Perf stat --null/offline CPU segv related fixes/tests
Ian Rogers
reviewerv2 cover Perf stat --null/offline CPU segv related fixes/tests
Ingo reported [1] that perf stat --null was segfaulting. Fix the underlying issue and add a test to the "perf stat tests". Do some related fixing/cleanup in the perf util cpumap code.
Thomas reported an issue fixed by the same patches [2] but caused by giving perf stat an offline CPU. Add test coverage for that and improve the "error" message that reports "success".
Ingo further pointed at broken signal handling in repeat mode [3]. I observed we weren't giving the best exit code, 0 rather than the expected 128+<signal number>. Add a patch fixing this.
[1] https://lore.kernel.org/linux-perf-users/[email protected]/
[2] https://lore.kernel.org/linux-perf-users/[email protected]/
[3] https://lore.kernel.org/lkml/[email protected]/
Ian Rogers (7):
perf stat: Allow no events to open if this is a "--null" run libperf cpumap: Fix perf_cpu_map__max for an empty/NULL map perf cpumap: Add "any" CPU handling to cpu_map__snprint_mask perf tests stat: Add "--null" coverage
perf stat: When no events, don't report an error if there is none perf tests stat: Add test for error for an offline CPU perf stat: Improve handling of termination by signal
tools/lib/perf/cpumap.c | 10 +++++---- tools/perf/builtin-stat.c | 29 ++++++++++++++++++------- tools/perf/tests/shell/stat.sh | 39 ++++++++++++++++++++++++++++++++++ tools/perf/util/cpumap.c | 9 ++++++-- 4 files changed, 73 insertions(+), 14 deletions(-)
Signature · 1 line
2.52.0.177.g9f829587af-goog
Ian Rogers
reviewerPATCH v2 1/7 perf stat: Allow no events to open if this is a "--null" run
It is intended that a "--null" run doesn't open any events.
--- tools/perf/builtin-stat.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/builtin-stat.c +1 -1
@@ -923,7 +923,7 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) goto err_out; } } -if (!has_supported_counters) { +if (!has_supported_counters && !stat_config.null_run) {evsel__open_strerror (evlist__first (evsel_list), &target, open_err, msg,sizeof (msg));ui__error ("No supported events found.\n%s\n" , msg);
Signature · 1 line
2.52.0.177.g9f829587af-goog
Ian Rogers
reviewerPATCH v2 2/7 libperf cpumap: Fix perf_cpu_map__max for an empty/NULL map
In reply to Ian RogersPassing an empty map to perf_cpu_map__max triggered a SEGV. Explicitly test for the empty map.
--- tools/lib/perf/cpumap.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-)
tools/lib/perf/cpumap.c +6 -4
@@ -368,10 +368,12 @@ struct perf_cpu perf_cpu_map__max(const struct perf_cpu_map *map) .cpu = -1 }; -// cpu_map__trim_new() qsort()s it, cpu_map__default_new() sorts it as well. -return __perf_cpu_map__nr (map) >0 - ?__perf_cpu_map__cpu (map,__perf_cpu_map__nr (map) -1 ) - : result; +if (!map) +return result; + +// The CPUs are always sorted and nr is always > 0 as 0 length map is +// encoded as NULL. +return __perf_cpu_map__cpu (map,__perf_cpu_map__nr (map) -1 ); }/** Is 'b' a subset of 'a'. */
Signature · 1 line
2.52.0.177.g9f829587af-goog
Ian Rogers
reviewerPATCH v2 3/7 perf cpumap: Add "any" CPU handling to cpu_map__snprint_mask
In reply to Ian RogersIf the perf_cpu_map is empty or is just the any CPU value, then early return. Don't process the "any" CPU when creating the bitmap.
