linux-stable.git/tools/perf/util, branch v6.5.2

x86/retpoline,kprobes: Skip optprobe check for indirect jumps with retpolines and IBT

2023-08-14T09:46:51+00:00

The kprobes optimization check can_optimize() calls
insn_is_indirect_jump() to detect indirect jump instructions in
a target function. If any is found, creating an optprobe is disallowed
in the function because the jump could be from a jump table and could
potentially land in the middle of the target optprobe.

With retpolines, insn_is_indirect_jump() additionally looks for calls to
indirect thunks which the compiler potentially used to replace original
jumps. This extra check is however unnecessary because jump tables are
disabled when the kernel is built with retpolines. The same is currently
the case with IBT.

Based on this observation, remove the logic to look for calls to
indirect thunks and skip the check for indirect jumps altogether if the
kernel is built with retpolines or IBT. Remove subsequently the symbols
__indirect_thunk_start and __indirect_thunk_end which are no longer
needed.

Dropping this logic indirectly fixes a problem where the range
[__indirect_thunk_start, __indirect_thunk_end] wrongly included also the
return thunk. It caused that machines which used the return thunk as
a mitigation and didn't have it patched by any alternative ended up not
being able to use optprobes in any regular function.

Fixes: 0b53c374b9ef ("x86/retpoline: Use -mfunction-return")
Suggested-by: Peter Zijlstra (Intel) 
Suggested-by: Masami Hiramatsu (Google) 
Signed-off-by: Petr Pavlu 
Signed-off-by: Peter Zijlstra (Intel) 
Signed-off-by: Borislav Petkov (AMD) 
Acked-by: Masami Hiramatsu (Google) 
Link: https://lore.kernel.org/r/20230711091952.27944-3-petr.pavlu@suse.com

perf stat: Don't display zero tool counts

2023-08-08T17:33:57+00:00

Andi reported (see link below) a regression when printing the
'duration_time' tool event, where it gets printed as "not counted" for
most of the CPUs, fix it by skipping zero counts for tool events.

Reported-by: Andi Kleen 
Signed-off-by: Ian Rogers 
Tested-by: Andi Kleen 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Athira Rajeev 
Cc: Claire Jensen 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Link: https://lore.kernel.org/all/ZMlrzcVrVi1lTDmn@tassilo/
Signed-off-by: Arnaldo Carvalho de Melo

Revert "perf report: Append inlines to non-DWARF callchains"

2023-08-08T13:52:17+00:00

This reverts commit 46d21ec067490ab9cdcc89b9de5aae28786a8b8e.

The tests were made with a specific workload, further tests on a
recently updated fedora 38 system with a system wide perf.data file
shows 'perf report' taking excessive time resolving inlines in vmlinux,
so lets revert this until a full investigation and improvement on the
addr2line support code is made.

Reported-by: Jesper Dangaard Brouer 
Acked-by: Artem Savkov 
Tested-by: Jesper Dangaard Brouer 
Cc: Andrii Nakryiko 
Cc: Namhyung Kim 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Mark Rutland 
Cc: Masami Hiramatsu 
Cc: Milian Wolff 
Cc: Peter Zijlstra 
Link: https://lore.kernel.org/r/ZMl8VyhdwhClTM5g@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo

perf test parse-events: Test complex name has required event format

2023-08-01T03:32:00+00:00

test__checkevent_complex_name will use an "event" format which if not
present, such as with a placeholder PMU, will cause test failures. Skip
the test in this case to avoid failures in restricted environments.

Add perf_pmu__has_format utility as a general PMU utility.

Fixes: 628eaa4e877af823 ("perf pmus: Add placeholder core PMU")
Signed-off-by: Ian Rogers 
Tested-by: Thomas Richter 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: James Clark 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Ravi Bangoria 
Cc: Rob Herring 
Link: https://lore.kernel.org/r/20230706183705.601412-2-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf pmus: Create placholder regardless of scanning core_only

2023-08-01T03:30:06+00:00

If scanning all PMUs the placeholder is still necessary if no core PMU
is found. This situation occurs in perf test's parse-events test,
when uncore events appear before core.

