topdown metrics
event closely after slots event in events regrouping and furtherly cause
the above issue.

Thus don't move topdown metrics events forward if they are already in a
group.

Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Yongwei Ma <yongwei.ma@intel.com>
Link: https://lore.kernel.org/r/20240913084712.13861-4-dapeng1.mi@linux.intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf x86/topdown: Correct leader selection with sample_read enabled

Addresses an issue where, in the absence of a topdown metrics event
within a sampling group, the slots event was incorrectly bypassed as
the sampling leader when sample_read was enabled.

perf record -e '{slots,branches}:S' -c 10000 -vv sleep 1

In this case, the slots event should be sampled as leader but the
branches event is sampled in fact like the verbose output shows.

perf_event_attr:
  type                             4 (cpu)
  size                             168
  config                           0x400 (slots)
  sample_type                      IP|TID|TIME|READ|CPU|IDENTIFIER
  read_format                      ID|GROUP|LOST
  disabled                         1
  sample_id_all                    1
  exclude_guest                    1
------------------------------------------------------------
sys_perf_event_open: pid -1  cpu 0  group_fd -1  flags 0x8 = 5
------------------------------------------------------------
perf_event_attr:
  type                             0 (PERF_TYPE_HARDWARE)
  size                             168
  config                           0x4 (PERF_COUNT_HW_BRANCH_INSTRUCTIONS)
  { sample_period, sample_freq }   10000
  sample_type                      IP|TID|TIME|READ|CPU|IDENTIFIER
  read_format                      ID|GROUP|LOST
  sample_id_all                    1
  exclude_guest                    1

The sample period of slots event instead of branches event is reset to
0.

This fix ensures the slots event remains the leader under these
conditions.

Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Yongwei Ma <yongwei.ma@intel.com>
Link: https://lore.kernel.org/r/20240913084712.13861-3-dapeng1.mi@linux.intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf x86/topdown: Complete topdown slots/metrics events check

It's not complete to check whether an event is a topdown slots or
topdown metrics event by only comparing the event name since user
may assign the event by RAW format, e.g.

perf stat -e '{instructions,cpu/r400/,cpu/r8300/}' sleep 1

 Performance counter stats for 'sleep 1':

     <not counted>      instructions
     <not counted>      cpu/r400/
   <not supported>      cpu/r8300/

       1.002917796 seconds time elapsed

       0.002955000 seconds user
       0.000000000 seconds sys

The RAW format slots and topdown-be-bound events are not recognized and
not regroup the events, and eventually cause error.

Thus add two helpers arch_is_topdown_slots()/arch_is_topdown_metrics()
to detect whether an event is topdown slots/metrics event by comparing
the event config directly, and use these two helpers to replace the
original event name comparisons.

Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Cc: Yongwei Ma <yongwei.ma@intel.com>
Link: https://lore.kernel.org/r/20240913084712.13861-2-dapeng1.mi@linux.intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf evsel: Reduce a variables scope

In __evsel__config_callchain avoid computing arch until code path that
uses it.

Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yang Jihong <yangjihong1@huawei.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Link: https://lore.kernel.org/r/20240918223116.127386-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf vender events arm64: Use "Topdown" as topdown metric group name

HiSilicon HIP08 does support Topdown metrics but perf tool complains
when trying to count Topdown metrics:
[root@localhost tracing]# perf stat --topdown
Topdown requested but the topdown metric groups aren't present.
(See perf list the metric groups have names like TopdownL1)

It's because tool's using "Topdown" as the metric group name[1] rather
than "TopDown", so follow the convention. This is introduced by [2]
which allows to use json metrics to support --topdown function.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/builtin-stat.c?h=v6.11-rc1#n1994
[2] commit 1647cd5b8802 ("perf stat: Implement --topdown using json metrics")

Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: prime.zeng@hisilicon.com
Cc: hejunhao3@huawei.com
Cc: linuxarm@huawei.com
Cc: shameerali.kolothum.thodi@huawei.com
Link: https://lore.kernel.org/r/20240912063903.31460-1-yangyicong@huawei.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Use ARRAY_SIZE for array length

Use of macro ARRAY_SIZE to calculate array size minimizes
the redundant code and improves code reusability.

  ./tools/perf/tests/demangle-java-test.c:31:34-35: WARNING: Use ARRAY_SIZE.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Closes: https://bugzilla.openanolis.cn/show_bug.cgi?id=11173
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Link: https://lore.kernel.org/r/20240929093045.10136-1-jiapeng.chong@linux.alibaba.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf/test: Speed up test case perf annotate basic tests

perf test 70 takes a long time. One culprit is the output of command
perf annotate. Per default enabled are
 - demangle symbol names
 - interleave source code with assembly code.
Disable demangle of symbols and abort the annotation
after the first 250 lines.

This speeds up the test case considerable, for example
on s390:

Output before:
 # time  perf test 70
 70: perf annotate basic tests             : Ok
 .....
 real   2m7.467s
 user   1m26.869s
 sys    0m34.086s
 #

 Output after:
 # time perf test 70
 70: perf annotate basic tests             : Ok

 real   0m3.341s
 user   0m1.606s
 sys    0m0.362s
 #

Signed-off-by: Thomas Richter <tmricht@linux.ibm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: sumanthk@linux.ibm.com
Link: https://lore.kernel.org/r/20240917085706.249691-1-tmricht@linux.ibm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf mem: Fix printing PERF_MEM_LVLNUM_{L2_MHB|MSC}

With commit 8ec9497d3ef34 ("tools/include: Sync uapi/linux/perf.h
with the kernel sources"), 'perf mem report' gives an incorrect memory
access string.
...
0.02% 1 3644 L5 hit [.] 0x0000000000009b0e mlc [.] 0x00007fce43f59480
...

This occurs because, if no entry exists in mem_lvlnum, perf_mem__lvl_scnprintf
will default to 'L%d, lvl', which in this case for PERF_MEM_LVLNUM_L2_MHB is 0x05.
Add entries for PERF_MEM_LVLNUM_L2_MHB and PERF_MEM_LVLNUM_MSC to mem_lvlnum,
so that the correct strings are printed.
...
0.02% 1 3644 L2 MHB hit [.] 0x0000000000009b0e mlc [.] 0x00007fce43f59480
...

Fixes: 8ec9497d3ef34 ("tools/include: Sync uapi/linux/perf.h with the kernel sources")
Suggested-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Thomas Falcon <thomas.falcon@intel.com>
Reviewed-by: Leo Yan <leo.yan@arm.com>
Link: https://lore.kernel.org/r/20240926144040.77897-1-thomas.falcon@intel.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf sched replay: Remove unused parts of the code

The sleep_sem semaphore and the specific_wait field (member of sched_atom)
are initialized but not used anywhere in the code, so this patch removes
them.

The SCHED_EVENT_MIGRATION case in perf_sched__process_event() is currently
not used and is also removed.

Additionally, prev_state in add_sched_event_sleep() is marked with
__maybe_unused and is not utilized anywhere in the function. This patch
removes the parameter.

If the task_state parameter was intended for future use, it can be
reintroduced when needed.

No functionality change intended.

Signed-off-by: Madadi Vineeth Reddy <vineethr@linux.ibm.com>
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Link: https://lore.kernel.org/r/20240917090100.42783-1-vineethr@linux.ibm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

libperf: Explicitly specify install-html dependencies

install_doc of tools/lib/perf/Makefile invokes install-man,
install-html, and install-examples of
tools/lib/perf/Documentation/Makefile at once. This invocation succeeds
when make runs in serial but can fail when make runs in parallel because
while install-man of tools/lib/perf/Documentation/Makefile depends on
all, install-html depends on nothing and can run ahead of all.

Explicitly specify the dependencies of install-html to ensure that
they are resolved before install-html.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Link: https://lore.kernel.org/r/20240915-perf-v1-1-cbfd9cd1d482@daynix.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Add a test for default perf stat command

Test that one cycles event is opened for each core PMU when "perf stat"
is run without arguments.

The event line can either be output as "pmu/cycles/" or just "cycles" if
there is only one PMU. Include 2 spaces for padding in the one PMU case
to avoid matching when the word cycles is included in metric
descriptions.

Acked-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: ak@linux.intel.com
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Yanteng Si <siyanteng@loongson.cn>
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20240926144851.245903-8-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Make stat test work on DT devices

PMUs aren't listed in /sys/devices/ on DT devices, so change the search
directory to /sys/bus/event_source/devices which works everywhere. Also
add armv8_cortex_* as a known PMU type to search for to make the test
run on more devices.

Acked-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Yunseong Kim <yskelg@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: ak@linux.intel.com
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20240926144851.245903-7-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf evsel: Remove pmu_name

"evsel->pmu_name" is only ever assigned a strdup of "pmu->name", a
strdup of "evsel->pmu_name" or NULL. As such, prefer to use
"pmu->name" directly and even to directly compare PMUs than PMU
names. For safety, add some additional NULL tests.

Acked-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Ian Rogers <irogers@google.com>
[ Fix arm-spe.c usage of pmu_name and empty PMU name ]
Acked-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: ak@linux.intel.com
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20240926144851.245903-6-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf evsel x86: Make evsel__has_perf_metrics work for legacy events

Use PMU interface to better detect core PMU for legacy events. Look
for slots event on core PMU if it is appropriate for the event.

Acked-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Yunseong Kim <yskelg@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: ak@linux.intel.com
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20240926144851.245903-5-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf stat: Remove evlist__add_default_attrs use strings

add_default_atttributes would add evsels by having pre-created
perf_event_attr, however, this needed fixing for hybrid as the
extended PMU type was necessary for each core PMU. The logic for this
was in an arch specific x86 function and wasn't present for ARM,
meaning that default events weren't being opened on all PMUs on
ARM. Change the creation of the default events to use parse_events and
strings as that will open the events on all PMUs.

Rather than try to detect events on PMUs before parsing, parse the
event but skip its output in stat-display.

The previous order of hardware events was: cycles,
stalled-cycles-frontend, stalled-cycles-backend, instructions. As
instructions is a more fundamental concept the order is changed to:
instructions, cycles, stalled-cycles-frontend, stalled-cycles-backend.

Closes: https://lore.kernel.org/lkml/CAP-5=fVABSBZnsmtRn1uF-k-G1GWM-L5SgiinhPTfHbQsKXb_g@mail.gmail.com/
Acked-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Ian Rogers <irogers@google.com>
[Don't display unsupported default events except 'cycles']
Acked-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: ak@linux.intel.com
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20240926144851.245903-4-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf stat: Uniquify event name improvements

Without aggregation on Intel:
```
$ perf stat -e instructions,cycles ...
```
Will use "cycles" for the name of the legacy cycles event but as
"instructions" has a sysfs name it will and a "[cpu]" PMU suffix. This
often breaks things as the space between the event and the PMU name
look like an extra column. The existing uniquify logic was also
uniquifying in cases when all events are core and not with uncore
events, it was not correctly handling modifiers, etc.

Change the logic so that an initial pass that can disable
uniquification is run. For individual counters, disable uniquification
in more cases such as for consistency with legacy events or for
libpfm4 events. Don't use the "[pmu]" style suffix in uniquification,
always use "pmu/.../". Change how modifiers/terms are handled in the
uniquification so that they look like parse-able events.

This fixes "102: perf stat metrics (shadow stat) test:" that has been
failing due to "instructions [cpu]" breaking its column/awk logic when
values aren't aggregated. This started happening when instructions
could match a sysfs rather than a legacy event, so the fixes tag
reflects this.

Fixes: 617824a7f0f7 ("perf parse-events: Prefer sysfs/JSON hardware events over legacy")
Acked-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Ian Rogers <irogers@google.com>
[ Fix Intel TPEBS counting mode test ]
Acked-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: ak@linux.intel.com
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20240926144851.245903-3-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf evsel: Add alternate_hw_config and use in evsel__match

There are cases where we want to match events like instructions and
cycles with legacy hardware values, in particular in stat-shadow's
hard coded metrics. An evsel's name isn't a good point of reference as
it gets altered, strstr would be too imprecise and re-parsing the
event from its name is silly. Instead, hold the legacy hardware event
name, determined during parsing, in the evsel for this matching case.

Inline evsel__match2 that is only used in builtin-diff.

Acked-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Yunseong Kim <yskelg@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Yang Li <yang.lee@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: ak@linux.intel.com
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20240926144851.245903-2-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Ignore security failures in all PMU test

Refactor code to have some more error diagnosis on traps, etc. and to
do less work on each line. Add an ignore situation for security failures.

Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Athira Jajeev <atrajeev@linux.vnet.ibm.com>
Link: https://lore.kernel.org/r/20240925173013.12789-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf symbol: Do not fixup end address of labels

When it loads symbols from an ELF file, it loads label symbols which is
0 size.  Sometimes it has the same address with other symbols and might
shadow the original symbols because it fixes up the size of the symbol.

For example, in my system __do_softirq is shadowed and only accepts the
__softirqentry_text_start instead.  But it should accept __do_softirq.

  $ readelf -sW vmlinux | grep -e __do_softirq -e __softirqentry_text_start
  105089: ffffffff82000000   814 FUNC    GLOBAL DEFAULT    1 __do_softirq
  111954: ffffffff82000000     0 NOTYPE  GLOBAL DEFAULT    1 __softirqentry_text_start

  $ perf annotate --stdio __do_softirq
  Error:
  The perf.data data has no samples!

  $ perf annotate --stdio __softirqentry_text_start | head
   Percent | Source code & Disassembly of vmlinux for cycles (26 samples, percent: local period)
  ---------------------------------------------------------------------------------------------------
           : 0                0xffffffff82000000 <__softirqentry_text_start>:
      0.00 :   ffffffff82000000:        nopl    (%rax,%rax)
     30.77 :   ffffffff82000005:        pushq   %rbp
      3.85 :   ffffffff82000006:        movq    %rsp, %rbp
      0.00 :   ffffffff82000009:        pushq   %r15
      3.85 :   ffffffff8200000b:        pushq   %r14
      3.85 :   ffffffff8200000d:        pushq   %r13
      0.00 :   ffffffff8200000f:        pushq   %r12

We can ignore NOTYPE symbols in the symbols__fixup_end() so that it can
pick the __do_softirq() in choose_best_symbol().  This should be fine
since most symbols have either STT_FUNC or STT_OBJECT.

Link: https://lore.kernel.org/r/20240912224208.3360116-1-namhyung@kernel.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf vendor events arm64: imx95: add imx95_bandwidth_usage.lpddr4x metric

Except lpddr5, i.MX95 also support lpddr4x. This will add a metric for
lpddr4x.

Signed-off-by: Xu Yang <xu.yang_2@nxp.com>
Cc: shawnguo@kernel.org
Cc: will@kernel.org
Cc: james.clark@linaro.org
Cc: mike.leach@linaro.org
Cc: imx@lists.linux.dev
Cc: john.g.garry@oracle.com
Cc: kernel@pengutronix.de
Cc: s.hauer@pengutronix.de
Link: https://lore.kernel.org/r/20240924030812.3211029-1-xu.yang_2@nxp.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf stat: Stop repeating when ref_perf_stat() returns -1

Exit when run_perf_stat() returns an error to avoid continuously
repeating the same error message. It's not expected that COUNTER_FATAL
or internal errors are recoverable so there's no point in retrying.

This fixes the following flood of error messages for permission issues,
for example when perf_event_paranoid==3:
  perf stat -r 1044 -- false

  Error:
  Access to performance monitoring and observability operations is limited.
  ...
  Error:
  Access to performance monitoring and observability operations is limited.
  ...
  (repeating for 1044 times).

Signed-off-by: Levi Yun <yeoreum.yun@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: nd@arm.com
Cc: howardchu95@gmail.com
Link: https://lore.kernel.org/r/20240925132022.2650180-3-yeoreum.yun@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf stat: Close cork_fd when create_perf_stat_counter() failed

When create_perf_stat_counter() failed, it doesn't close workload.cork_fd
open in evlist__prepare_workload(). This could make too many open file
error while __run_perf_stat() repeats.

Introduce evlist__cancel_workload to close workload.cork_fd and
wait workload.child_pid until exit to clear child process
when create_perf_stat_counter() is failed.

Signed-off-by: Levi Yun <yeoreum.yun@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Cc: nd@arm.com
Cc: howardchu95@gmail.com
Link: https://lore.kernel.org/r/20240925132022.2650180-2-yeoreum.yun@arm.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf evsel: display dmesg command of showing a hardcoded path

In non-FHS compliant distros like NixOS, nothing resides in `/bin`
and `/usr/bin`. Instead dynamically symlinked into
`/run/current-system/sw/bin/`, the executable resides in `/nix/store`.

With this patch,`/bin` prefix from the dmesg command in the error
message is stripped.

Link: https://github.com/NixOS/nixpkgs/pull/258027
Signed-off-by: Masum Reza <masumrezarock100@gmail.com>
Cc: Yunseong Kim <yskelg@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yang Jihong <yangjihong1@huawei.com>
Link: https://lore.kernel.org/r/20240922112619.149429-1-masumrezarock100@gmail.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: cs-etm: Test Coresight disassembly script

Run a few samples through the disassembly script and check to see that
at least one branch instruction is printed.

Signed-off-by: James Clark <james.clark@linaro.org>
Reviewed-by: Leo Yan <leo.yan@arm.com>
Tested-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Cc: Ben Gainey <ben.gainey@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Ruidong Tian <tianruidong@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: coresight@lists.linaro.org
Cc: John Garry <john.g.garry@oracle.com>
Cc: scclevenger@os.amperecomputing.com
Link: https://lore.kernel.org/r/20240916135743.1490403-8-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf scripts python cs-etm: Add start and stop arguments

Make it possible to only disassemble a range of timestamps or sample
indexes. This will be used by the test to limit the runtime, but it's
also useful for users.

Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Cc: Ben Gainey <ben.gainey@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Ruidong Tian <tianruidong@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: coresight@lists.linaro.org
Cc: John Garry <john.g.garry@oracle.com>
Cc: scclevenger@os.amperecomputing.com
Link: https://lore.kernel.org/r/20240916135743.1490403-7-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf scripts python cs-etm: Improve arguments

Make vmlinux detection automatic and use Perf's default objdump
when -d is specified. This will make it easier for a test to use the
script without having to provide arguments. And similarly for users.

Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Cc: Ben Gainey <ben.gainey@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Ruidong Tian <tianruidong@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: coresight@lists.linaro.org
Cc: John Garry <john.g.garry@oracle.com>
Cc: scclevenger@os.amperecomputing.com
Link: https://lore.kernel.org/r/20240916135743.1490403-6-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf scripts python cs-etm: Update to use argparse

optparse is deprecated and less flexible than argparse so update it.

Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Cc: Ben Gainey <ben.gainey@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Ruidong Tian <tianruidong@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: coresight@lists.linaro.org
Cc: John Garry <john.g.garry@oracle.com>
Cc: scclevenger@os.amperecomputing.com
Link: https://lore.kernel.org/r/20240916135743.1490403-5-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf scripting python: Add function to get a config value

This can be used to get config values like which objdump Perf uses for
disassembly.

Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Cc: Ben Gainey <ben.gainey@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Ruidong Tian <tianruidong@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: coresight@lists.linaro.org
Cc: John Garry <john.g.garry@oracle.com>
Cc: scclevenger@os.amperecomputing.com
Link: https://lore.kernel.org/r/20240916135743.1490403-4-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf cs-etm: Use new OpenCSD consistency checks

Previously when the incorrect binary was used for decode, Perf would
silently continue to generate incorrect samples. With OpenCSD 1.5.4 we
can enable consistency checks that do a best effort to detect a mismatch
in the image. When one is detected a warning is printed and sample
generation stops until the trace resynchronizes with a good part of the
image.

Reported-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Closes: https://lore.kernel.org/all/20240719092619.274730-1-gankulkarni@os.amperecomputing.com/
Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Cc: Ben Gainey <ben.gainey@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Ruidong Tian <tianruidong@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: coresight@lists.linaro.org
Cc: John Garry <john.g.garry@oracle.com>
Cc: scclevenger@os.amperecomputing.com
Link: https://lore.kernel.org/r/20240916135743.1490403-3-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf cs-etm: Don't flush when packet_queue fills up

cs_etm__flush(), like cs_etm__sample() is an operation that generates a
sample and then swaps the current with the previous packet. Calling
flush after processing the queues results in two swaps which corrupts
the next sample. Therefore it wasn't appropriate to call flush here so
remove it.

Flushing is still done on a discontinuity to explicitly clear the last
branch buffer, but when the packet_queue fills up before reaching a
timestamp, that's not a discontinuity and the call to
cs_etm__process_traceid_queue() already generated samples and drained
the buffers correctly.

This is visible by looking for a branch that has the same target as the
previous branch and the following source is before the address of the
last target, which is impossible as execution would have had to have
gone backwards:

  ffff800080849d40 _find_next_and_bit+0x78 => ffff80008011cadc update_sg_lb_stats+0x94
   (packet_queue fills here before a timestamp, resulting in a flush and
    branch target ffff80008011cadc is duplicated.)
  ffff80008011cb1c update_sg_lb_stats+0xd4 => ffff80008011cadc update_sg_lb_stats+0x94
  ffff8000801117c4 cpu_util+0x24 => ffff8000801117d4 cpu_util+0x34

After removing the flush the correct branch target is used for the
second sample, and ffff8000801117c4 is no longer before the previous
address:

  ffff800080849d40 _find_next_and_bit+0x78 => ffff80008011cadc update_sg_lb_stats+0x94
  ffff80008011cb1c update_sg_lb_stats+0xd4 => ffff8000801117a0 cpu_util+0x0
  ffff8000801117c4 cpu_util+0x24 => ffff8000801117d4 cpu_util+0x34

Make sure that a final branch stack is output at the end of the trace
by calling cs_etm__end_block(). This is already done for both the
timeless decode paths.

Fixes: 21fe8dc1191a ("perf cs-etm: Add support for CPU-wide trace scenarios")
Reported-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Closes: https://lore.kernel.org/all/20240719092619.274730-1-gankulkarni@os.amperecomputing.com/
Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>
Cc: Ben Gainey <ben.gainey@arm.com>
Cc: Suzuki K Poulose <suzuki.poulose@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Ruidong Tian <tianruidong@linux.alibaba.com>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: coresight@lists.linaro.org
Cc: John Garry <john.g.garry@oracle.com>
Cc: scclevenger@os.amperecomputing.com
Link: https://lore.kernel.org/r/20240916135743.1490403-2-james.clark@linaro.org
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf test: Be more tolerant of metricgroup failures

Previously "set -e" meant any non-zero exit code from perf stat would
cause a test failure. As a non-zero exit happens when there aren't
sufficient permissions, check for this case and make the exit code
2/skip for it.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Veronika Molnarova <vmolnaro@redhat.com>
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Link: https://lore.kernel.org/r/20240502223115.2357499-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

perf trace: Mark the 'head' arg in the set_robust_list syscall as coming from user space

With that it uses the generic BTF based pretty printer:

This one we need to think about, not being acquainted with this syscall,
should we _traverse_ that list somehow? Would that be useful?

  root@number:~# perf trace -e set_robust_list sleep 1
       0.000 ( 0.004 ms): sleep/1206493 set_robust_list(head: (struct robust_list_head){.list = (struct robust_list){.next = (struct robust_list *)0x7f48a9a02a20,},.futex_offset = (long int)-32,}, len: 24) =
  root@number:~#

strace prints the default integer args:

  root@number:~# strace -e set_robust_list sleep 1
  set_robust_list(0x7efd99559a20, 24)     = 0
  +++ exited with 0 +++
  root@number:~#

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alan Maguire <alan.maguire@oracle.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org
Link: https://lore.kernel.org/lkml/ZuH6MquMraBvODRp@x1
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Mark the 'rseq' arg in the rseq syscall as coming from user space

With that it uses the generic BTF based pretty printer:

  root@number:~# grep -w rseq /sys/kernel/tracing/events/syscalls/sys_enter_rseq/format
   field:struct rseq * rseq; offset:16; size:8; signed:0;
  print fmt: "rseq: 0x%08lx, rseq_len: 0x%08lx, flags: 0x%08lx, sig: 0x%08lx", ((unsigned long)(REC->rseq)), ((unsigned long)(REC->rseq_len)), ((unsigned long)(REC->flags)), ((unsigned long)(REC->sig))
  root@number:~#

Before:

  root@number:~# perf trace -e rseq
       0.000 ( 0.017 ms): Isolated Web C/1195452 rseq(rseq: 0x7ff0ecfe6fe0, rseq_len: 32, sig: 1392848979)             = 0
      74.018 ( 0.006 ms): :1195453/1195453 rseq(rseq: 0x7f2af20fffe0, rseq_len: 32, sig: 1392848979)             = 0
    1817.220 ( 0.009 ms): Isolated Web C/1195454 rseq(rseq: 0x7f5c9ec7dfe0, rseq_len: 32, sig: 1392848979)             = 0
    2515.526 ( 0.034 ms): :1195455/1195455 rseq(rseq: 0x7f61503fffe0, rseq_len: 32, sig: 1392848979)             = 0
  ^Croot@number:~#