--- tools/perf/util/cpumap.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)
tools/perf/util/cpumap.c +7 -2
@@ -684,16 +684,21 @@ size_t cpu_map__snprint_mask(struct perf_cpu_map *map, char *buf, size_t size)unsigned char *bitmap;struct perf_cpu c, last_cpu =perf_cpu_map__max (map); -if (buf ==NULL ) +if (buf ==NULL || size ==0 )return 0 ; +if (last_cpu.cpu <0 ) { + buf[0 ] ='\0' ; +return 0 ; + } + bitmap =zalloc (last_cpu.cpu /8 +1 );if (bitmap ==NULL ) { buf[0 ] ='\0' ;return 0 ; } -perf_cpu_map__for_each_cpu (c, idx, map) +perf_cpu_map__for_each_cpu_skip_any (c, idx, map) bitmap[c.cpu /8 ] |=1 << (c.cpu %8 );for (int cpu = last_cpu.cpu /4 *4 ; cpu >=0 ; cpu -=4 ) {
Signature · 1 line
2.52.0.177.g9f829587af-goog
Ian Rogers
reviewerPATCH v2 4/7 perf tests stat: Add "--null" coverage
In reply to Ian RogersEnsure "--null" does a minimal run.
--- tools/perf/tests/shell/stat.sh | 12 ++++++++++++ 1 file changed, 12 insertions(+)
tools/perf/tests/shell/stat.sh +12 -0
@@ -16,6 +16,17 @@ test_default_stat() { echo "Basic stat command test [Success]" } +test_null_stat() { + echo "Null stat command test" + if ! perf stat --null true 2>&1 | grep -E -q "Performance counter stats for 'true':" + then + echo "Null stat command test [Failed]" + err=1 + return + fi + echo "Null stat command test [Success]" +} + test_stat_record_report() { echo "stat record and report test" if ! perf stat record -e task-clock -o - true | perf stat report -i - 2>&1 | \
@@ -212,6 +223,7 @@ test_hybrid() { } test_default_stat +test_null_stat test_stat_record_report test_stat_record_script test_stat_repeat_weak_groups
Signature · 1 line
2.52.0.177.g9f829587af-goog
Ian Rogers
reviewerPATCH v2 5/7 perf stat: When no events, don't report an error if there is none
In reply to Ian RogersEvents may fail to open as no supported CPUs were specified on the command line. In this case a confusing "error" message of "success" can be reported. Let's skip the error in that case.
Before:
``` $ perf stat -C2048 -e cycles -- true WARNING: A requested CPU in '2048' is not supported by PMU 'cpu' (CPUs 0-7) for event 'cycles' Error: No supported events found. The sys_perf_event_open() syscall returned with 0 (Success) for event (cpu/unknown-hardware/). "dmesg | grep -i perf" may provide additional information. ```
After:
``` $ perf stat -C2048 -e cycles -- true WARNING: A requested CPU in '2048' is not supported by PMU 'cpu' (CPUs 0-7) for event 'cycles' Error: No supported events found. ```
--- tools/perf/builtin-stat.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
tools/perf/builtin-stat.c +4 -2
@@ -924,8 +924,10 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) } }if (!has_supported_counters && !stat_config.null_run) { -evsel__open_strerror (evlist__first (evsel_list), &target, open_err, - msg,sizeof (msg)); +if (open_err) { +evsel__open_strerror (evlist__first (evsel_list), &target, open_err, + msg,sizeof (msg)); + }ui__error ("No supported events found.\n%s\n" , msg);if (child_pid !=-1 )
Signature · 1 line
2.52.0.177.g9f829587af-goog
Ian Rogers
reviewerPATCH v2 6/7 perf tests stat: Add test for error for an offline CPU
In reply to Ian RogersAdd a test that if an offline CPU is requested perf stat will fail.
