- 14 Mar, 2023 23 commits
-
-
Ian Rogers authored
Update from 1.00 to 1.01, some event description updates. Signed-off-by:
Ian Rogers <irogers@google.com> Reviewed-by:
Kan Liang <kan.liang@linux.intel.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Edward Baker <edward.baker@intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Xing Zhengjun <zhengjun.xing@linux.intel.com> Link: https://lore.kernel.org/r/20230314053312.3237390-2-irogers@google.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Roman Lozko authored
Integers are not converted to floats during division in Python 2 which results in incorrect IPC values. Fix by switching to new division behavior. Fixes: a483e64c ("perf scripting python: intel-pt-events.py: Add --insn-trace and --src-trace") Signed-off-by:
Roman Lozko <lozko.roma@gmail.com> Acked-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: Adrian Hunter <adrian.hunter@intel.com> Link: https://lore.kernel.org/r/20230310150445.2925841-1-lozko.roma@gmail.comSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Now that BPF skel based tools will be built by default if the toolchain pieces that are needed are available, building directly on the source tree will produce a vmlinux.h from the BTF info that needs to get ignored. Signed-off-by: -
Ian Rogers authored
A table was renamed and needed to be renamed in the empty case. Fixes: 62774db2 ("perf jevents: Generate metrics and events as separate tables") Reviewed-by:
John Garry <john.g.garry@oracle.com> Signed-off-by:
-
Namhyung Kim authored
Show lock type names after the symbol of locks if any. This can be useful especially when it doesn't show the lock symbols. The indentation before the lock type parenthesis is to recognize lock symbols more easily. $ sudo ./perf lock con -abl -- sleep 1 contended total wait max wait avg wait address symbol 44 6.13 ms 284.49 us 139.28 us ffffffff92e06080 tasklist_lock (rwlock) 159 983.38 us 12.38 us 6.18 us ffff8cc717c90000 siglock (spinlock) 10 679.90 us 153.35 us 67.99 us ffff8cdc2872aaf8 mmap_lock (rwsem) 9 558.11 us 180.67 us 62.01 us ffff8cd647914038 mmap_lock (rwsem) 78 228.56 us 7.82 us 2.93 us ffff8cc700061c00 (spinlock) 5 41.60 us 16.93 us 8.32 us ffffd853acb41468 (spinlock) 10 37.24 us 5.87 us 3.72 us ffff8cd560b5c200 siglock (spinlock) 4 11.17 us 3.97 us 2.79 us ffff8d053ddf0c80 rq_lock (spinlock) 1 7.86 us 7.86 us 7.86 us ffff8cd64791404c (spinlock) 1 4.13 us 4.13 us 4.13 us ffff8d053d930c80 rq_lock (spinlock) 7 3.98 us 1.67 us 568 ns ffff8ccb92479440 (mutex) 2 2.62 us 2.33 us 1.31 us ffff8cc702e6ede0 (rwlock) Signed-off-by:Namhyung Kim <namhyung@kernel.org> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Hao Luo <haoluo@google.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Juri Lelli <juri.lelli@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Song Liu <song@kernel.org> Cc: Stephane Eranian <eranian@google.com> Cc: Waiman Long <longman@redhat.com> Cc: Will Deacon <will@kernel.org> Cc: bpf@vger.kernel.org Link: https://lore.kernel.org/r/20230313204825.2665483-2-namhyung@kernel.orgSigned-off-by:
-
Namhyung Kim authored
Using the BPF_PROG_RUN mechanism, we can run a raw_tp BPF program to collect some semi-global locks like per-cpu locks. Let's add runqueue locks using bpf_per_cpu_ptr() helper. $ sudo ./perf lock con -abl -- sleep 1 contended total wait max wait avg wait address symbol 248 3.25 ms 32.23 us 13.10 us ffff8cc75cfd2940 siglock 60 217.91 us 9.69 us 3.63 us ffff8cc700061c00 8 70.23 us 13.86 us 8.78 us ffff8cc703629484 4 56.32 us 35.81 us 14.08 us ffff8cc78b66f778 mmap_lock 4 16.70 us 5.18 us 4.18 us ffff8cc7036a0684 3 4.99 us 2.65 us 1.66 us ffff8d053da30c80 rq_lock 2 3.44 us 2.28 us 1.72 us ffff8d053dcf0c80 rq_lock 9 2.51 us 371 ns 278 ns ffff8ccb92479440 2 2.11 us 1.24 us 1.06 us ffff8d053db30c80 rq_lock 2 2.06 us 1.69 us 1.03 us ffff8d053d970c80 rq_lock Signed-off-by: -
Namhyung Kim authored