After:

  root@number:~# perf trace -e rseq
       0.000 ( 0.019 ms): Isolated Web C/1197258 rseq(rseq: (struct rseq){.cpu_id_start = (__u32)4,.cpu_id = (__u32)4,.mm_cid = (__u32)5,}, rseq_len: 32, sig: 1392848979) = 0
    1663.835 ( 0.019 ms): Isolated Web C/1197259 rseq(rseq: (struct rseq){.cpu_id_start = (__u32)24,.cpu_id = (__u32)24,.mm_cid = (__u32)2,}, rseq_len: 32, sig: 1392848979) = 0
    4750.444 ( 0.018 ms): Isolated Web C/1197260 rseq(rseq: (struct rseq){.cpu_id_start = (__u32)8,.cpu_id = (__u32)8,.mm_cid = (__u32)4,}, rseq_len: 32, sig: 1392848979) = 0
    4994.132 ( 0.018 ms): Isolated Web C/1197261 rseq(rseq: (struct rseq){.cpu_id_start = (__u32)10,.cpu_id = (__u32)10,.mm_cid = (__u32)1,}, rseq_len: 32, sig: 1392848979) = 0
    4997.578 ( 0.011 ms): Isolated Web C/1197263 rseq(rseq: (struct rseq){.cpu_id_start = (__u32)16,.cpu_id = (__u32)16,.mm_cid = (__u32)4,}, rseq_len: 32, sig: 1392848979) = 0
    4997.462 ( 0.014 ms): Isolated Web C/1197262 rseq(rseq: (struct rseq){.cpu_id_start = (__u32)17,.cpu_id = (__u32)17,.mm_cid = (__u32)3,}, rseq_len: 32, sig: 1392848979) = 0
  ^Croot@number:~#

We'll probably need to come up with some way for using the BTF info to
synthesize a test that then gets used and captures the output of the
'perf trace' output to check if the arguments are the ones synthesized,
randomically, for now, lets make do manually:

  root@number:~# cat ~acme/c/rseq.c
  #include <sys/syscall.h>     /* Definition of SYS_* constants */
  #include <linux/rseq.h>
  #include <errno.h>
  #include <string.h>
  #include <unistd.h>
  #include <stdint.h>
  #include <stdio.h>

  /* Provide own rseq stub because glibc doesn't */
  __attribute__((weak))
  int sys_rseq(struct rseq *rseq, __u32 rseq_len, int flags, __u32 sig)
  {
   return syscall(SYS_rseq, rseq, rseq_len, flags, sig);
  }

  int main(int argc, char *argv[])
  {
   struct rseq rseq = {
   .cpu_id_start = 12,
   .cpu_id = 34,
   .rseq_cs = 56,
   .flags = 78,
   .node_id = 90,
   .mm_cid = 12,
   };
   int err = sys_rseq(&rseq, sizeof(rseq), 98765, 0xdeadbeaf);

   printf("sys_rseq({ .cpu_id_start = 12, .cpu_id = 34, .rseq_cs = 56, .flags = 78, .node_id = 90, .mm_cid = 12, }, %d, 0) = %d (%s)\n", sizeof(rseq), err, strerror(errno));
   return err;
  }
  root@number:~# perf trace -e rseq ~acme/c/rseq
  sys_rseq({ .cpu_id_start = 12, .cpu_id = 34, .rseq_cs = 56, .flags = 78, .node_id = 90, .mm_cid = 12, }, 32, 0) = -1 (Invalid argument)
       0.000 ( 0.003 ms): rseq/1200640 rseq(rseq: (struct rseq){}, rseq_len: 32, sig: 1392848979)            =
       0.064 ( 0.001 ms): rseq/1200640 rseq(rseq: (struct rseq){.cpu_id_start = (__u32)12,.cpu_id = (__u32)34,.rseq_cs = (__u64)56,.flags = (__u32)78,.node_id = (__u32)90,.mm_cid = (__u32)12,}, rseq_len: 32, flags: 98765, sig: 3735928495) = -1 EINVAL (Invalid argument)
  root@number:~#root@number:~# cat ~acme/c/rseq.c
  #include <sys/syscall.h>     /* Definition of SYS_* constants */
  #include <linux/rseq.h>
  #include <errno.h>
  #include <string.h>
  #include <unistd.h>
  #include <stdint.h>
  #include <stdio.h>

  /* Provide own rseq stub because glibc doesn't */
  __attribute__((weak))
  int sys_rseq(struct rseq *rseq, __u32 rseq_len, int flags, __u32 sig)
  {
   return syscall(SYS_rseq, rseq, rseq_len, flags, sig);
  }

  int main(int argc, char *argv[])
  {
   struct rseq rseq = {
   .cpu_id_start = 12,
   .cpu_id = 34,
   .rseq_cs = 56,
   .flags = 78,
   .node_id = 90,
   .mm_cid = 12,
   };
   int err = sys_rseq(&rseq, sizeof(rseq), 98765, 0xdeadbeaf);

   printf("sys_rseq({ .cpu_id_start = 12, .cpu_id = 34, .rseq_cs = 56, .flags = 78, .node_id = 90, .mm_cid = 12, }, %d, 0) = %d (%s)\n", sizeof(rseq), err, strerror(errno));
   return err;
  }
  root@number:~# perf trace -e rseq ~acme/c/rseq
  sys_rseq({ .cpu_id_start = 12, .cpu_id = 34, .rseq_cs = 56, .flags = 78, .node_id = 90, .mm_cid = 12, }, 32, 0) = -1 (Invalid argument)
       0.000 ( 0.003 ms): rseq/1200640 rseq(rseq: (struct rseq){}, rseq_len: 32, sig: 1392848979)            =
       0.064 ( 0.001 ms): rseq/1200640 rseq(rseq: (struct rseq){.cpu_id_start = (__u32)12,.cpu_id = (__u32)34,.rseq_cs = (__u64)56,.flags = (__u32)78,.node_id = (__u32)90,.mm_cid = (__u32)12,}, rseq_len: 32, flags: 98765, sig: 3735928495) = -1 EINVAL (Invalid argument)
  root@number:~#

Interesting, glibc seems to be using rseq here, as in addition to the
totally fake one this test case uses, we have this one, around these
other syscalls:

     0.175 ( 0.001 ms): rseq/1201095 set_tid_address(tidptr: 0x7f6def759a10)                               = 1201095 (rseq)
     0.177 ( 0.001 ms): rseq/1201095 set_robust_list(head: 0x7f6def759a20, len: 24)                        = 0
     0.178 ( 0.001 ms): rseq/1201095 rseq(rseq: (struct rseq){}, rseq_len: 32, sig: 1392848979)            =
     0.231 ( 0.005 ms): rseq/1201095 mprotect(start: 0x7f6def93f000, len: 16384, prot: READ)               = 0
     0.238 ( 0.003 ms): rseq/1201095 mprotect(start: 0x403000, len: 4096, prot: READ)                      = 0
     0.244 ( 0.004 ms): rseq/1201095 mprotect(start: 0x7f6def99c000, len: 8192, prot: READ)

Matches strace (well, not really as the strace in fedora:40 doesn't know
about rseq, printing just integer values in hex):

  set_robust_list(0x7fbc6acc7a20, 24)     = 0
  rseq(0x7fbc6acc8060, 0x20, 0, 0x53053053) = 0
  mprotect(0x7fbc6aead000, 16384, PROT_READ) = 0
  mprotect(0x403000, 4096, PROT_READ)     = 0
  mprotect(0x7fbc6af0a000, 8192, PROT_READ) = 0
  prlimit64(0, RLIMIT_STACK, NULL, {rlim_cur=8192*1024, rlim_max=RLIM64_INFINITY}) = 0
  munmap(0x7fbc6aebd000, 81563)           = 0
  rseq(0x7fff15bb9920, 0x20, 0x181cd, 0xdeadbeaf) = -1 EINVAL (Invalid argument)
  fstat(1, {st_mode=S_IFCHR|0620, st_rdev=makedev(0x88, 0x9), ...}) = 0
  getrandom("\xd0\x34\x97\x17\x61\xc2\x2b\x10", 8, GRND_NONBLOCK) = 8
  brk(NULL)                               = 0x18ff4000
  brk(0x19015000)                         = 0x19015000
  write(1, "sys_rseq({ .cpu_id_start = 12, ."..., 136sys_rseq({ .cpu_id_start = 12, .cpu_id = 34, .rseq_cs = 56, .flags = 78, .node_id = 90, .mm_cid = 12, }, 32, 0) = -1 (Invalid argument)
  ) = 136
  exit_group(-1)                          = ?
  +++ exited with 255 +++
  root@number:~#

And also the focus for the v6.13 should be to have a better, strace
like BTF pretty printer as one of the outputs we can get from the libbpf
BTF dumper.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alan Maguire <alan.maguire@oracle.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/lkml/ZuH2K1LLt1pIDkbd@x1
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf env: Find correct branch counter info on hybrid

No event is printed in the "Branch Counter" column on hybrid machines.

For example,

  $ perf record -e "{cpu_core/branch-instructions/pp,cpu_core/branches/}:S" -j any,counter
  $ perf report --total-cycles

  # Branch counter abbr list:
  # cpu_core/branch-instructions/pp = A
  # cpu_core/branches/ = B
  # '-' No event occurs
  # '+' Event occurrences may be lost due to branch counter saturated
  #
  # Sampled Cycles%  Sampled Cycles  Avg Cycles%  Avg Cycles  Branch Counter
  # ...............  ..............  ...........  ..........  ..............
            44.54%          727.1K        0.00%           1   |+   |+   |
            36.31%          592.7K        0.00%           2   |+   |+   |
            17.83%          291.1K        0.00%           1   |+   |+   |

The branch counter information (br_cntr_width and br_cntr_nr) in the
perf_env is retrieved from the CPU_PMU_CAPS. However, the CPU_PMU_CAPS
is not available on hybrid machines. Without the width information, the
number of occurrences of an event cannot be calculated.

For a hybrid machine, the caps information should be retrieved from the
PMU_CAPS, and stored in the perf_env->pmu_caps.

Add a perf_env__find_br_cntr_info() to return the correct branch counter
information from the corresponding fields.

Committer notes:

While testing I couldn't s ee those "Branch counter" columns enabled by
pressing 'B' on the TUI, after reporting it to the list Kan explained
the situation:

<quote Kan Liang>
For a hybrid client, the "Branch Counter" feature is only supported
starting from the just released Lunar Lake. Perf falls back to only
"ANY" on your Raptor Lake.

The "The branch counter is not available" message is expected.

Here is the 'perf evlist' result from my Lunar Lake machine,

  # perf evlist -v
  cpu_core/branch-instructions/pp: type: 4 (cpu_core), size: 136, config: 0xc4 (branch-instructions), { sample_period, sample_freq }: 4000, sample_type: IP|TID|TIME|READ|PERIOD|BRANCH_STACK|IDENTIFIER, read_format: ID|GROUP|LOST, disabled: 1, freq: 1, enable_on_exec: 1, precise_ip: 2, sample_id_all: 1, exclude_guest: 1, branch_sample_type: ANY|COUNTERS
  #
</quote>

Fixes: 6f9d8d1de2c61288 ("perf script: Add branch counters")
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240909184201.553519-1-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evlist: Print hint for group

An event group is a critical relationship. There is a -g option that can
display the relationship. But it's hard for a user to know when should
this option be applied.

If there is an event group in the perf record, print a hint to suggest
the user apply the -g to display the group information.

With the patch,

  $ perf record -e "{cycles,instructions},instructions" sleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.024 MB perf.data (4 samples) ]
  $

  $ perf evlist
  cycles
  instructions
  instructions
  # Tip: use 'perf evlist -g' to show group information

  $ perf evlist -g
  {cycles,instructions}
  instructions
  $

Committer testing:

So for a perf.data file _with_ a group:

  root@number:~# perf evlist -g
  {cpu_core/branch-instructions/pp,cpu_core/branches/}
  dummy:u
  root@number:~# perf evlist
  cpu_core/branch-instructions/pp
  cpu_core/branches/
  dummy:u
  # Tip: use 'perf evlist -g' to show group information
  root@number:~#

Then for something _without_ a group, no hint:

  root@number:~# perf record ls
  <SNIP>
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.035 MB perf.data (7 samples) ]
  root@number:~# perf evlist
  cpu_atom/cycles/P
  cpu_core/cycles/P
  dummy:u
  root@number:~#

No suggestion, good.