``` $ perf test -vv "perf stat tests" 101: perf stat tests: --- start --- test child forked, pid 46965 Basic stat command test Basic stat command test [Success] Null stat command test Null stat command test [Success] Offline CPU stat command test (cpu 8) Offline CPU stat command test [Success] stat record and report test stat record and report test [Success] stat record and script test stat record and script test [Success] stat repeat weak groups test stat repeat weak groups test [Success] Topdown event group test Topdown event group test [Success] Topdown weak groups test Topdown weak groups test [Skipped event parsing failed] cputype test cputype test [Success] hybrid test hybrid test [Success] ---- end(0) ---- 101: perf stat tests : Ok ```
--- tools/perf/tests/shell/stat.sh | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
tools/perf/tests/shell/stat.sh +27 -0
@@ -27,6 +27,32 @@ test_null_stat() { echo "Null stat command test [Success]" } +find_offline_cpu() { + for i in $(seq 1 4096) + do + if [[ ! -f /sys/devices/system/cpu/cpu$i/online || \ + $(cat /sys/devices/system/cpu/cpu$i/online) == "0" ]] + then + echo $i + return + fi + done + echo "Failed to find offline CPU" + exit 1 +} + +test_offline_cpu_stat() { + cpu=$(find_offline_cpu) + echo "Offline CPU stat command test (cpu $cpu)" + if ! perf stat "-C$cpu" -e cycles true 2>&1 | grep -E -q "No supported events found." + then + echo "Offline CPU stat command test [Failed]" + err=1 + return + fi + echo "Offline CPU stat command test [Success]" +} + test_stat_record_report() { echo "stat record and report test" if ! perf stat record -e task-clock -o - true | perf stat report -i - 2>&1 | \
@@ -224,6 +250,7 @@ test_hybrid() { test_default_stat test_null_stat +test_offline_cpu_stat test_stat_record_report test_stat_record_script test_stat_repeat_weak_groups
Signature · 1 line
2.52.0.177.g9f829587af-goog
Ian Rogers
reviewerPATCH v2 7/7 perf stat: Improve handling of termination by signal
In reply to Ian RogersWhen interrupting perf stat in repeat mode with a signal the signal is passed to the child process but the repeat doesn't terminate:
```
$ perf stat -v --null --repeat 10 sleep 1
Control descriptor is not initialized
[ perf stat: executing run #1 ... ]
[ perf stat: executing run #2 ... ]
^Csleep: Interrupt
[ perf stat: executing run #3 ... ]
[ perf stat: executing run #4 ... ]
[ perf stat: executing run #5 ... ]
[ perf stat: executing run #6 ... ]
[ perf stat: executing run #7 ... ]
[ perf stat: executing run #8 ... ]
[ perf stat: executing run #9 ... ]
[ perf stat: executing run #10 ... ]
Performance counter stats for 'sleep 1' (10 runs):
0.9500 +- 0.0512 seconds time elapsed ( +- 5.39% )
0.01user 0.02system 0:09.53elapsed 0%CPU (0avgtext+0avgdata 18940maxresident)k
29944inputs+0outputs (0major+2629minor)pagefaults 0swaps
```Terminate the repeated run and give a reasonable exit value:
```
$ perf stat -v --null --repeat 10 sleep 1
Control descriptor is not initialized
[ perf stat: executing run #1 ... ]
[ perf stat: executing run #2 ... ]
[ perf stat: executing run #3 ... ]
^Csleep: Interrupt
Performance counter stats for 'sleep 1' (10 runs):
0.680 +- 0.321 seconds time elapsed ( +- 47.16% )
Command exited with non-zero status 130
0.00user 0.01system 0:02.05elapsed 0%CPU (0avgtext+0avgdata 70688maxresident)k
0inputs+0outputs (0major+5002minor)pagefaults 0swaps
```Note, this also changes the exit value for non-repeat runs when interrupted by a signal.