Likewise, we can display siglock by following the pointer like current->sighand->siglock. $ sudo ./perf lock con -abl -- sleep 1 contended total wait max wait avg wait address symbol 16 2.18 ms 305.35 us 136.34 us ffffffff92e06080 tasklist_lock 28 521.78 us 31.16 us 18.63 us ffff8cc703783ec4 7 119.03 us 23.55 us 17.00 us ffff8ccb92479440 15 88.29 us 10.06 us 5.89 us ffff8cd560b5f380 siglock 7 37.67 us 9.16 us 5.38 us ffff8d053daf0c80 5 8.81 us 4.92 us 1.76 us ffff8d053d6b0c80 Signed-off-by: -
Namhyung Kim authored
Sometimes there are severe contentions on the mmap_lock and we want see it in the -l/--lock-addr output. However it cannot symbolize the mmap_lock because it's allocated dynamically without symbols. Stephane and Hao gave me an idea separately to display mmap_lock by following the current->mm pointer. I added a flag to mark mmap_lock after comparing the lock address so that it can show them differently. With this change it can show mmap_lock like below: $ sudo ./perf lock con -abl -- sleep 10 contended total wait max wait avg wait address symbol ... 16344 312.30 ms 2.22 ms 19.11 us ffff8cc702595640 17686 310.08 ms 1.49 ms 17.53 us ffff8cc7025952c0 3 84.14 ms 45.79 ms 28.05 ms ffff8cc78114c478 mmap_lock 3557 76.80 ms 68.75 us 21.59 us ffff8cc77ca3af58 1 68.27 ms 68.27 ms 68.27 ms ffff8cda745dfd70 9 54.53 ms 7.96 ms 6.06 ms ffff8cc7642a48b8 mmap_lock 14629 44.01 ms 60.00 us 3.01 us ffff8cc7625f9ca0 3481 42.63 ms 140.71 us 12.24 us ffffffff937906ac vmap_area_lock 16194 38.73 ms 42.15 us 2.39 us ffff8cd397cbc560 11 38.44 ms 10.39 ms 3.49 ms ffff8ccd6d12fbb8 mmap_lock 1 5.43 ms 5.43 ms 5.43 ms ffff8cd70018f0d8 1674 5.38 ms 422.93 us 3.21 us ffffffff92e06080 tasklist_lock 581 4.51 ms 130.68 us 7.75 us ffff8cc9b1259058 5 3.52 ms 1.27 ms 703.23 us ffff8cc754510070 112 3.47 ms 56.47 us 31.02 us ffff8ccee38b3120 381 3.31 ms 73.44 us 8.69 us ffffffff93790690 purge_vmap_area_lock 255 3.19 ms 36.35 us 12.49 us ffff8d053ce30c80 Note that mmap_lock was renamed some time ago and it needs to support old kernels with a different name 'mmap_sem'. Suggested-by:Hao Luo <haoluo@google.com> Suggested-by:
Stephane Eranian <eranian@google.com> Signed-off-by:
-
Ian Rogers authored
Building without libelf support is going disable a lot of functionality. Require that the NO_LIBELF=1 build option is passed if this is intentional. Committer notes: Add NO_LIBELF=1 to the 'make_static' target in tools/perf/tests/make so that 'make -C tools/perf build-test' works. Signed-off-by:
-
Ian Rogers authored
The option controlled nothing and no code depends, conditional or otherwise, on libnewt. Signed-off-by:
-
Ian Rogers authored
If libtraceevent isn't present, the build will warn and continue. This disables a number of features and so isn't desirable. This change makes the build error for this case. The build can still be made to happen by adding NO_LIBTRACEEVENT=1. Committer notes: Add NO_LIBTRACEEVENT=1 to the 'make_static' target in tools/perf/tests/make so that 'make -C tools/perf build-test' works. Signed-off-by:
-
Ian Rogers authored
If libpfm4 passes the feature test, it would be nice to have it enabled rather than also requiring the LIBPFM4=1 build flag. Signed-off-by:
-
Ian Rogers authored
Refactor C++ demangling out of symbol-elf into its own files similar to other languages. Add abi::__cxa_demangle support. As the other demanglers are not shippable with distributions, this brings back C++ demangling in a common case. It isn't perfect as the support for optionally demangling arguments and modifiers isn't present. Signed-off-by:
-
Ian Rogers authored
cxxabi.h is part of libsdtc++ and LLVM's libcxx, providing abi::__cxa_demangle a portable C++ demangler. Add a feature test to detect that the function is available. Signed-off-by:
-
Ian Rogers authored
