- 12 Jan, 2022 35 commits
-
-
Ian Rogers authored
Correct in print_counter() where an index was being used as a cpu. Signed-off-by:
Ian Rogers <irogers@google.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: John Garry <john.garry@huawei.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> 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: Paul Clarke <pc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Stephane Eranian <eranian@google.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Vineet Singh <vineet.singh@intel.com> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Cc: zhengjun.xing@intel.com Link: https://lore.kernel.org/r/20220105061351.120843-32-irogers@google.comSigned-off-by:
Arnaldo Carvalho de Melo <acme@redhat.com>
-
Ian Rogers authored
Trying to make cpu maps less error prone. Signed-off-by:
-
Ian Rogers authored
Try to reduce confusion in particular when the cpu map doesn't contain an entry for every CPU. Signed-off-by:
-
Ian Rogers authored
Fix issue where evsel's CPU map index was being used as the mmap cpu. Signed-off-by:
-
Ian Rogers authored
Modify variable names and adopt perf_cpu_map__for_each_cpu() in perf_evsel__open(). Renaming is done by looking for consistency in API usage. Signed-off-by:
-
Ian Rogers authored
Passing the number of CPUs and threads allows for an evsel's counts to be mismatched to its cpu map. To avoid this always derive the counts size from the cpu map. Change openat-syscall-all-cpus to set the cpus to allow for this to work. Signed-off-by:
-
Ian Rogers authored
Correct use of cpumap index in print_no_aggr_metric(). Signed-off-by:
-
Ian Rogers authored
With no aggregration, such as 'perf stat -A', the aggr_cpu_id lacks a way to describe per CPU aggregation and the core is set to the CPU in places like print_counter_aggrdata in stat-display.c. Setting the core to the CPU is undesirable as the CPU will exceed valid core values and lead to confusion. Add a CPU variable to address this. Signed-off-by:
-
Ian Rogers authored
If the key is already present then free the key used for lookup. Found with: $ perf stat -M IO_Read_BW /bin/true ==1749112==ERROR: LeakSanitizer: detected memory leaks Direct leak of 32 byte(s) in 4 object(s) allocated from: #0 0x7f6f6fa7d7cf in __interceptor_malloc ../../../../src/libsanitizer/asan/asan_malloc_linux.cpp:145 #1 0x55acecd9d7a6 in check_per_pkg util/stat.c:343 #2 0x55acecd9d9c5 in process_counter_values util/stat.c:365 #3 0x55acecd9e0ab in process_counter_maps util/stat.c:421 #4 0x55acecd9e292 in perf_stat_process_counter util/stat.c:443 #5 0x55aceca8553e in read_counters ./tools/perf/builtin-stat.c:470 #6 0x55aceca88fe3 in __run_perf_stat ./tools/perf/builtin-stat.c:1023 #7 0x55aceca89146 in run_perf_stat ./tools/perf/builtin-stat.c:1048 #8 0x55aceca90858 in cmd_stat ./tools/perf/builtin-stat.c:2555 #9 0x55acecc05fa5 in run_builtin ./tools/perf/perf.c:313 #10 0x55acecc064fe in handle_internal_command ./tools/perf/perf.c:365 #11 0x55acecc068bb in run_argv ./tools/perf/perf.c:409 #12 0x55acecc070aa in main ./tools/perf/perf.c:539 Reviewed-by:James Clark <james.clark@arm.com> Signed-off-by:
-
Ian Rogers authored
cpu_aggr_map__new() removes duplicates, when this happens shrink the array. Reviewed-by:
-
Ian Rogers authored
Move cpu_aggr_map__empty_new() to be with other cpu_aggr_map function. Reviewed-by:
-
Ian Rogers authored
Make the cpu map argument const for consistency with the rest of the API. Modify cpu_map__idx accordingly. Reviewed-by:
-
Ian Rogers authored
The functions don't use a cpu_map so reduce them to being like constructors of aggr_cpu_id. Reviewed-by:
-
Ian Rogers authored
Turn it into a cpu_aggr_map__new(). Pass helper functions. Refactor builtin-stat calls to manually pass function pointers. Try to reduce some copy-paste code. Reviewed-by:
-
Ian Rogers authored