--- tools/perf/builtin-stat.c | 21 ++++++++++++++++----- 1 file changed, 16 insertions(+), 5 deletions(-)
tools/perf/builtin-stat.c +16 -5
@@ -1007,10 +1007,20 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) goto err_out; } -if (WIFSIGNALED (status)) +if (WIFSIGNALED (status)) { +/* +* We want to indicate failure to stop a repeat run, +* hence negative. We want the value to be the exit code +* of perf, which for termination by a signal is 128 +* plus the signal number. +*/ + err =0 - (128 +WTERMSIG (status));psignal (WTERMSIG (status), argv[0 ]); + }else { + err =WEXITSTATUS (status); + } }else { - status =dispatch_events (forks, timeout, interval, ×); + err =dispatch_events (forks, timeout, interval, ×); }disable_counters ();
@@ -1050,7 +1060,7 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)if (!STAT_RECORD )evlist__close (evsel_list); -return WEXITSTATUS (status); +return err; err_out:if (forks)
@@ -2969,7 +2979,7 @@ int cmd_stat(int argc, const char **argv)evlist__reset_prev_raw_counts (evsel_list); status =run_perf_stat (argc, argv, run_idx); -if (status ==-1 ) +if (status <0 )break ;if (forever && !interval) {
@@ -3039,5 +3049,6 @@ int cmd_stat(int argc, const char **argv)evlist__close_control (stat_config.ctl_fd, stat_config.ctl_fd_ack, &stat_config.ctl_fd_close); -return status; +/* Only the low byte of status becomes the exit code. */ +return abs (status); }
Signature · 1 line
2.52.0.177.g9f829587af-goog
Quoted 25 lines · Ian Rogers
Ingo reported [1] that
perf stat --nullwas segfaulting. Fix the underlying issue and add a test to the "perf stat tests". Do some related fixing/cleanup in the perf util cpumap code.Thomas reported an issue fixed by the same patches [2] but caused by giving perf stat an offline CPU. Add test coverage for that and improve the "error" message that reports "success".
Ingo further pointed at broken signal handling in repeat mode [3]. I observed we weren't giving the best exit code, 0 rather than the expected 128+<signal number>. Add a patch fixing this.
[1] https://lore.kernel.org/linux-perf-users/[email protected]/
[2] https://lore.kernel.org/linux-perf-users/[email protected]/
[3] https://lore.kernel.org/lkml/[email protected]/Ian Rogers (7):
perf stat: Allow no events to open if this is a "--null" run libperf cpumap: Fixperf_cpu_map__maxfor an empty/NULL map perf cpumap: Add "any" CPU handling tocpu_map__snprint_maskperf tests stat: Add "--null" coverage
perf stat: When no events, don't report an error if there is none perf tests stat: Add test for error for an offline CPU perf stat: Improve handling of termination by signal
Quoted 6 lines
tools/lib/perf/cpumap.c | 10 +++++---- tools/perf/builtin-stat.c | 29 ++++++++++++++++++------- tools/perf/tests/shell/stat.sh | 39 ++++++++++++++++++++++++++++++++++ tools/perf/util/cpumap.c | 9 ++++++-- 4 files changed, 73 insertions(+), 14 deletions(-)
For the whole series:
Thanks for fixing this, Ian
Signature · 9 lines
Thomas Richter, Dept 3303, IBM s390 Linux Development, Boeblingen, Germany -- IBM Deutschland Research & Development GmbH Vorsitzender des Aufsichtsrats: Wolfgang Wendt Geschäftsführung: David Faller Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294
Quoted 9 lines · Ian Rogers
Ingo reported [1] that
perf stat --nullwas segfaulting. Fix the underlying issue and add a test to the "perf stat tests". Do some related fixing/cleanup in the perf util cpumap code.Thomas reported an issue fixed by the same patches [2] but caused by giving perf stat an offline CPU. Add test coverage for that and improve the "error" message that reports "success".
[...]
Applied to perf-tools-next, thanks!