binutils is GPLv3 so distributions cannot ship perf linked against libbfd and libiberty as the licenses are incompatible. Rather than defaulting the build to opting in to libbfd and libiberty support and opting out via NO_LIBBFD=1 and NO_DEMANGLE=1, make building against the libraries optional and enabled with BUILD_NONDISTRO=1. Signed-off-by:
-
Ian Rogers authored
Rather than disabling jevents if a sufficient python isn't present error in the build. This avoids the build progressing but the binary being degraded. The build can still succeed by specifying NO_JEVENTS=1 to the build and this is conveyed in the error message. Signed-off-by:
-
Ian Rogers authored
sched_getcpu may not be present and so a feature test and definition exist to workaround this in the build. The feature test is used to define HAVE_SCHED_GETCPU_SUPPORT and so this is sufficient to know whether the local sched_getcpu is needed and a weak symbol can be avoided. Signed-off-by:
-
Ian Rogers authored
HAVE_GLIBC_SUPPORT is only used in `perf version --build-options` but doesn't control any behavior. Remove from the build to simplify it. Signed-off-by:
-
Ian Rogers authored
BPF skeleton support is now key to a number of perf features. Rather than making it so that BPF support must be enabled for the build, make this the default and error if the build lacks a clang and libbpf that are sufficient. To avoid the error and build without BPF skeletons the NO_BPF_SKEL=1 flag can be used. Add a build-options flag to 'perf version' to enable detection of the BPF skeleton support and use this in the offcpu shell test. Signed-off-by:
-
Ian Rogers authored
Add a build target to echo the python/perf.so's name from Makefile.perf. Use it in tests/make so the correct target is built and tested for. Fixes: caec5470 ("perf build: Fix python/perf.so library's name") Signed-off-by:
-
Ian Rogers authored
Tests are no longer applicable as libbpf 1.0 can be assumed. Signed-off-by:
-
Ian Rogers authored
The feature tests were necessary for libbpf pre-1.0, but as the libbpf implies at least 1.0 we can remove these now. Committer notes: Modified tools/perf/Makefile.config to better reflect the reason for failure when the libbpf present is < 1.0 and LIBBPF_DYNAMIC=1 was asked for. Signed-off-by:
Jiri Olsa <jolsa@kernel.org> Tested-by:
-
Ian Rogers authored
libbpf 1.0 represented a cleanup and stabilization of APIs. Simplify development by only passing the feature test if libbpf 1.0 is installed. Committer notes: Change 'make -C tools/perf build-test' so that the LIBBPF_DYNAMIC=1 test runs only if libbpf is >= 1.0. Signed-off-by:
-
- 13 Mar, 2023 17 commits
-
-
Ian Rogers authored
__has_builtin was passed the macro rather than the actual builtin feature. The builtin test isn't sufficient and a clang version test also needs to be performed. Fixes: 1bece135 ("perf lock contention: Support old rw_semaphore type") Reviewed-by:
-
James Clark authored
When checking for the presence of ETE, a register is read that may not be present on older kernels or if ETE isn't available. cs_etm_get_ro() will print a warning if it doesn't exist, so check for the existence first before accessing it. Reviewed-by:
Leo Yan <leo.yan@linaro.org> Signed-off-by:
James Clark <james.clark@arm.com> Cc: Al Grant <al.grant@arm.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ian Rogers <irogers@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: John Garry <john.g.garry@oracle.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mathieu Poirier <mathieu.poirier@linaro.org> Cc: Mike Leach <mike.leach@linaro.org> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Will Deacon <will@kernel.org> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Link: https://lore.kernel.org/r/20230308094843.287093-1-james.clark@arm.comSigned-off-by:
-
James Clark authored
This is printed as a warning but it is normal behavior that users shouldn't be expected to do anything about. Reduce the warning level to debug3 so it's only seen in verbose mode to avoid confusion. Reviewed-by: Leo Yan <leo.yan@linaro.org Signed-off-by:
-
Thomas Richter authored
Add the event description for the IBM z16 pai_ext PMU released with commit c432fefe ("s390/pai: Add support for PAI Extension 1 NNPA counters") The document SA22-7832-13 "z/Architecture Principles of Operation", published May, 2022, contains the description of the Processor Activity Instrumentation Facility and the NNPA counter set., See Pages 5-113 to 5-116 and chapter 26 for details. Signed-off-by:
Thomas Richter <tmricht@linux.ibm.com> Acked-by:
Sumanth Korikkar <sumanthk@linux.ibm.com> Cc: Heiko Carstens <hca@linux.ibm.com> Cc: Sven Schnelle <svens@linux.ibm.com> Cc: Vasily Gorbik <gor@linux.ibm.com> Link: https://lore.kernel.org/r/20230308125326.2195613-1-tmricht@linux.ibm.comSigned-off-by:
-
Thomas Richter authored
Add metrics for s390 z16 - Percentage sourced from Level 2 cache - Percentage sourced from Level 3 on same chip cache - Percentage sourced from Level 4 Local cache on same book - Percentage sourced from Level 4 Remote cache on different book - Percentage sourced from memory For details about the formulas see this documentation: https://www.ibm.com/support/pages/system/files/inline-files/CPU%20MF%20Formulas%20including%20z16%20-%20May%202022_1.pdf Output after: # ./perf stat -M l4rp -- dd if=/dev/zero of=/dev/null bs=10M count=10K .... dd output deleted Performance counter stats for 'dd if=/dev/zero of=/dev/null bs=10M count=10K': 0 IDCW_OFF_DRAWER_CHIP_HIT # 0.00 l4rp 431,866 L1I_DIR_WRITES 2,395 IDCW_OFF_DRAWER_IV 0 ICW_OFF_DRAWER 0 IDCW_OFF_DRAWER_DRAWER_HIT 1,437 DCW_OFF_DRAWER 425,960,793 L1D_DIR_WRITES 12.165030699 seconds time elapsed 0.001037000 seconds user 12.162140000 seconds sys # Signed-off-by:
-
Thomas Richter authored
Add 3 metrics for s390 machines: - Cycles per instruction: Amount of CPU cycles used per instructions, named cpi. - Problem state ratio: Ratio of instructions executed in problem state compared to total number of instructions, named prbstate. - Level one instruction and data cache misses per 100 instructions, named l1mp. For details about the formulas see this documentation: https://www.ibm.com/support/pages/system/files/inline-files/CPU%20MF%20Formulas%20including%20z16%20-%20May%202022_1.pdf Output after: # ./perf stat -M cpi -- dd if=/dev/zero of=/dev/null bs=1M count=10K 10240+0 records in 10240+0 records out 10737418240 bytes (11 GB, 10 GiB) copied, 1.30151 s, 8.2 GB/s Performance counter stats for 'dd if=/dev/zero of=/dev/null .....': 6,779,778,802 CPU_CYCLES # 1.96 cpi 3,461,975,090 INSTRUCTIONS 1.306873021 seconds time elapsed 0.001034000 seconds user 1.305677000 seconds sys # Signed-off-by:
-
Ian Rogers authored
Use if an event is reordered or the number of groups increases to signal that regrouping has happened and warn about it. Disable the warning in the case wild card PMU names are used and for metrics. Signed-off-by:
-
Ian Rogers authored
Maintaining the number of groups during event parsing is problematic and since changing to sort/regroup events can only be computed by a linear pass over the evlist. As the value is generally only used in tests, rather than hold it in a variable compute it by passing over the evlist when necessary. This change highlights that libpfm's counting of groups with a single entry disagreed with regular event parsing. The libpfm tests are updated accordingly. Signed-off-by:
-
Ian Rogers authored
This flag used to be used when regrouping uncore events in particular due to wildcard matches. This is now handled by sorting evlist and so the flag is redundant. Signed-off-by:
-
Ian Rogers authored