Switch the remaining few users of cpu_map__cpu() to perf_cpu_map__cpu() and remove the function. Reviewed-by:
-
Ian Rogers authored
Move to the cpu name and document for consistency. Reviewed-by:
-
Ian Rogers authored
cpu_map__get_node_id() isn't used externally and merely delegates to cpu__get_node(). Reviewed-by:
-
Ian Rogers authored
Remove cpu_map from name as a cpu_map isn't used. Pass a const pointer rather than by value to avoid unnecessary copying. Reviewed-by:
-
Ian Rogers authored
Rename cpu_map__compare_aggr_cpu_id() to aggr_cpu_id__equal(), the cpu_map part of the name is misleading. Equal better describes the function than compare. Switch to const pointer rather than value as struct given the number of variables in aggr_cpu_id(). Reviewed-by:
-
Ian Rogers authored
Unused function so remove. Reviewed-by:
-
Ian Rogers authored
This code is already tested in topology.c. Reviewed-by:
-
Ian Rogers authored
Migrate final users to appropriate cpu variant. Reviewed-by:
-
Ian Rogers authored
Migrate final users to appropriate cpu variant. Reviewed-by:
-
Ian Rogers authored
Migrate final users to appropriate cpu variant. Reviewed-by:
-
Ian Rogers authored
Migrate final users to appropriate cpu variant. Reviewed-by:
-
Ian Rogers authored
Avoid error prone cpu_map + idx variant. Remove now unused functions. Committer notes: Remove by now unused perf_env__get_cpu(). Reviewed-by:
-
Ian Rogers authored
Avoid possible bugs where the wrong index is passed with the cpu_map. Reviewed-by:
-
Ian Rogers authored
Tidy up the use of cpu and index to hopefully make the code less error prone. Avoid unused warnings with (void) which will be removed in a later patch. Reviewed-by:
-
Ian Rogers authored
Switch the perf_cpu_map in aggr_update_shadow from the evlist to the counter's cpu map, so the index is appropriate. This addresses a problem where uncore counts, with a cpumap like: $ cat /sys/devices/uncore_imc_0/cpumask 0,18 Don't aggregate counts in CPUs based on the index of those values in the cpumap (0 and 1) but on the actual CPU (0 and 18). Thereby correcting metric calculations in per-socket mode for counters without a full cpumask. On a SkylakeX with a tweaked DRAM_BW_Use metric, to remove unnecessary scaling, this gives: Before: $ /perf stat --per-socket -M DRAM_BW_Use -I 1000 1.001102293 S0 1 27.01 MiB uncore_imc/cas_count_write/ # 103.00 DRAM_BW_Use 1.001102293 S0 1 30.22 MiB uncore_imc/cas_count_read/ 1.001102293 S0 1 1,001,102,293 ns duration_time 1.001102293 S1 1 20.10 MiB uncore_imc/cas_count_write/ # 0.00 DRAM_BW_Use 1.001102293 S1 1 32.74 MiB uncore_imc/cas_count_read/ 1.001102293 S1 0 <not counted> ns duration_time 2.003517973 S0 1 83.04 MiB uncore_imc/cas_count_write/ # 920.00 DRAM_BW_Use 2.003517973 S0 1 145.95 MiB uncore_imc/cas_count_read/ 2.003517973 S0 1 1,002,415,680 ns duration_time 2.003517973 S1 1 302.45 MiB uncore_imc/cas_count_write/ # 0.00 DRAM_BW_Use 2.003517973 S1 1 290.99 MiB uncore_imc/cas_count_read/ 2.003517973 S1 0 <not counted> ns duration_time After: $ perf stat --per-socket -M DRAM_BW_Use -I 1000 1.001080840 S0 1 24.96 MiB uncore_imc/cas_count_write/ # 54.00 DRAM_BW_Use 1.001080840 S0 1 33.64 MiB uncore_imc/cas_count_read/ 1.001080840 S0 1 1,001,080,840 ns duration_time 1.001080840 S1 1 42.43 MiB uncore_imc/cas_count_write/ # 84.00 DRAM_BW_Use 1.001080840 S1 1 47.05 MiB uncore_imc/cas_count_read/ 1.001080840 S1 0 <not counted> ns duration_time Signed-off-by:John Garry <john.garry@huawei.com> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Clark <james.clark@arm.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Leo Yan <leo.yan@linaro.org> 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: Paul Clarke <pc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Riccardo Mancini <rickyman7@gmail.com> Cc: Stephane Eranian <eranian@google.com> Cc: Suzuki Poulouse <suzuki.poulose@arm.com> Cc: Vineet Singh <vineet.singh@intel.com> Cc: coresight@lists.linaro.org Cc: linux-arm-kernel@lists.infradead.org Cc: zhengjun.xing@intel.com Link: https://lore.kernel.org/r/20220105061351.120843-4-irogers@google.comSigned-off-by:
-
Ian Rogers authored
The cpu_map and index can get confused. Add variants of the cpu_map__get routines that are passed a cpu. Make the existing cpu_map__get routines use the new functions with a view to remove them when no longer used. Reviewed-by:
-
Ian Rogers authored
A particular observed problem is confusing the index with the CPU value, documentation should hopefully reduce this type of problem. Reviewed-by:
-
Ian Rogers authored
When a group has multiple events and the leader fails it can yield errors like: $ perf stat -e '{uncore_imc/cas_count_read/},instructions' /bin/true Error: The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (uncore_imc/cas_count_read/). /bin/dmesg | grep -i perf may provide additional information. However, when not the group leader <not supported> is given: $ perf stat -e '{instructions,uncore_imc/cas_count_read/}' /bin/true ... 1,619,057 instructions <not supported> MiB uncore_imc/cas_count_read/ This is necessary because get_group_fd will fail if the leader fails and is the direct result of the check on line 750 of builtin-stat.c in stat_handle_error that returns COUNTER_SKIP for the latter case. This patch improves the error message to: $ perf stat -e '{uncore_imc/cas_count_read/},instructions' /bin/true Error: Invalid event (uncore_imc/cas_count_read/) in per-thread mode, enable system wide with '-a'. v2. Changed the test to use !target__has_cpu as suggested by Namhyung Kim. Signed-off-by: -
Arnaldo Carvalho de Melo authored
This was in a patchkit mixing up kernel with tools/ parts and I mistakenly got it merged in the perf tools tree, revert it, it'll go via the PowerPC kernel tree. This reverts commit af2b24f2. Cc: kajoljain <kjain@linux.ibm.com> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Stephen Rothwell <sfr@canb.auug.org.au> Link: http://lore.kernel.org/lkml/20220112171659.531d22ce@canb.auug.org.auSigned-off-by:
-
Arnaldo Carvalho de Melo authored
Revert "perf powerpc: Add encodings to represent data based on newer composite PERF_MEM_LVLNUM* fields" This was in a patchkit mixing up kernel with tools/ parts and I mistakenly got it merged in the perf tools tree, revert it, it'll go via the PowerPC kernel tree. This reverts commit 0ebce3d6. Cc: kajoljain <kjain@linux.ibm.com> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Stephen Rothwell <sfr@canb.auug.org.au> Link: http://lore.kernel.org/lkml/20220112171659.531d22ce@canb.auug.org.auSigned-off-by:
-
Adrian Hunter authored
Using grep -C with perf script -D can give erroneous results as grep loses lines due to non-printable characters, for example, below the 0020, 0060 and 0070 lines are missing: $ perf script -D | grep -C10 AUX | head . 0010: 08 00 00 00 00 00 00 00 1f 00 00 00 00 00 00 00 ................ . 0030: 01 00 00 00 00 00 00 00 00 04 00 00 00 00 00 00 ................ . 0040: 00 08 00 00 00 00 00 00 02 00 00 00 00 00 00 00 ................ . 0050: 00 00 00 00 00 00 00 00 01 00 00 00 00 00 00 00 ................ . 0080: 02 00 00 00 00 00 00 00 1b 00 00 00 00 00 00 00 ................ . 0090: 00 00 00 00 00 00 00 00 ........ 0 0 0x450 [0x98]: PERF_RECORD_AUXTRACE_INFO type: 1 PMU Type 8 Time Shift 31 perf's isprint() is a custom implementation from the kernel, but the kernel's _ctype appears to include characters from Latin-1 Supplement which is not compatible with, for example, UTF-8. Fix by checking also isascii(). After: $ tools/perf/perf script -D | grep -C10 AUX | head . 0010: 08 00 00 00 00 00 00 00 1f 00 00 00 00 00 00 00 ................ . 0020: 03 84 32 2f 00 00 00 00 63 7c 4f d2 fa ff ff ff ..2/....c|O..... . 0030: 01 00 00 00 00 00 00 00 00 04 00 00 00 00 00 00 ................ . 0040: 00 08 00 00 00 00 00 00 02 00 00 00 00 00 00 00 ................ . 0050: 00 00 00 00 00 00 00 00 01 00 00 00 00 00 00 00 ................ . 0060: 00 02 00 00 00 00 00 00 00 c0 03 00 00 00 00 00 ................ . 0070: e2 00 00 00 00 00 00 00 02 00 00 00 00 00 00 00 ................ . 0080: 02 00 00 00 00 00 00 00 1b 00 00 00 00 00 00 00 ................ . 0090: 00 00 00 00 00 00 00 00 ........ Fixes: 3052ba56 ("tools perf: Move from sane_ctype.h obtained from git to the Linux's original") Signed-off-by:
Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Link: http://lore.kernel.org/lkml/20220112085057.277205-1-adrian.hunter@intel.comSigned-off-by:
-
- 11 Jan, 2022 1 commit
-
-
Ian Rogers authored
Uncore events as group leaders fail in per-thread mode causing exit errors. Enable system-wide for metricgroup testing. This fixes the HPC metric group when tested on skylakex. Fixes: 4a87dea9 ("perf test: Workload test of metric and metricgroups") Signed-off-by:
-
- 10 Jan, 2022 4 commits
-
-
Dario Petrillo authored
In 'perf report', entering a recursive function from inside of itself (either directly of indirectly through some other function) results in calling symbol__annotate2 multiple() times, and freeing the whole disassembly when exiting from the innermost instance. The first issue causes the function's disassembly to be duplicated, and the latter a heap use-after-free (and crash) when trying to access the disassembly again. I reproduced the bug on perf 5.11.22 (Ubuntu 20.04.3 LTS) and 5.16.rc8 with the following testcase (compile with gcc recursive.c -o recursive). To reproduce: - perf record ./recursive - perf report - enter fibonacci and annotate it - move the cursor on one of the "callq fibonacci" instructions and press enter - at this point there will be two copies of the function in the disassembly - go back by pressing q, and perf will crash #include <stdio.h> int fibonacci(int n) { if(n <= 2) return 1; return fibonacci(n-1) + fibonacci(n-2); } int main() { printf("%d\n", fibonacci(40)); } This patch addresses the issue by annotating a function and freeing the associated memory on exit only if no annotation is already present, so that a recursive function is only annotated on entry. Signed-off-by:Dario Petrillo <dario.pk1@gmail.com> Tested-by:
-
Athira Rajeev authored
Update the arch_support_sort_key() function in powerpc to enable presenting local and global variants of sort key 'p_stage_cyc'. Update the "se_header" strings for these in arch_perf_header_entry() along with instruction latency. Reported-by:
Namhyung Kim <namhyung@kernel.org> Signed-off-by:
Athira Jajeev <atrajeev@linux.vnet.ibm.com> Tested-by:
Nageswara R Sastry <rnsastry@linux.ibm.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Kajol Jain <kjain@linux.ibm.com> Cc: Madhavan Srinivasan <maddy@linux.vnet.ibm.com> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: linuxppc-dev@lists.ozlabs.org Link: https://lore.kernel.org/r/20211203022038.48240-2-atrajeev@linux.vnet.ibm.comSigned-off-by:
-
Athira Rajeev authored
Sort key 'p_stage_cyc' is used to present the latency cycles spent in pipeline stages. perf has local 'p_stage_cyc' sort key to display this info. There is no global variant available for this sort key. The local variant shows latency in a single sample, whereas the global value will be useful to present the total latency (sum of latencies) in the hist entry. It represents the latency number multiplied by the number of samples. Add global ('p_stage_cyc') and local variant ('local_p_stage_cyc') for this sort key. Use 'local_p_stage_cyc' as default option for "mem" sort mode. Also add this to the list of dynamic sort keys and made the "dynamic_headers" and "arch_specific_sort_keys" as static. Reported-by: -
Arnaldo Carvalho de Melo authored
To pick up fixes. Signed-off-by:
-