Best regards,
Namhyung
Quoted 25 lines · Ian Rogers
Ingo reported [1] that
perf stat --nullwas segfaulting. Fix the underlying issue and add a test to the "perf stat tests". Do some related fixing/cleanup in the perf util cpumap code.Thomas reported an issue fixed by the same patches [2] but caused by giving perf stat an offline CPU. Add test coverage for that and improve the "error" message that reports "success".
Ingo further pointed at broken signal handling in repeat mode [3]. I observed we weren't giving the best exit code, 0 rather than the expected 128+<signal number>. Add a patch fixing this.
[1] https://lore.kernel.org/linux-perf-users/[email protected]/
[2] https://lore.kernel.org/linux-perf-users/[email protected]/
[3] https://lore.kernel.org/lkml/[email protected]/Ian Rogers (7):
perf stat: Allow no events to open if this is a "--null" run libperf cpumap: Fixperf_cpu_map__maxfor an empty/NULL map perf cpumap: Add "any" CPU handling tocpu_map__snprint_maskperf tests stat: Add "--null" coverage
perf stat: When no events, don't report an error if there is none perf tests stat: Add test for error for an offline CPU perf stat: Improve handling of termination by signal
Quoted 5 lines
tools/lib/perf/cpumap.c | 10 +++++---- tools/perf/builtin-stat.c | 29 ++++++++++++++++++------- tools/perf/tests/shell/stat.sh | 39 ++++++++++++++++++++++++++++++++++ tools/perf/util/cpumap.c | 9 ++++++-- 4 files changed, 73 insertions(+), 14 deletions(-)
A belated:
And thank you a lot for doing these QoL fixes!
Ingo
Quoted 33 lines · Ingo Molnar
Quoted 25 lines · Ian Rogers
Ingo reported [1] that
perf stat --nullwas segfaulting. Fix the underlying issue and add a test to the "perf stat tests". Do some related fixing/cleanup in the perf util cpumap code.Thomas reported an issue fixed by the same patches [2] but caused by giving perf stat an offline CPU. Add test coverage for that and improve the "error" message that reports "success".
Ingo further pointed at broken signal handling in repeat mode [3]. I observed we weren't giving the best exit code, 0 rather than the expected 128+<signal number>. Add a patch fixing this.
[1] https://lore.kernel.org/linux-perf-users/[email protected]/
[2] https://lore.kernel.org/linux-perf-users/[email protected]/
[3] https://lore.kernel.org/lkml/[email protected]/Ian Rogers (7):
perf stat: Allow no events to open if this is a "--null" run libperf cpumap: Fixperf_cpu_map__maxfor an empty/NULL map perf cpumap: Add "any" CPU handling tocpu_map__snprint_maskperf tests stat: Add "--null" coverage
perf stat: When no events, don't report an error if there is none perf tests stat: Add test for error for an offline CPU perf stat: Improve handling of termination by signalQuoted 5 lines
tools/lib/perf/cpumap.c | 10 +++++---- tools/perf/builtin-stat.c | 29 ++++++++++++++++++------- tools/perf/tests/shell/stat.sh | 39 ++++++++++++++++++++++++++++++++++ tools/perf/util/cpumap.c | 9 ++++++-- 4 files changed, 73 insertions(+), 14 deletions(-)
Quoted 5 lines
A belated:
Tested-by Ingo Molnar <[email protected]>And thank you a lot for doing these QoL fixes!