Suggested-by: Arnaldo Carvalho de Melo <acme@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Closes: https://lore.kernel.org/lkml/ZttgvduaKsVn1r4p@x1/
Link: https://lore.kernel.org/r/20240908202847.176280-1-kan.liang@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools: Drop nonsensical -O6

-O6 is very much not-a-thing. Really, this should've been dropped
entirely in 49b3cd306e60b9d8 ("tools: Set the maximum optimization level
according to the compiler being used") instead of just passing it for
not-Clang.

Just collapse it down to -O3, instead of "-O6 unless Clang, in which case
-O3".

GCC interprets > -O3 as -O3. It doesn't even interpret > -O3 as -Ofast,
which is a good thing, given -Ofast has specific (non-)requirements for
code built using it. So, this does nothing except look a bit daft.

Remove the silliness and also save a few lines in the Makefiles accordingly.

Reviewed-by: Ian Rogers <irogers@google.com>
Reviewed-by: Jesper Juhl <jesperjuhl76@gmail.com>
Signed-off-by: Sam James <sam@gentoo.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Bill Wendling <morbo@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Justin Stitt <justinstitt@google.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: llvm@lists.linux.dev
Link: https://lore.kernel.org/r/4f01524fa4ea91c7146a41e26ceaf9dae4c127e4.1725821201.git.sam@gentoo.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf pmu: To info add event_type_desc

All PMU events are assumed to be "Kernel PMU event", however, this
isn't true for fake PMUs and won't be true with the addition of more
software PMUs. Make the PMU's type description name configurable -
largely for printing callbacks.

Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20240907050830.6752-5-irogers@google.com
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Sandipan Das <sandipan.das@amd.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Clément Le Goffic <clement.legoffic@foss.st.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Changbin Du <changbin.du@huawei.com>
Cc: Junhao He <hejunhao3@huawei.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Will Deacon <will@kernel.org>
Cc: James Clark <james.clark@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Oliver Upton <oliver.upton@linux.dev>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Athira Jajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: Tiezhu Yang <yangtiezhu@loongson.cn>
Cc: Xu Yang <xu.yang_2@nxp.com>
Cc: John Garry <john.g.garry@oracle.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Veronika Molnarova <vmolnaro@redhat.com>
Cc: Dr. David Alan Gilbert <linux@treblig.org>
Cc: linux-kernel@vger.kernel.org
Cc: linux-perf-users@vger.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf evsel: Add accessor for tool_event

Currently tool events use a dedicated variable within the evsel. Later
changes will move this to the unused struct perf_event_attr config for
these events. Add an accessor to allow the later change to be well
typed and avoid changing all uses.

Signed-off-by: Ian Rogers <irogers@google.com>
Link: https://lore.kernel.org/r/20240907050830.6752-4-irogers@google.com
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Sandipan Das <sandipan.das@amd.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Clément Le Goffic <clement.legoffic@foss.st.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Changbin Du <changbin.du@huawei.com>
Cc: Junhao He <hejunhao3@huawei.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Will Deacon <will@kernel.org>
Cc: James Clark <james.clark@linaro.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Oliver Upton <oliver.upton@linux.dev>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Athira Jajeev <atrajeev@linux.vnet.ibm.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: Tiezhu Yang <yangtiezhu@loongson.cn>
Cc: Xu Yang <xu.yang_2@nxp.com>
Cc: John Garry <john.g.garry@oracle.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Veronika Molnarova <vmolnaro@redhat.com>
Cc: Dr. David Alan Gilbert <linux@treblig.org>
Cc: linux-kernel@vger.kernel.org
Cc: linux-perf-users@vger.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf pmus: Fake PMU clean up

Rather than passing a fake PMU around, just pass that the fake PMU
should be used - true when doing testing. Move the fake PMU into
pmus.[ch] and try to abstract the PMU's properties in pmu.c, ie so
there is less "if fake_pmu" in non-PMU code. Give the fake PMU a made
up type number.

Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: Changbin Du <changbin.du@huawei.com>
Cc: Clément Le Goffic <clement.legoffic@foss.st.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Dr. David Alan Gilbert <linux@treblig.org>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: John Garry <john.g.garry@oracle.com>
Cc: Junhao He <hejunhao3@huawei.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oliver Upton <oliver.upton@linux.dev>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Sandipan Das <sandipan.das@amd.com>
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: Tiezhu Yang <yangtiezhu@loongson.cn>
Cc: Veronika Molnarova <vmolnaro@redhat.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Xu Yang <xu.yang_2@nxp.com>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: https://lore.kernel.org/r/20240907050830.6752-3-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf list: Avoid potential out of bounds memory read

If a desc string is 0 length then -1 will be out of bounds, add a
check.

Signed-off-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Cc: Benjamin Gray <bgray@linux.ibm.com>
Cc: Changbin Du <changbin.du@huawei.com>
Cc: Clément Le Goffic <clement.legoffic@foss.st.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: Dr. David Alan Gilbert <linux@treblig.org>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: James Clark <james.clark@linaro.org>
Cc: Jing Zhang <renyu.zj@linux.alibaba.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: John Garry <john.g.garry@oracle.com>
Cc: Junhao He <hejunhao3@huawei.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Leo Yan <leo.yan@linux.dev>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Oliver Upton <oliver.upton@linux.dev>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Sandipan Das <sandipan.das@amd.com>
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Cc: Tiezhu Yang <yangtiezhu@loongson.cn>
Cc: Veronika Molnarova <vmolnaro@redhat.com>
Cc: Weilin Wang <weilin.wang@intel.com>
Cc: Will Deacon <will@kernel.org>
Cc: Xu Yang <xu.yang_2@nxp.com>
Cc: Yang Jihong <yangjihong@bytedance.com>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Ze Gao <zegao2021@gmail.com>
Cc: linux-arm-kernel@lists.infradead.org
Link: https://lore.kernel.org/r/20240907050830.6752-2-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf help: Fix a typo ("bellow")

Fix a typo in comments.

Reported-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Signed-off-by: Andrew Kreimer <algonell@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: kernel-janitors@vger.kernel.org
Link: https://lore.kernel.org/r/20240907131006.18510-1-algonell@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf ftrace: Detect whether ftrace is enabled on system

To make error messages more accurate, this change detects whether ftrace is
enabled on system by checking trace file "set_ftrace_pid".

Before:

  # perf ftrace
  failed to reset ftrace
  #

After:

  # perf ftrace
  ftrace is not supported on this system
  #

Committer testing:

Doing it in an unprivileged toolbox container on Fedora 40:

Before:

  acme@number:~/git/perf-tools-next$ toolbox enter perf
  ⬢[acme@toolbox perf-tools-next]$ sudo su -
  ⬢[root@toolbox ~]# ~acme/bin/perf ftrace
  failed to reset ftrace
  ⬢[root@toolbox ~]#

After this patch:

  ⬢[root@toolbox ~]# ~acme/bin/perf ftrace
  ftrace is not supported on this system
  ⬢[root@toolbox ~]#

Maybe we could check if we are in such as situation, inside an
unprivileged container, and provide a HINT line?

Reviewed-by: James Clark <james.clark@linaro.org>
Signed-off-by: Changbin Du <changbin.du@huawei.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Changbin Du <changbin.du@huawei.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20240911100126.900779-1-changbin.du@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf test shell probe_vfs_getname: Remove extraneous '=' from probe line number regex

Thomas reported the vfs_getname perf tests failing on s/390, it seems it
was just to some extraneous '=' somehow getting into the regexp, remove
it, now:

  root@x1:~# perf test getname
   91: Add vfs_getname probe to get syscall args filenames             : Ok
   93: Use vfs_getname probe to get syscall args filenames             : FAILED!
  126: Check open filename arg using perf trace + vfs_getname          : Ok
  root@x1:~#

Second one remains a mistery, have to take some time to nail it down.

Reported-by: Thomas Richter <tmricht@linux.ibm.com>
Tested-by: Thomas Richter <tmricht@linux.ibm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Vasily Gorbik <gor@linux.ibm.com>,
Link: https://lore.kernel.org/lkml/1d7f3b7b-9edc-4d90-955c-9345428563f1@linux.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf build: Require at least clang 16.0.6 to build BPF skeletons

Howard reported problems using perf features that use BPF:

  perf $ clang -v
  Debian clang version 15.0.6
  Target: x86_64-pc-linux-gnu
  Thread model: posix
  InstalledDir: /bin
  Found candidate GCC installation: /bin/../lib/gcc/x86_64-linux-gnu/12
  Selected GCC installation: /bin/../lib/gcc/x86_64-linux-gnu/12
  Candidate multilib: .;@m64
  Selected multilib: .;@m64
  perf $ ./perf trace -e write --max-events=1
  libbpf: prog 'sys_enter_rename': BPF program load failed: Permission denied
  libbpf: prog 'sys_enter_rename': -- BEGIN PROG LOAD LOG --
  0: R1=ctx() R10=fp0

But it works with:

  perf $ clang -v
  Debian clang version 16.0.6 (15~deb12u1)
  Target: x86_64-pc-linux-gnu
  Thread model: posix
  InstalledDir: /bin
  Found candidate GCC installation: /bin/../lib/gcc/x86_64-linux-gnu/12
  Selected GCC installation: /bin/../lib/gcc/x86_64-linux-gnu/12
  Candidate multilib: .;@m64
  Selected multilib: .;@m64
  perf $ ./perf trace -e write --max-events=1
       0.000 ( 0.009 ms): gmain/1448 write(fd: 4, buf: \1\0\0\0\0\0\0\0, count: 8)                         = 8 (kworker/0:0-eve)
  perf $

So lets make that the required version, if you happen to have a slightly
older version where this work, please report so that we can adjust the
minimum required version.

Reported-by: Howard Chu <howardchu95@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alan Maguire <alan.maguire@oracle.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/lkml/ZuGL9ROeTV2uXoSp@x1
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: If a syscall arg is marked as 'const', assume it is coming _from_ userspace

We need to decide where to copy syscall arg contents, if at the
syscalls:sys_entry hook, meaning is something that is coming from
user to kernel space, or if it is a response, i.e. if it is something
the _kernel_ is filling in and thus going to userspace.

Since we have 'const' used in those syscalls, and unsure about this
being consistent, doing:

  root@number:~# echo $(grep const /sys/kernel/tracing/events/syscalls/sys_enter_*/format  | grep struct | cut -c47- | cut -d'/' -f1)
  clock_nanosleep clock_settime epoll_pwait2 futex io_pgetevents landlock_create_ruleset listmount mq_getsetattr mq_notify mq_timedreceive mq_timedsend preadv2 preadv prlimit64 process_madvise process_vm_readv process_vm_readv process_vm_writev process_vm_writev pwritev2 pwritev readv rt_sigaction rt_sigtimedwait semtimedop statmount timerfd_settime timer_settime vmsplice writev
  root@number:~#

Seems to indicate that we can use that for the ones that have the
'const' to mark it as coming from user space, do it.