Fixes: 628eaa4e877af823 ("perf pmus: Add placeholder core PMU")
Signed-off-by: Ian Rogers 
Tested-by: Thomas Richter 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: James Clark 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Ravi Bangoria 
Cc: Rob Herring 
Link: https://lore.kernel.org/r/20230706183705.601412-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf parse-events: Only move force grouped evsels when sorting

2023-07-27T13:32:12+00:00

Prior to this change, events without a group would be sorted as if they
were from the location of the first event without a group. For example
instructions and cycles are without a group:

  instructions,{imc_free_running/data_read/,imc_free_running/data_write/},cycles

parse events would create an eventual evlist like:

  instructions,cycles,{uncore_imc_free_running_0/data_read/,uncore_imc_free_running_1/data_read/,uncore_imc_free_running_0/data_write/,uncore_imc_free_running_1/data_write/}

This is done so that perf metric events, that must always be in a
group, will be adjacent and so can be forced into a group.

This change modifies the sorting so that only force grouped events,
like perf metrics, are sorted and all other events keep their position
with respect to groups in the evlist. The location of the force
grouped event is chosen to match the first force grouped event.

For architectures without force grouped events, ie anything not Intel
Icelake or newer, this should mean sorting and fixing doesn't modify
the event positions except when fixing the grouping for PMUs of things
like uncore events.

Fixes: 347c2f0a0988c59c ("perf parse-events: Sort and group parsed events")
Reported-by: Andi Kleen 
Signed-off-by: Ian Rogers 
Tested-by: Andi Kleen 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Xing Zhengjun 
Link: https://lore.kernel.org/r/20230719001836.198363-4-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf parse-events: When fixing group leaders always set the leader

2023-07-27T13:31:32+00:00

The evsel grouping fix iterates over evsels tracking the leader group
and the current position's group, updating the current position's leader
if an evsel is being forced into a group or groups changed. However,
groups changing isn't a sufficient condition as sorting may have
reordered events and the leader may no longer come first. For this
reason update all leaders whenever they disagree.

This change breaks certain Icelake+ metrics due to bugs in the
kernel. For example, tma_l3_bound with threshold enabled tries to
program the events:

  {topdown-retiring,slots,CYCLE_ACTIVITY.STALLS_L2_MISS,topdown-fe-bound,EXE_ACTIVITY.BOUND_ON_STORES,EXE_ACTIVITY.1_PORTS_UTIL,topdown-be-bound,cpu/INT_MISC.RECOVERY_CYCLES,cmask=1,edge/,CYCLE_ACTIVITY.STALLS_L3_MISS,CPU_CLK_UNHALTED.THREAD,CYCLE_ACTIVITY.STALLS_MEM_ANY,EXE_ACTIVITY.2_PORTS_UTIL,CYCLE_ACTIVITY.STALLS_TOTAL,topdown-bad-spec}:W

fixing the perf metric event order gives:

  {slots,topdown-retiring,topdown-fe-bound,topdown-be-bound,topdown-bad-spec,CYCLE_ACTIVITY.STALLS_L2_MISS,EXE_ACTIVITY.BOUND_ON_STORES,EXE_ACTIVITY.1_PORTS_UTIL,cpu/INT_MISC.RECOVERY_CYCLES,cmask=1,edge/,CYCLE_ACTIVITY.STALLS_L3_MISS,CPU_CLK_UNHALTED.THREAD,CYCLE_ACTIVITY.STALLS_MEM_ANY,EXE_ACTIVITY.2_PORTS_UTIL,CYCLE_ACTIVITY.STALLS_TOTAL}:W

Both of these return "" for all events, whilst they work
with the group removed respecting that the perf metric events must still
be grouped. A vendor events update will need to add METRIC_NO_GROUP to
these metrics to workaround the kernel PMU driver issue.

Fixes: a90cc5a9eeab45ea ("perf evsel: Don't let evsel__group_pmu_name() traverse unsorted group")
Signed-off-by: Ian Rogers 
Tested-by: Andi Kleen 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Xing Zhengjun 
Link: https://lore.kernel.org/r/20230719001836.198363-3-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf parse-events: Extra care around force grouped events

2023-07-27T13:27:12+00:00

Perf metric (topdown) events on Intel Icelake+ machines require a
group, however, they may be next to events that don't require a group.
Consider:

  cycles,slots,topdown-fe-bound

The cycles event needn't be grouped but slots and topdown-fe-bound need
grouping.

Prior to this change, as slots and topdown-fe-bound need a group forcing
and all events share the same PMU, slots and topdown-fe-bound would be
forced into a group with cycles.