This change is intended to be a no-op for most current cases, the default sort order is the order the events were parsed. Where it varies is in how groups are handled. Previously an uncore and core event that are grouped would most often cause the group to be removed: ``` $ perf stat -e '{instructions,uncore_imc_free_running_0/data_total/}' -a sleep 1 WARNING: grouped events cpus do not match, disabling group: anon group { instructions, uncore_imc_free_running_0/data_total/ } ... ``` However, when wildcards are used the events should be re-sorted and re-grouped in parse_events__set_leader, but this currently fails for simple examples: ``` $ perf stat -e '{uncore_imc_free_running/data_read/,uncore_imc_free_running/data_write/}' -a sleep 1 Performance counter stats for 'system wide': <not counted> MiB uncore_imc_free_running/data_read/ <not counted> MiB uncore_imc_free_running/data_write/ 1.000996992 seconds time elapsed ``` A futher failure mode, fixed in this patch, is to force topdown events into a group. This change moves sorting the evsels in the evlist after parsing. It requires parsing to set up groups. First the evsels are sorted respecting the existing groupings and parse order, but also reordering to ensure evsels of the same PMU and group appear together. So that software and aux events respect groups, their pmu_name is taken from the group leader. The sorting is done with list_sort removing a memory allocation. After sorting a pass is done to correct the group leaders and for topdown events ensuring they have a group leader. This fixes the problems seen before: ``` $ perf stat -e '{uncore_imc_free_running/data_read/,uncore_imc_free_running/data_write/}' -a sleep 1 Performance counter stats for 'system wide': 727.42 MiB uncore_imc_free_running/data_read/ 81.84 MiB uncore_imc_free_running/data_write/ 1.000948615 seconds time elapsed ``` As well as making groups not fail for cases like: ``` $ perf stat -e '{imc_free_running_0/data_total/,imc_free_running_1/data_total/}' -a sleep 1 Performance counter stats for 'system wide': 256.47 MiB imc_free_running_0/data_total/ 256.48 MiB imc_free_running_1/data_total/ 1.001165442 seconds time elapsed ``` Signed-off-by: -
Ian Rogers authored
Pass ownership of the group name rather than copying and freeing the original. This saves a memory allocation and copy. Signed-off-by:
-
Ian Rogers authored
The computed name respects software events and aux event groups, such that the pmu_name is changed to be that of the aux event leader or group leader for software events. This is done as a later change will split events that are in different PMUs into different groups. Committer notes: Added a stub for this new function so that 'perf test python' passes. Signed-off-by:
-
Ian Rogers authored
List sorting, added later to evlist, passes const elements requiring helper functions to also be const. Make the argument to evsel__find_pmu, evsel__is_aux_event and evsel__leader const. Signed-off-by:
-
Ian Rogers authored
Currently nr_members is 0 for an event with no group, however, they are always a leader of their own group. A later change will make that count 1 because the event is its own leader. Make the find_stat logic consistent with this, an improvement suggested by Namhyung Kim. Suggested-by:
-
Ian Rogers authored
This allows event parsing to use the evsel__is_aux_event function, which is important when determining event grouping. Suggested-by:
-
Ian Rogers authored
If the events in an evlist's CPU map differ then the entire group is removed. For example: ``` $ perf stat -e '{imc_free_running/data_read/,imc_free_running/data_write/,cs}' -a sleep 1 WARNING: grouped events cpus do not match, disabling group: anon group { imc_free_running/data_read/, imc_free_running/data_write/, cs } ``` Change the behavior so that just the events not matching the leader are removed. So in the example above, just 'cs' will be removed. Modify the warning so that it is produced once for each group, rather than once for the entire evlist. Shrink the scope and size of the warning text buffer. Signed-off-by: -
Ian Rogers authored
Setting the leader iterates the list, so rather than use idx (which may be changed through list reordering) just count the elements and set afterwards. Signed-off-by:
-