Most notable/frequent syscall that now gets BTF pretty printed in a
system wide 'perf trace' session is:

  root@number:~# perf trace
     21.160 (         ): MediaSu~isor #/1028597 futex(uaddr: 0x7f49e1dfe964, op: WAIT_BITSET|PRIVATE_FLAG, utime: (struct __kernel_timespec){.tv_sec = (__kernel_time64_t)50290,.tv_nsec = (long long int)810362837,}, val3: MATCH_ANY) ...
      21.166 ( 0.000 ms): RemVidChild/6995 futex(uaddr: 0x7f49fcc7fa00, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      21.169 ( 0.001 ms): RemVidChild/6995 sendmsg(fd: 25<socket:[78915]>, msg: 0x7f49e9af9da0, flags: DONTWAIT) = 280
      21.172 ( 0.289 ms): RemVidChild/6995 futex(uaddr: 0x7f49fcc7fa58, op: WAIT_BITSET|PRIVATE_FLAG|CLOCK_REALTIME, val3: MATCH_ANY) = 0
      21.463 ( 0.000 ms): RemVidChild/6995 futex(uaddr: 0x7f49fcc7fa00, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      21.467 ( 0.001 ms): RemVidChild/6995 futex(uaddr: 0x7f49e28bb964, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
      21.160 ( 0.314 ms): MediaSu~isor #/1028597  ... [continued]: futex())                                            = 0
      21.469 (         ): RemVidChild/6995 futex(uaddr: 0x7f49fcc7fa5c, op: WAIT_BITSET|PRIVATE_FLAG|CLOCK_REALTIME, val3: MATCH_ANY) ...
      21.475 ( 0.000 ms): MediaSu~isor #/1028597 futex(uaddr: 0x7f49d0223040, op: WAKE|PRIVATE_FLAG, val: 1)           = 0
      21.478 ( 0.001 ms): MediaSu~isor #/1028597 futex(uaddr: 0x7f49e26ac964, op: WAKE|PRIVATE_FLAG, val: 1)           = 1
  ^Croot@number:~#
  root@number:~# cat /sys/kernel/tracing/events/syscalls/sys_enter_futex/format
  name: sys_enter_futex
  ID: 454
  format:
   field:unsigned short common_type; offset:0; size:2; signed:0;
   field:unsigned char common_flags; offset:2; size:1; signed:0;
   field:unsigned char common_preempt_count; offset:3; size:1; signed:0;
   field:int common_pid; offset:4; size:4; signed:1;

   field:int __syscall_nr; offset:8; size:4; signed:1;
   field:u32 * uaddr; offset:16; size:8; signed:0;
   field:int op; offset:24; size:8; signed:0;
   field:u32 val; offset:32; size:8; signed:0;
   field:const struct __kernel_timespec * utime; offset:40; size:8; signed:0;
   field:u32 * uaddr2; offset:48; size:8; signed:0;
   field:u32 val3; offset:56; size:8; signed:0;

  print fmt: "uaddr: 0x%08lx, op: 0x%08lx, val: 0x%08lx, utime: 0x%08lx, uaddr2: 0x%08lx, val3: 0x%08lx", ((unsigned long)(REC->uaddr)), ((unsigned long)(REC->op)), ((unsigned long)(REC->val)), ((unsigned long)(REC->utime)), ((unsigned long)(REC->uaddr2)), ((unsigned long)(REC->val3))
  root@number:~#

Suggested-by: Ian Rogers <irogers@google.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alan Maguire <alan.maguire@oracle.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/lkml/CAP-5=fWnuQrrBoTn6Rrn6vM_xQ2fCoc9i-AitD7abTcNi-4o1Q@mail.gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf parse-events: Remove duplicated include in parse-events.c

The header files parse-events.h is included twice in parse-events.c,
so one inclusion of each can be removed.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Reviewed-by: James Clark <james.clark@linaro.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Closes: https://bugzilla.openanolis.cn/show_bug.cgi?id=10822
Link: https://lore.kernel.org/r/20240910005522.35994-1-yang.lee@linux.alibaba.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf callchain: Allow symbols to be optional when resolving a callchain

In uses like 'perf inject' it is not necessary to gather the symbol for
each call chain location, the map for the sample IP is wanted so that
build IDs and the like can be injected. Make gathering the symbol in the
callchain_cursor optional.

For a 'perf inject -B' command this lowers the peak RSS from 54.1MB to
29.6MB by avoiding loading symbols.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Anne Macedo <retpolanne@posteo.net>
Cc: Casey Chen <cachen@purestorage.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Link: https://lore.kernel.org/r/20240909203740.143492-5-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf inject: Lazy build-id mmap2 event insertion

Add -B option that lazily inserts mmap2 events thereby dropping all
mmap events without samples. This is similar to the behavior of -b
where only build_id events are inserted when a dso is accessed in a
sample.

File size savings can be significant in system-wide mode, consider:

  $ perf record -g -a -o perf.data sleep 1
  $ perf inject -B -i perf.data -o perf.new.data
  $ ls -al perf.data perf.new.data
           5147049 perf.data
           2248493 perf.new.data

Give test coverage of the new option in pipe test.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Anne Macedo <retpolanne@posteo.net>
Cc: Casey Chen <cachen@purestorage.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Link: https://lore.kernel.org/r/20240909203740.143492-4-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf inject: Add new mmap2-buildid-all option

Add an option that allows all mmap or mmap2 events to be rewritten as
mmap2 events with build IDs.

This is similar to the existing -b/--build-ids and --buildid-all options
except instead of adding a build_id event an existing mmap/mmap2 event
is used as a template and a new mmap2 event synthesized from it.

As mmap2 events are typical this avoids the insertion of build_id
events.

Add test coverage to the pipe test.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Anne Macedo <retpolanne@posteo.net>
Cc: Casey Chen <cachen@purestorage.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Link: https://lore.kernel.org/r/20240909203740.143492-3-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf inject: Fix build ID injection

Build ID injection wasn't inserting a sample ID and aligning events to
64 bytes rather than 8. No sample ID means events are unordered and two
different build_id events for the same path, as happens when a file is
replaced, can't be differentiated.

Add in sample ID insertion for the build_id events alongside some
refactoring. The refactoring better aligns the function arguments for
different use cases, such as synthesizing build_id events without
needing to have a dso. The misc bits are explicitly passed as with
callchains the maps/dsos may span user and kernel land, so using
sample->cpumode isn't good enough.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Anne Macedo <retpolanne@posteo.net>
Cc: Casey Chen <cachen@purestorage.com>
Cc: Colin Ian King <colin.i.king@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sun Haiyong <sunhaiyong@loongson.cn>
Link: https://lore.kernel.org/r/20240909203740.143492-2-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate-data: Add pr_debug_scope()

The pr_debug_scope() is to print more information about the scope DIE
during the instruction tracking so that it can help finding relevant
debug info and the source code like inlined functions more easily.

  $ perf --debug type-profile annotate --data-type
  ...
  -----------------------------------------------------------
  find data type for 0(reg0, reg12) at set_task_cpu+0xdd
  CU for kernel/sched/core.c (die:0x1268dae)
  frame base: cfa=1 fbreg=7
  scope: [3/3] (die:12b6d28) [inlined] set_task_rq       <<<--- (here)
  bb: [9f - dd]
  var [9f] reg3 type='struct task_struct*' size=0x8 (die:0x126aff0)
  var [9f] reg6 type='unsigned int' size=0x4 (die:0x1268e0d)

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20240909214251.3033827-2-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf annotate: Treat 'call' instruction as stack operation

I found some portion of mem-store events sampled on CALL instruction
which has no memory access.  But it actually saves a return address
into stack.  It should be considered as a stack operation like RET
instruction.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://lore.kernel.org/r/20240909214251.3033827-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf build: Remove unused feature test target

llvm-version was removed in commit 56b11a2126bf ("perf bpf: Remove
support for embedding clang for compiling BPF events (-e foo.c)") but
some parts were left in the Makefile so finish removing them.

Signed-off-by: James Clark <james.clark@linaro.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Bill Wendling <morbo@google.com>
Cc: Changbin Du <changbin.du@huawei.com>
Cc: Daniel Wagner <dwagner@suse.de>
Cc: Guilherme Amadio <amadio@gentoo.org>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Justin Stitt <justinstitt@google.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Leo Yan <leo.yan@arm.com>
Cc: Manu Bretelle <chantr4@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Quentin Monnet <qmo@kernel.org>
Cc: Steinar H. Gunderson <sesse@google.com>
Link: https://lore.kernel.org/r/20240910140405.568791-2-james.clark@linaro.org
[ Removed one leftover, 'llvm-version' from FEATURE_TESTS_EXTRA ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf build: Autodetect minimum required llvm-dev version

The new LLVM addr2line feature requires a minimum version of 13 to
compile. Add a feature check for the version so that NO_LLVM=1 doesn't
need to be explicitly added. Leave the existing llvm feature check
intact because it's used by tools other than Perf.

This fixes the following compilation error when the llvm-dev version
doesn't match:

  util/llvm-c-helpers.cpp: In function 'char* llvm_name_for_code(dso*, const char*, u64)':
  util/llvm-c-helpers.cpp:178:21: error: 'std::remove_reference_t<llvm::DILineInfo>' {aka 'struct llvm::DILineInfo'} has no member named 'StartAddress'
    178 |   addr, res_or_err->StartAddress ? *res_or_err->StartAddress : 0);

Fixes: c3f8644c21df9b7d ("perf report: Support LLVM for addr2line()")
Signed-off-by: James Clark <james.clark@linaro.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Bill Wendling <morbo@google.com>
Cc: Changbin Du <changbin.du@huawei.com>
Cc: Guilherme Amadio <amadio@gentoo.org>
Cc: Ian Rogers <irogers@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Justin Stitt <justinstitt@google.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Leo Yan <leo.yan@arm.com>
Cc: Manu Bretelle <chantr4@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Quentin Monnet <qmo@kernel.org>
Cc: Steinar H. Gunderson <sesse@google.com>
Link: https://lore.kernel.org/r/20240910140405.568791-1-james.clark@linaro.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Mark the rlim arg in the prlimit64 and setrlimit syscalls as coming from user space

With that it uses the generic BTF based pretty printer:

  root@number:~# perf trace -e prlimit64
       0.000 ( 0.004 ms): :3417020/3417020 prlimit64(resource: NOFILE, old_rlim: 0x7fb8842fe3b0)      = 0
       0.126 ( 0.003 ms): Chroot Helper/3417022 prlimit64(resource: NOFILE, old_rlim: 0x7fb8842fdfd0) = 0
      12.557 ( 0.005 ms): firefox/3417020 prlimit64(resource: STACK, old_rlim: 0x7ffe9ade1b80)        = 0
      26.640 ( 0.006 ms): MainThread/3417020 prlimit64(resource: STACK, old_rlim: 0x7ffe9ade1780)     = 0
      27.553 ( 0.002 ms): Web Content/3417020 prlimit64(resource: AS, old_rlim: 0x7ffe9ade1660)       = 0
      29.405 ( 0.003 ms): Web Content/3417020 prlimit64(resource: NOFILE, old_rlim: 0x7ffe9ade0c80)   = 0
      30.471 ( 0.002 ms): Web Content/3417020 prlimit64(resource: RTTIME, old_rlim: 0x7ffe9ade1370)   = 0
      30.485 ( 0.001 ms): Web Content/3417020 prlimit64(resource: RTTIME, new_rlim: (struct rlimit64){.rlim_cur = (__u64)50000,.rlim_max = (__u64)200000,}) = 0
      31.779 ( 0.001 ms): Web Content/3417020 prlimit64(resource: STACK, old_rlim: 0x7ffe9ade1670)    = 0
  ^Croot@number:~#

Better than before, still needs improvements in the configurability of
the libbpf BTF dumper to get it to the strace output standard.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alan Maguire <alan.maguire@oracle.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/lkml/ZuBQI-f8CGpuhIdH@x1
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Support collecting 'union's with the BPF augmenter

And reuse the BTF based struct pretty printer, with that we can offer
initial support for the 'bpf' syscall's second argument, a 'union
bpf_attr' pointer.

But this is not that satisfactory as the libbpf btf dumper will pretty
print _all_ the union, we need to have a way to say that the first arg
selects the type for the union member to be pretty printed, something
like what pahole does translating the PERF_RECORD_ selector into a name,
and using that name to find a matching struct.