This is a bug on two fronts, cycles wasn't supposed to be grouped and
cycles can't be a group leader with a perf metric event.

This change adds recognition that cycles isn't force grouped and so it
shouldn't be force grouped with slots and topdown-fe-bound.

Fixes: a90cc5a9eeab45ea ("perf evsel: Don't let evsel__group_pmu_name() traverse unsorted group")
Signed-off-by: Ian Rogers 
Tested-by: Andi Kleen 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Xing Zhengjun 
Link: https://lore.kernel.org/r/20230719001836.198363-2-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf parse-events: Avoid SEGV if PMU lookup fails for legacy cache terms

2023-07-12T19:03:24+00:00

libfuzzer found the following command could SEGV:

  $ perf stat -e cpu/L2,L2/ true

This is because the L2 term rewrites the perf_event_attr type to
PERF_TYPE_HW_CACHE which then fails the PMU lookup for the second
legacy cache term.

The new failure is consistent with repeated hardware terms:

  $ perf stat -e cpu/L2,L2/ true
  event syntax error: 'cpu/L2,L2/'
                              \___ Failed to find PMU for type 3

  Initial error:
  event syntax error: 'cpu/L2,L2/'
                              \___ Failed to find PMU for type 3
  Run 'perf list' for a list of valid events

   Usage: perf stat [] []

      -e, --event    event selector. use 'perf list' to list available events
  $ perf stat -e cpu/cycles,cycles/ true
  event syntax error: 'cpu/cycles,cycles/'
                                  \___ Failed to find PMU for type 0

  Initial error:
  event syntax error: 'cpu/cycles,cycles/'
                                  \___ Failed to find PMU for type 0
  Run 'perf list' for a list of valid events

   Usage: perf stat [] []

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

Committer testing:

Before:

  $ perf stat -e cpu/L2,L2/ true
  Segmentation fault (core dumped)
  $

After:

  $ perf stat -e cpu/L2,L2/ true
  event syntax error: 'cpu/L2,L2/'
                              \___ Failed to find PMU for type 3

  Initial error:
  event syntax error: 'cpu/L2,L2/'
                              \___ Failed to find PMU for type 3
  Run 'perf list' for a list of valid events

   Usage: perf stat [] []

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

Fixes: 6fd1e5191591f9d5 ("perf parse-events: Support PMUs for legacy cache events")
Signed-off-by: Ian Rogers 
Tested-by: Arnaldo Carvalho de Melo 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Kan Liang 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: Xing Zhengjun 
Link: https://lore.kernel.org/r/20230712065250.1450306-1-irogers@google.com
Signed-off-by: Arnaldo Carvalho de Melo

perf probe: Read DWARF files from the correct CU

2023-07-11T14:41:07+00:00

After switching from dwarf_decl_file() to die_get_decl_file(), it is not
possible to add probes for certain functions:

  $ perf probe -x /usr/lib/systemd/systemd-logind match_unit_removed
  A function DIE doesn't have decl_line. Maybe broken DWARF?
  A function DIE doesn't have decl_line. Maybe broken DWARF?
  Probe point 'match_unit_removed' not found.
     Error: Failed to add events.

The problem is that die_get_decl_file() uses the wrong CU to search for
the file. elfutils commit e1db5cdc9f has some good explanation for this:

    dwarf_decl_file uses dwarf_attr_integrate to get the DW_AT_decl_file
    attribute. This means the attribute might come from a different DIE
    in a different CU. If so, we need to use the CU associated with the
    attribute, not the original DIE, to resolve the file name.

This patch uses the same source of information as elfutils: use attribute
DW_AT_decl_file and use this CU to search for the file.

Fixes: dc9a5d2ccd5c823c ("perf probe: Fix to get declared file name from clang DWARF5")
Signed-off-by: Georg Müller 
Acked-by: Masami Hiramatsu (Google) 
Cc: Adrian Hunter 
Cc: Alexander Shishkin 
Cc: Ian Rogers 
Cc: Ingo Molnar 
Cc: Jiri Olsa 
Cc: Mark Rutland 
Cc: Namhyung Kim 
Cc: Peter Zijlstra 
Cc: regressions@lists.linux.dev
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20230628084551.1860532-6-georgmueller@gmx.net
Signed-off-by: Arnaldo Carvalho de Melo