In the case of 'union bpf_attr' it would map PROG_LOAD to one of the
union members, but unfortunately there is no such mapping:

  root@number:~# pahole bpf_attr
  union bpf_attr {
   struct {
   __u32              map_type;           /*     0     4 */
   __u32              key_size;           /*     4     4 */
   __u32              value_size;         /*     8     4 */
   __u32              max_entries;        /*    12     4 */
   __u32              map_flags;          /*    16     4 */
   __u32              inner_map_fd;       /*    20     4 */
   __u32              numa_node;          /*    24     4 */
   char               map_name[16];       /*    28    16 */
   __u32              map_ifindex;        /*    44     4 */
   __u32              btf_fd;             /*    48     4 */
   __u32              btf_key_type_id;    /*    52     4 */
   __u32              btf_value_type_id;  /*    56     4 */
   __u32              btf_vmlinux_value_type_id; /*    60     4 */
   /* --- cacheline 1 boundary (64 bytes) --- */
   __u64              map_extra;          /*    64     8 */
   __s32              value_type_btf_obj_fd; /*    72     4 */
   __s32              map_token_fd;       /*    76     4 */
   };                                             /*     0    80 */
   struct {
   __u32              map_fd;             /*     0     4 */

   /* XXX 4 bytes hole, try to pack */

   __u64              key;                /*     8     8 */
   union {
   __u64      value;              /*    16     8 */
   __u64      next_key;           /*    16     8 */
   };                                     /*    16     8 */
   __u64              flags;              /*    24     8 */
   };                                             /*     0    32 */
   struct {
   __u64              in_batch;           /*     0     8 */
   __u64              out_batch;          /*     8     8 */
   __u64              keys;               /*    16     8 */
   __u64              values;             /*    24     8 */
   __u32              count;              /*    32     4 */
   __u32              map_fd;             /*    36     4 */
   __u64              elem_flags;         /*    40     8 */
   __u64              flags;              /*    48     8 */
   } batch;                                       /*     0    56 */
   struct {
   __u32              prog_type;          /*     0     4 */
   __u32              insn_cnt;           /*     4     4 */
   __u64              insns;              /*     8     8 */
   __u64              license;            /*    16     8 */
   __u32              log_level;          /*    24     4 */
   __u32              log_size;           /*    28     4 */
   __u64              log_buf;            /*    32     8 */
   __u32              kern_version;       /*    40     4 */
   __u32              prog_flags;         /*    44     4 */
   char               prog_name[16];      /*    48    16 */
   /* --- cacheline 1 boundary (64 bytes) --- */
   __u32              prog_ifindex;       /*    64     4 */
   __u32              expected_attach_type; /*    68     4 */
   __u32              prog_btf_fd;        /*    72     4 */
   __u32              func_info_rec_size; /*    76     4 */
   __u64              func_info;          /*    80     8 */
   __u32              func_info_cnt;      /*    88     4 */
   __u32              line_info_rec_size; /*    92     4 */
   __u64              line_info;          /*    96     8 */
   __u32              line_info_cnt;      /*   104     4 */
   __u32              attach_btf_id;      /*   108     4 */
   union {
   __u32      attach_prog_fd;     /*   112     4 */
   __u32      attach_btf_obj_fd;  /*   112     4 */
   };                                     /*   112     4 */
   __u32              core_relo_cnt;      /*   116     4 */
   __u64              fd_array;           /*   120     8 */
   /* --- cacheline 2 boundary (128 bytes) --- */
   __u64              core_relos;         /*   128     8 */
   __u32              core_relo_rec_size; /*   136     4 */
   __u32              log_true_size;      /*   140     4 */
   __s32              prog_token_fd;      /*   144     4 */
   };                                             /*     0   152 */
   struct {
   __u64              pathname;           /*     0     8 */
   __u32              bpf_fd;             /*     8     4 */
   __u32              file_flags;         /*    12     4 */
   __s32              path_fd;            /*    16     4 */
   };                                             /*     0    24 */
   struct {
   union {
   __u32      target_fd;          /*     0     4 */
   __u32      target_ifindex;     /*     0     4 */
   };                                     /*     0     4 */
   __u32              attach_bpf_fd;      /*     4     4 */
   __u32              attach_type;        /*     8     4 */
   __u32              attach_flags;       /*    12     4 */
   __u32              replace_bpf_fd;     /*    16     4 */
   union {
   __u32      relative_fd;        /*    20     4 */
   __u32      relative_id;        /*    20     4 */
   };                                     /*    20     4 */
   __u64              expected_revision;  /*    24     8 */
   };                                             /*     0    32 */
   struct {
   __u32              prog_fd;            /*     0     4 */
   __u32              retval;             /*     4     4 */
   __u32              data_size_in;       /*     8     4 */
   __u32              data_size_out;      /*    12     4 */
   __u64              data_in;            /*    16     8 */
   __u64              data_out;           /*    24     8 */
   __u32              repeat;             /*    32     4 */
   __u32              duration;           /*    36     4 */
   __u32              ctx_size_in;        /*    40     4 */
   __u32              ctx_size_out;       /*    44     4 */
   __u64              ctx_in;             /*    48     8 */
   __u64              ctx_out;            /*    56     8 */
   /* --- cacheline 1 boundary (64 bytes) --- */
   __u32              flags;              /*    64     4 */
   __u32              cpu;                /*    68     4 */
   __u32              batch_size;         /*    72     4 */
   } test;                                        /*     0    80 */
   struct {
   union {
   __u32      start_id;           /*     0     4 */
   __u32      prog_id;            /*     0     4 */
   __u32      map_id;             /*     0     4 */
   __u32      btf_id;             /*     0     4 */
   __u32      link_id;            /*     0     4 */
   };                                     /*     0     4 */
   __u32              next_id;            /*     4     4 */
   __u32              open_flags;         /*     8     4 */
   };                                             /*     0    12 */
   struct {
   __u32              bpf_fd;             /*     0     4 */
   __u32              info_len;           /*     4     4 */
   __u64              info;               /*     8     8 */
   } info;                                        /*     0    16 */
   struct {
   union {
   __u32      target_fd;          /*     0     4 */
   __u32      target_ifindex;     /*     0     4 */
   };                                     /*     0     4 */
   __u32              attach_type;        /*     4     4 */
   __u32              query_flags;        /*     8     4 */
   __u32              attach_flags;       /*    12     4 */
   __u64              prog_ids;           /*    16     8 */
   union {
   __u32      prog_cnt;           /*    24     4 */
   __u32      count;              /*    24     4 */
   };                                     /*    24     4 */

   /* XXX 4 bytes hole, try to pack */

   __u64              prog_attach_flags;  /*    32     8 */
   __u64              link_ids;           /*    40     8 */
   __u64              link_attach_flags;  /*    48     8 */
   __u64              revision;           /*    56     8 */
   } query;                                       /*     0    64 */
   struct {
   __u64              name;               /*     0     8 */
   __u32              prog_fd;            /*     8     4 */

   /* XXX 4 bytes hole, try to pack */

   __u64              cookie;             /*    16     8 */
   } raw_tracepoint;                              /*     0    24 */
   struct {
   __u64              btf;                /*     0     8 */
   __u64              btf_log_buf;        /*     8     8 */
   __u32              btf_size;           /*    16     4 */
   __u32              btf_log_size;       /*    20     4 */
   __u32              btf_log_level;      /*    24     4 */
   __u32              btf_log_true_size;  /*    28     4 */
   __u32              btf_flags;          /*    32     4 */
   __s32              btf_token_fd;       /*    36     4 */
   };                                             /*     0    40 */
   struct {
   __u32              pid;                /*     0     4 */
   __u32              fd;                 /*     4     4 */
   __u32              flags;              /*     8     4 */
   __u32              buf_len;            /*    12     4 */
   __u64              buf;                /*    16     8 */
   __u32              prog_id;            /*    24     4 */
   __u32              fd_type;            /*    28     4 */
   __u64              probe_offset;       /*    32     8 */
   __u64              probe_addr;         /*    40     8 */
   } task_fd_query;                               /*     0    48 */
   struct {
   union {
   __u32      prog_fd;            /*     0     4 */
   __u32      map_fd;             /*     0     4 */
   };                                     /*     0     4 */
   union {
   __u32      target_fd;          /*     4     4 */
   __u32      target_ifindex;     /*     4     4 */
   };                                     /*     4     4 */
   __u32              attach_type;        /*     8     4 */
   __u32              flags;              /*    12     4 */
   union {
   __u32      target_btf_id;      /*    16     4 */
   struct {
   __u64 iter_info;       /*    16     8 */
   __u32 iter_info_len;   /*    24     4 */
   };                             /*    16    16 */
   struct {
   __u64 bpf_cookie;      /*    16     8 */
   } perf_event;                  /*    16     8 */
   struct {
   __u32 flags;           /*    16     4 */
   __u32 cnt;             /*    20     4 */
   __u64 syms;            /*    24     8 */
   __u64 addrs;           /*    32     8 */
   __u64 cookies;         /*    40     8 */
   } kprobe_multi;                /*    16    32 */
   struct {
   __u32 target_btf_id;   /*    16     4 */

   /* XXX 4 bytes hole, try to pack */

   __u64 cookie;          /*    24     8 */
   } tracing;                     /*    16    16 */
   struct {
   __u32 pf;              /*    16     4 */
   __u32 hooknum;         /*    20     4 */
   __s32 priority;        /*    24     4 */
   __u32 flags;           /*    28     4 */
   } netfilter;                   /*    16    16 */
   struct {
   union {
   __u32  relative_fd; /*    16     4 */
   __u32  relative_id; /*    16     4 */
   };                     /*    16     4 */

   /* XXX 4 bytes hole, try to pack */

   __u64 expected_revision; /*    24     8 */
   } tcx;                         /*    16    16 */
   struct {
   __u64 path;            /*    16     8 */
   __u64 offsets;         /*    24     8 */
   __u64 ref_ctr_offsets; /*    32     8 */
   __u64 cookies;         /*    40     8 */
   __u32 cnt;             /*    48     4 */
   __u32 flags;           /*    52     4 */
   __u32 pid;             /*    56     4 */
   } uprobe_multi;                /*    16    48 */
   struct {
   union {
   __u32  relative_fd; /*    16     4 */
   __u32  relative_id; /*    16     4 */
   };                     /*    16     4 */

   /* XXX 4 bytes hole, try to pack */

   __u64 expected_revision; /*    24     8 */
   } netkit;                      /*    16    16 */
   };                                     /*    16    48 */
   } link_create;                                 /*     0    64 */
   struct {
   __u32              link_fd;            /*     0     4 */
   union {
   __u32      new_prog_fd;        /*     4     4 */
   __u32      new_map_fd;         /*     4     4 */
   };                                     /*     4     4 */
   __u32              flags;              /*     8     4 */
   union {
   __u32      old_prog_fd;        /*    12     4 */
   __u32      old_map_fd;         /*    12     4 */
   };                                     /*    12     4 */
   } link_update;                                 /*     0    16 */
   struct {
   __u32              link_fd;            /*     0     4 */
   } link_detach;                                 /*     0     4 */
   struct {
   __u32              type;               /*     0     4 */
   } enable_stats;                                /*     0     4 */
   struct {
   __u32              link_fd;            /*     0     4 */
   __u32              flags;              /*     4     4 */
   } iter_create;                                 /*     0     8 */
   struct {
   __u32              prog_fd;            /*     0     4 */
   __u32              map_fd;             /*     4     4 */
   __u32              flags;              /*     8     4 */
   } prog_bind_map;                               /*     0    12 */
   struct {
   __u32              flags;              /*     0     4 */
   __u32              bpffs_fd;           /*     4     4 */
   } token_create;                                /*     0     8 */
  };

  root@number:~#

So this is one case where BTF gets us only that far, not getting all
the way to automate the pretty printing of unions designed like 'union
bpf_attr', we will need a custom pretty printer for this union, as using
the libbpf union BTF dumper is way too verbose:

  root@number:~# perf trace --max-events 1 -e bpf bpftool map
       0.000 ( 0.054 ms): bpftool/3409073 bpf(cmd: PROG_LOAD, uattr: (union bpf_attr){(struct){.map_type = (__u32)1,.key_size = (__u32)2,.value_size = (__u32)2755142048,.max_entries = (__u32)32764,.map_flags = (__u32)150263906,.inner_map_fd = (__u32)21920,},(struct){.map_fd = (__u32)1,.key = (__u64)140723063628192,(union){.value = (__u64)94145833392226,.next_key = (__u64)94145833392226,},},.batch = (struct){.in_batch = (__u64)8589934593,.out_batch = (__u64)140723063628192,.keys = (__u64)94145833392226,},(struct){.prog_type = (__u32)1,.insn_cnt = (__u32)2,.insns = (__u64)140723063628192,.license = (__u64)94145833392226,},(struct){.pathname = (__u64)8589934593,.bpf_fd = (__u32)2755142048,.file_flags = (__u32)32764,.path_fd = (__s32)150263906,},(struct){(union){.target_fd = (__u32)1,.target_ifindex = (__u32)1,},.attach_bpf_fd = (__u32)2,.attach_type = (__u32)2755142048,.attach_flags = (__u32)32764,.replace_bpf_fd = (__u32)150263906,(union){.relative_fd = (__u32)21920,.relative_id = (__u32)21920,},},.test = (struct){.prog_fd = (__u32)1,.retval = (__u32)2,.data_size_in = (__u32)2755142048,.data_size_out = (__u32)32764,.data_in = (__u64)94145833392226,},(struct){(union){.start_id = (__u32)1,.prog_id = (__u32)1,.map_id = (__u32)1,.btf_id = (__u32)1,.link_id = (__u32)1,},.next_id = (__u32)2,.open_flags = (__u32)2755142048,},.info = (struct){.bpf_fd = (__u32)1,.info_len = (__u32)2,.info = (__u64)140723063628192,},.query = (struct){(union){.target_fd = (__u32)1,.target_ifindex = (__u32)1,},.attach_type = (__u32)2,.query_flags = (__u32)2755142048,.attach_flags = (__u32)32764,.prog_ids = (__u64)94145833392226,},.raw_tracepoint = (struct){.name = (__u64)8589934593,.prog_fd = (__u32)2755142048,.cookie = (__u64)94145833392226,},(struct){.btf = (__u64)8589934593,.btf_log_buf = (__u64)140723063628192,.btf_size = (__u32)150263906,.btf_log_size = (__u32)21920,},.task_fd_query = (struct){.pid = (__u32)1,.fd = (__u32)2,.flags = (__u32)2755142048,.buf_len = (__u32)32764,.buf = (__u64)94145833392226,},.link_create = (struct){(union){.prog_fd = (__u32)1,.map_fd = (__u32)1,},(u) = 3
  root@number:~# 2: prog_array  name hid_jmp_table  flags 0x0
   key 4B  value 4B  max_entries 1024  memlock 8440B
   owner_prog_type tracing  owner jited
  13: hash_of_maps  name cgroup_hash  flags 0x0
   key 8B  value 4B  max_entries 2048  memlock 167584B
   pids systemd(1)
  960: array  name libbpf_global  flags 0x0
   key 4B  value 32B  max_entries 1  memlock 280B
  961: array  name pid_iter.rodata  flags 0x480
   key 4B  value 4B  max_entries 1  memlock 8192B
   btf_id 1846  frozen
   pids bpftool(3409073)
  962: array  name libbpf_det_bind  flags 0x0
   key 4B  value 32B  max_entries 1  memlock 280B

  root@number:~#

For simpler unions this may be better than not seeing any payload, so
keep it there.

Acked-by: Howard Chu <howardchu95@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alan Maguire <alan.maguire@oracle.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/lkml/ZuBLat8cbadILNLA@x1
[ Removed needless parenteses in the if block leading to the trace__btf_scnprintf() call, as per Howard's review comments ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Add --force-btf for debugging

If --force-btf is enabled, prefer btf_dump general pretty printer to
perf trace's customized pretty printers.

Mostly for debug purposes.

Committer testing:

diff before/after shows we need several improvements to be able to
compare the changes, first we need to cut off/disable mutable data such
as pids and timestamps, then what is left are the buffer addresses
passed from userspace, returned from kernel space, maybe we can ask
'perf trace' to go on making those reproducible.

That would entail a Pointer Address Translation (PAT) like for
networking, that would, for simple, reproducible if not for these
details, workloads, that we would then use in our regression tests.

Enough digression, this is one such diff:

   openat(dfd: CWD, filename: "/usr/share/locale/locale.alias", flags: RDONLY|CLOEXEC) = 3
  -fstat(fd: 3, statbuf: 0x7fff01f212a0)                                 = 0
  -read(fd: 3, buf: 0x5596bab2d630, count: 4096)                         = 2998
  -read(fd: 3, buf: 0x5596bab2d630, count: 4096)                         = 0
  +fstat(fd: 3, statbuf: 0x7ffc163cf0e0)                                 = 0
  +read(fd: 3, buf: 0x55b4e0631630, count: 4096)                         = 2998
  +read(fd: 3, buf: 0x55b4e0631630, count: 4096)                         = 0
   close(fd: 3)                                                          = 0
   openat(dfd: CWD, filename: "/usr/share/locale/en_US.UTF-8/LC_MESSAGES/coreutils.mo") = -1 ENOENT (No such file or directory)
   openat(dfd: CWD, filename: "/usr/share/locale/en_US.utf8/LC_MESSAGES/coreutils.mo") = -1 ENOENT (No such file or directory)
  @@ -45,7 +45,7 @@
   openat(dfd: CWD, filename: "/usr/share/locale/en.UTF-8/LC_MESSAGES/coreutils.mo") = -1 ENOENT (No such file or directory)
   openat(dfd: CWD, filename: "/usr/share/locale/en.utf8/LC_MESSAGES/coreutils.mo") = -1 ENOENT (No such file or directory)
   openat(dfd: CWD, filename: "/usr/share/locale/en/LC_MESSAGES/coreutils.mo") = -1 ENOENT (No such file or directory)
  -{ .tv_sec: 1, .tv_nsec: 0 }, rmtp: 0x7fff01f21990) = 0
  +(struct __kernel_timespec){.tv_sec = (__kernel_time64_t)1,}, rmtp: 0x7ffc163cf7d0) =

The problem more close to our hands is to make the libbpf BTF pretty
printer to have a mode that closely resembles what we're trying to
resemble: strace output.

Being able to run something with 'perf trace' and with 'strace' and get
the exact same output should be of interest of anybody wanting to have
strace and 'perf trace' regression tested against each other.

That last part is 'perf trace' shot at being something so useful as
strace... ;-)

Suggested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Howard Chu <howardchu95@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240824163322.60796-8-howardchu95@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Collect augmented data using BPF

Include trace_augment.h for TRACE_AUG_MAX_BUF, so that BPF reads
TRACE_AUG_MAX_BUF bytes of buffer maximum.

Determine what type of argument and how many bytes to read from user space, us ing the
value in the beauty_map. This is the relation of parameter type and its corres ponding
value in the beauty map, and how many bytes we read eventually:

string: 1                          -> size of string (till null)
struct: size of struct             -> size of struct
buffer: -1 * (index of paired len) -> value of paired len (maximum: TRACE_AUG_ MAX_BUF)

After reading from user space, we output the augmented data using
bpf_perf_event_output().

If the struct augmenter, augment_sys_enter() failed, we fall back to
using bpf_tail_call().

I have to make the payload 6 times the size of augmented_arg, to pass the
BPF verifier.

Signed-off-by: Howard Chu <howardchu95@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240815013626.935097-10-howardchu95@gmail.com
Link: https://lore.kernel.org/r/20240824163322.60796-7-howardchu95@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Pretty print buffer data

Define TRACE_AUG_MAX_BUF in trace_augment.h data, which is the maximum
buffer size we can augment. BPF will include this header too.

Print buffer in a way that's different than just printing a string, we
print all the control characters in \digits (such as \0 for null, and
\10 for newline, LF).

For character that has a bigger value than 127, we print the digits
instead of the character itself as well.

Committer notes:

Simplified the buffer scnprintf to avoid using multiple buffers as
discussed in the patch review thread.

We can't really all 'buf' args to SCA_BUF as we're collecting so far
just on the sys_enter path, so we would be printing the previous 'read'
arg buffer contents, not what the kernel puts there.

So instead of:
   static int syscall_fmt__cmp(const void *name, const void *fmtp)
  @@ -1987,8 +1989,6 @@ syscall_arg_fmt__init_array(struct syscall_arg_fmt *arg, struct tep_format_field
  -               else if (strstr(field->type, "char *") && strstr(field->name, "buf"))
  -                       arg->scnprintf = SCA_BUF;

Do:

static const struct syscall_fmt syscall_fmts[] = {
  +       { .name     = "write",      .errpid = true,
  +         .arg = { [1] = { .scnprintf = SCA_BUF /* buf */, from_user = true, }, }, },

Signed-off-by: Howard Chu <howardchu95@gmail.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240815013626.935097-8-howardchu95@gmail.com
Link: https://lore.kernel.org/r/20240824163322.60796-6-howardchu95@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Pretty print struct data

Change the arg->augmented.args to arg->augmented.args->value to skip the
header for customized pretty printers, since we collect data in BPF
using the general augment_sys_enter(), which always adds the header.

Use btf_dump API to pretty print augmented struct pointer.

Prefer existed pretty-printer than btf general pretty-printer.

set compact = true and skip_names = true, so that no newline character
and argument name are printed.

Committer notes:

Simplified the btf_dump_snprintf callback to avoid using multiple
buffers, as discussed in the thread accessible via the Link tag below.

Also made it do:

  dump_data_opts.skip_names = !arg->trace->show_arg_names;

I.e. show the type and struct field names according to that tunable, we
probably need another tunable just for this, but for now if the user
wants to see syscall names in addition to its value, it makes sense to
see the struct field names according to that tunable.

Committer testing:

The following have explicitely set beautifiers (SCA_FILENAME,
SCA_SOCKADDR and SCA_PERF_ATTR), SCA_FILENAME is here just because we
have been wiring up the "renameat2" ("renameat" until recently), so it
doesn't use the introduced generic fallback (btf_struct_scnprintf(), see
the definition of SCA_PERF_ATTR, SCA_SOCKADDR to see the more feature
rich beautifiers, that are not using BTF):

  root@number:~# rm -f 987654 ; touch 123456 ; perf trace -e rename* mv 123456 987654
       0.000 ( 0.039 ms): mv/258478 renameat2(olddfd: CWD, oldname: "123456", newdfd: CWD, newname: "987654", flags: NOREPLACE) = 0
  root@number:~# perf trace -e connect,sendto ping -c 1 www.google.com
       0.000 ( 0.014 ms): ping/258481 connect(fd: 5, uservaddr: { .family: LOCAL, path: /run/systemd/resolve/io.systemd.Resolve }, addrlen: 42) = 0
       0.040 ( 0.003 ms): ping/258481 sendto(fd: 5, buff: 0x55bc317a6980, len: 97, flags: DONTWAIT|NOSIGNAL) = 97
      18.742 ( 0.020 ms): ping/258481 sendto(fd: 5, buff: 0x7ffc04768df0, len: 20, addr: { .family: NETLINK }, addr_len: 0xc) = 20
  PING www.google.com (142.251.129.68) 56(84) bytes of data.
      18.783 ( 0.012 ms): ping/258481 connect(fd: 5, uservaddr: { .family: INET6, port: 0, addr: 2800:3f0:4004:810::2004 }, addrlen: 28) = 0
      18.797 ( 0.001 ms): ping/258481 connect(fd: 5, uservaddr: { .family: UNSPEC }, addrlen: 16)           = 0
      18.800 ( 0.004 ms): ping/258481 connect(fd: 5, uservaddr: { .family: INET, port: 0, addr: 142.251.129.68 }, addrlen: 16) = 0
      18.815 ( 0.002 ms): ping/258481 connect(fd: 5, uservaddr: { .family: INET, port: 1025, addr: 142.251.129.68 }, addrlen: 16) = 0
      18.862 ( 0.023 ms): ping/258481 sendto(fd: 3, buff: 0x55bc317a0ac0, len: 64, addr: { .family: INET, port: 0, addr: 142.251.129.68 }, addr_len: 0x10) = 64
      63.330 ( 0.038 ms): ping/258481 connect(fd: 5, uservaddr: { .family: LOCAL, path: /run/systemd/resolve/io.systemd.Resolve }, addrlen: 42) = 0
      63.435 ( 0.010 ms): ping/258481 sendto(fd: 5, buff: 0x55bc317a8340, len: 110, flags: DONTWAIT|NOSIGNAL) = 110
  64 bytes from rio07s07-in-f4.1e100.net (142.251.129.68): icmp_seq=1 ttl=49 time=44.2 ms

  --- www.google.com ping statistics ---
  1 packets transmitted, 1 received, 0% packet loss, time 0ms
  rtt min/avg/max/mdev = 44.158/44.158/44.158/0.000 ms
  root@number:~# perf trace -e perf_event_open perf stat -e instructions,cache-misses,syscalls:sys_enter*sleep* sleep 1.23456789
       0.000 ( 0.010 ms): :258487/258487 perf_event_open(attr_uptr: { type: 0 (PERF_TYPE_HARDWARE), config: 0xa00000000, disabled: 1, { bp_len, config2 }: 0x900000000, branch_sample_type: USER|COUNTERS, sample_regs_user: 0x3f1b7ffffffff, sample_stack_user: 258487, clockid: -599052088, sample_regs_intr: 0x60a000003eb, sample_max_stack: 14, sig_data: 120259084288 }, cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 3
       0.016 ( 0.002 ms): :258487/258487 perf_event_open(attr_uptr: { type: 0 (PERF_TYPE_HARDWARE), config: 0x400000000, disabled: 1, { bp_len, config2 }: 0x900000000, branch_sample_type: USER|COUNTERS, sample_regs_user: 0x3f1b7ffffffff, sample_stack_user: 258487, clockid: -599044082, sample_regs_intr: 0x60a000003eb, sample_max_stack: 14, sig_data: 120259084288 }, cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
       1.838 ( 0.006 ms): perf/258487 perf_event_open(attr_uptr: { type: 0 (PERF_TYPE_HARDWARE), size: 136, config: 0xa00000001, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258488 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 5
       1.846 ( 0.002 ms): perf/258487 perf_event_open(attr_uptr: { type: 0 (PERF_TYPE_HARDWARE), size: 136, config: 0x400000001, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258488 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 6
       1.849 ( 0.002 ms): perf/258487 perf_event_open(attr_uptr: { type: 0 (PERF_TYPE_HARDWARE), size: 136, config: 0xa00000003, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258488 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 7
       1.851 ( 0.002 ms): perf/258487 perf_event_open(attr_uptr: { type: 0 (PERF_TYPE_HARDWARE), size: 136, config: 0x400000003, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258488 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 9
       1.853 ( 0.600 ms): perf/258487 perf_event_open(attr_uptr: { type: 2 (tracepoint), size: 136, config: 0x190 (syscalls:sys_enter_nanosleep), sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258488 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 10
       2.456 ( 0.016 ms): perf/258487 perf_event_open(attr_uptr: { type: 2 (tracepoint), size: 136, config: 0x196 (syscalls:sys_enter_clock_nanosleep), sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258488 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 11

   Performance counter stats for 'sleep 1.23456789':

           1,402,839      cpu_atom/instructions/
       <not counted>      cpu_core/instructions/                                                  (0.00%)
              11,066      cpu_atom/cache-misses/
       <not counted>      cpu_core/cache-misses/                                                  (0.00%)
                   0      syscalls:sys_enter_nanosleep
                   1      syscalls:sys_enter_clock_nanosleep

         1.236246714 seconds time elapsed

         0.000000000 seconds user
         0.001308000 seconds sys

  root@number:~#

Now if we use it even for the ones we have a specific beautifier in
tools/perf/trace/beauty, i.e. use btf_struct_scnprintf() for all
structs, by adding the following patch:

  @@ -2316,7 +2316,7 @@ static size_t syscall__scnprintf_args(struct syscall *sc, char *bf, size_t size,

    default_scnprintf = sc->arg_fmt[arg.idx].scnprintf;

  - if (default_scnprintf == NULL || default_scnprintf == SCA_PTR) {
  + if (1 || (default_scnprintf == NULL || default_scnprintf == SCA_PTR)) {
    btf_printed = trace__btf_scnprintf(trace, &arg, bf + printed,
       size - printed, val, field->type);
    if (btf_printed) {

We get:

  root@number:~# perf trace -e connect,sendto ping -c 1 www.google.com
  PING www.google.com (142.251.129.68) 56(84) bytes of data.
       0.000 ( 0.015 ms): ping/283259 connect(fd: 5, uservaddr: (struct sockaddr){.sa_family = (sa_family_t)1,(union){.sa_data_min = (char[14])['/','r','u','n','/','s','y','s','t','e','m','d','/','r',],},}, addrlen: 42) = 0
       0.046 ( 0.004 ms): ping/283259 sendto(fd: 5, buff: 0x559b008ae980, len: 97, flags: DONTWAIT|NOSIGNAL) = 97
       0.353 ( 0.012 ms): ping/283259 sendto(fd: 5, buff: 0x7ffc01294960, len: 20, addr: (struct sockaddr){.sa_family = (sa_family_t)16,}, addr_len: 0xc) = 20
       0.377 ( 0.006 ms): ping/283259 connect(fd: 5, uservaddr: (struct sockaddr){.sa_family = (sa_family_t)2,}, addrlen: 16) = 0
       0.388 ( 0.010 ms): ping/283259 connect(fd: 5, uservaddr: (struct sockaddr){.sa_family = (sa_family_t)10,}, addrlen: 28) = 0
       0.402 ( 0.001 ms): ping/283259 connect(fd: 5, uservaddr: (struct sockaddr){.sa_family = (sa_family_t)2,(union){.sa_data_min = (char[14])[4,1,142,251,129,'D',],},}, addrlen: 16) = 0
       0.425 ( 0.045 ms): ping/283259 sendto(fd: 3, buff: 0x559b008a8ac0, len: 64, addr: (struct sockaddr){.sa_family = (sa_family_t)2,}, addr_len: 0x10) = 64
  64 bytes from rio07s07-in-f4.1e100.net (142.251.129.68): icmp_seq=1 ttl=49 time=44.1 ms

  --- www.google.com ping statistics ---
  1 packets transmitted, 1 received, 0% packet loss, time 0ms
  rtt min/avg/max/mdev = 44.113/44.113/44.113/0.000 ms
      44.849 ( 0.038 ms): ping/283259 connect(fd: 5, uservaddr: (struct sockaddr){.sa_family = (sa_family_t)1,(union){.sa_data_min = (char[14])['/','r','u','n','/','s','y','s','t','e','m','d','/','r',],},}, addrlen: 42) = 0
      44.927 ( 0.006 ms): ping/283259 sendto(fd: 5, buff: 0x559b008b03d0, len: 110, flags: DONTWAIT|NOSIGNAL) = 110
  root@number:~#

Which looks sane, i.e.:

  18.800 ( 0.004 ms): ping/258481 connect(fd: 5, uservaddr: { .family: INET, port: 0, addr: 142.251.129.68 }, addrlen: 16) = 0

Becomes:

   0.402 ( 0.001 ms): ping/283259 connect(fd: 5, uservaddr: (struct sockaddr){.sa_family = (sa_family_t)2,(union){.sa_data_min = (char[14])[4,1,142,251,129,'D',],},}, addrlen: 16) = 0

And.

  #define AF_UNIX         1       /* Unix domain sockets          */
  #define AF_LOCAL        1       /* POSIX name for AF_UNIX       */
  #define AF_INET         2       /* Internet IP Protocol         */
  <SNIP>
  #define AF_INET6        10      /* IP version 6                 */

And 'D' == 68, so the preexisting sockaddr BPF collector is working with
the new generic BTF pretty printer (btf_struct_scnprintf()), its just
that it doesn't know about 'struct sockaddr' besides what is in BTF,
i.e. its an array of bytes, not an IPv4 address that needs extra
massaging.

Ditto for the 'struct perf_event_attr' case:

       1.851 ( 0.002 ms): perf/258487 perf_event_open(attr_uptr: { type: 0 (PERF_TYPE_HARDWARE), size: 136, config: 0x400000003, sample_type: IDENTIFIER, read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING, disabled: 1, inherit: 1, enable_on_exec: 1, exclude_guest: 1 }, pid: 258488 (perf), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 9

Becomes:

       2.081 ( 0.002 ms): :283304/283304 perf_event_open(attr_uptr: (struct perf_event_attr){.size = (__u32)136,.config = (__u64)17179869187,.sample_type = (__u64)65536,.read_format = (__u64)3,.disabled = (__u64)0x1,.inherit = (__u64)0x1,.enable_on_exec = (__u64)0x1,.exclude_guest = (__u64)0x1,}, pid: 283305 (sleep), cpu: -1, group_fd: -1, flags: FD_CLOEXEC) = 9

hex(17179869187) = 0x400000003, etc.

read_format: TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING is

enum perf_event_read_format {
        PERF_FORMAT_TOTAL_TIME_ENABLED          = 1U << 0,
        PERF_FORMAT_TOTAL_TIME_RUNNING          = 1U << 1,

and so on.

We need to work with the libbpf btf dump api to get one output that
matches the 'perf trace'/strace expectations/format, but having this in
this current form is already an improvement to 'perf trace', so lets
improve from what we have.

Signed-off-by: Howard Chu <howardchu95@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240815013626.935097-7-howardchu95@gmail.com
Link: https://lore.kernel.org/r/20240824163322.60796-5-howardchu95@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Add trace__bpf_sys_enter_beauty_map() to prepare for fetching data in BPF

Set up beauty_map, load it to BPF, in such format: if argument No.3 is a
struct of size 32 bytes (of syscall number 114) beauty_map[114][2] = 32;

if argument No.3 is a string (of syscall number 114) beauty_map[114][2] =
1;

if argument No.3 is a buffer, its size is indicated by argument No.4 (of
syscall number 114) beauty_map[114][2] = -4; /* -1 ~ -6, we'll read this
buffer size in BPF  */

Committer notes:

Moved syscall_arg_fmt__cache_btf_struct() from a ifdef
HAVE_LIBBPF_SUPPORT to closer to where it is used, that is ifdef'ed on
HAVE_BPF_SKEL and thus breaks the build when building with
BUILD_BPF_SKEL=0, as detected using 'make -C tools/perf build-test'.

Also add 'struct beauty_map_enter' to tools/perf/util/bpf_skel/augmented_raw_syscalls.bpf.c
as we're using it in this patch, otherwise we get this while trying to
build at this point in the original patch series:

  builtin-trace.c: In function ‘trace__init_syscalls_bpf_prog_array_maps’:
  builtin-trace.c:3725:58: error: ‘struct <anonymous>’ has no member named ‘beauty_map_enter’
   3725 |         int beauty_map_fd = bpf_map__fd(trace->skel->maps.beauty_map_enter);
        |

We also have to take into account syscall_arg_fmt.from_user when telling
the kernel what to copy in the sys_enter generic collector, we don't
want to collect bogus data in buffers that will only be available to us
at sys_exit time, i.e. after the kernel has filled it, so leave this for
when we have such a sys_exit based collector.

Committer testing:

Not wired up yet, so all continues to work, using the existing BPF
collector and userspace beautifiers that are augmentation aware:

  root@number:~# rm -f 987654 ; touch 123456 ; perf trace -e rename* mv 123456 987654
       0.000 ( 0.031 ms): mv/20888 renameat2(olddfd: CWD, oldname: "123456", newdfd: CWD, newname: "987654", flags: NOREPLACE) = 0
  root@number:~# perf trace -e connect,sendto ping -c 1 www.google.com
       0.000 ( 0.014 ms): ping/20892 connect(fd: 5, uservaddr: { .family: LOCAL, path: /run/systemd/resolve/io.systemd.Resolve }, addrlen: 42) = 0
       0.040 ( 0.003 ms): ping/20892 sendto(fd: 5, buff: 0x560b4ff17980, len: 97, flags: DONTWAIT|NOSIGNAL) = 97
       0.480 ( 0.017 ms): ping/20892 sendto(fd: 5, buff: 0x7ffd82d07150, len: 20, addr: { .family: NETLINK }, addr_len: 0xc) = 20
       0.526 ( 0.014 ms): ping/20892 connect(fd: 5, uservaddr: { .family: INET6, port: 0, addr: 2800:3f0:4004:810::2004 }, addrlen: 28) = 0
       0.542 ( 0.002 ms): ping/20892 connect(fd: 5, uservaddr: { .family: UNSPEC }, addrlen: 16)           = 0
       0.544 ( 0.004 ms): ping/20892 connect(fd: 5, uservaddr: { .family: INET, port: 0, addr: 142.251.135.100 }, addrlen: 16) = 0
       0.559 ( 0.002 ms): ping/20892 connect(fd: 5, uservaddr: { .family: INET, port: 1025, addr: 142.251.135.100 }, addrlen: 16PING www.google.com (142.251.135.100) 56(84) bytes of data.
  ) = 0
       0.589 ( 0.058 ms): ping/20892 sendto(fd: 3, buff: 0x560b4ff11ac0, len: 64, addr: { .family: INET, port: 0, addr: 142.251.135.100 }, addr_len: 0x10) = 64
      45.250 ( 0.029 ms): ping/20892 connect(fd: 5, uservaddr: { .family: LOCAL, path: /run/systemd/resolve/io.systemd.Resolve }, addrlen: 42) = 0
      45.344 ( 0.012 ms): ping/20892 sendto(fd: 5, buff: 0x560b4ff19340, len: 111, flags: DONTWAIT|NOSIGNAL) = 111
  64 bytes from rio09s08-in-f4.1e100.net (142.251.135.100): icmp_seq=1 ttl=49 time=44.4 ms

  --- www.google.com ping statistics ---
  1 packets transmitted, 1 received, 0% packet loss, time 0ms
  rtt min/avg/max/mdev = 44.361/44.361/44.361/0.000 ms
  root@number:~#

Signed-off-by: Howard Chu <howardchu95@gmail.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240815013626.935097-4-howardchu95@gmail.com
Link: https://lore.kernel.org/r/20240824163322.60796-3-howardchu95@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Mark bpf's attr as from_user

This one has no specific pretty printer right now, so will be handled by
the generic BTF based one later in this patch series.

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

perf trace: Introduce SCA_TIMESPEC_FROM_USER() to set .from_user = true

Paving the way for the generic BPF BTF based syscall arg augmenter.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Introduce SCA_SOCKADDR_FROM_USER() to set .from_user = true

Paving the way for the generic BPF BTF based syscall arg augmenter.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

perf trace: Introduce SCA_PERF_ATTR_FROM_USER() to set .from_user = true

Paving the way for the generic BPF BTF based syscall arg augmenter.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Howard Chu <howardchu95@gmail.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>