There's one more perf stat QoL bug I'd like to report - I frequently do repeated runs of perf stat --repeat and grep the output, to get a feel for the run-to-run stability of a particular benchmark:
starship:~/tip> while :; do perf stat --null --repeat 3 sleep 0.1 2>&1 | grep elapsed; done
0.1017997 +- 0.0000771 seconds time elapsed ( +- 0.08% )
0.1017627 +- 0.0000795 seconds time elapsed ( +- 0.08% )
0.1018106 +- 0.0000650 seconds time elapsed ( +- 0.06% )
0.1017844 +- 0.0000601 seconds time elapsed ( +- 0.06% )
0.101883 +- 0.000169 seconds time elapsed ( +- 0.17% ) <====
0.1017757 +- 0.0000532 seconds time elapsed ( +- 0.05% )
0.1017991 +- 0.0000720 seconds time elapsed ( +- 0.07% )
0.1018024 +- 0.0000704 seconds time elapsed ( +- 0.07% )
0.1018074 +- 0.0000946 seconds time elapsed ( +- 0.09% )
0.1019797 +- 0.0000524 seconds time elapsed ( +- 0.05% )
0.1018407 +- 0.0000658 seconds time elapsed ( +- 0.06% )
0.1017907 +- 0.0000605 seconds time elapsed ( +- 0.06% )
0.1018328 +- 0.0000868 seconds time elapsed ( +- 0.09% )
0.1017469 +- 0.0000285 seconds time elapsed ( +- 0.03% )
0.1019589 +- 0.0000549 seconds time elapsed ( +- 0.05% )
0.1018465 +- 0.0000891 seconds time elapsed ( +- 0.09% )
0.101868 +- 0.000117 seconds time elapsed ( +- 0.12% ) <====
0.1017705 +- 0.0000590 seconds time elapsed ( +- 0.06% )
0.1017728 +- 0.0000718 seconds time elapsed ( +- 0.07% )
0.1017821 +- 0.0000419 seconds time elapsed ( +- 0.04% )
0.1018328 +- 0.0000581 seconds time elapsed ( +- 0.06% )
0.1017836 +- 0.0000853 seconds time elapsed ( +- 0.08% )
0.1018124 +- 0.0000765 seconds time elapsed ( +- 0.08% )
0.1018706 +- 0.0000639 seconds time elapsed ( +- 0.06% )Note the two outliers, which happen due to some misguided output optimization feature in perf shortening zero-ended numbers unnecessarily, and adding noise to the grepped output's vertical alignment.
Those two lines should be:
0.1017844 +- 0.0000601 seconds time elapsed ( +- 0.06% )
0.1018830 +- 0.0001690 seconds time elapsed ( +- 0.17% ) <====
0.1017757 +- 0.0000532 seconds time elapsed ( +- 0.05% ) 0.1018465 +- 0.0000891 seconds time elapsed ( +- 0.09% )
0.1018680 +- 0.0001170 seconds time elapsed ( +- 0.12% ) <====
0.1017705 +- 0.0000590 seconds time elapsed ( +- 0.06% )(The zeroes are printed fully, to full precision.)
Basically random chance causing an apparent lack of significant numbers doesn't mean the tool should strip them from the output.
Thanks,
Ingo
Ian Rogers
reviewerQuoted 83 lines · Ingo Molnar
Quoted 33 lines · Ingo Molnar
Quoted 25 lines · Ian Rogers
Ingo reported [1] that
perf stat --nullwas segfaulting. Fix the underlying issue and add a test to the "perf stat tests". Do some related fixing/cleanup in the perf util cpumap code.Thomas reported an issue fixed by the same patches [2] but caused by giving perf stat an offline CPU. Add test coverage for that and improve the "error" message that reports "success".
Ingo further pointed at broken signal handling in repeat mode [3]. I observed we weren't giving the best exit code, 0 rather than the expected 128+<signal number>. Add a patch fixing this.
[1] https://lore.kernel.org/linux-perf-users/[email protected]/
[2] https://lore.kernel.org/linux-perf-users/[email protected]/
[3] https://lore.kernel.org/lkml/[email protected]/Ian Rogers (7):
perf stat: Allow no events to open if this is a "--null" run libperf cpumap: Fixperf_cpu_map__maxfor an empty/NULL map perf cpumap: Add "any" CPU handling tocpu_map__snprint_maskperf tests stat: Add "--null" coverage
perf stat: When no events, don't report an error if there is none perf tests stat: Add test for error for an offline CPU perf stat: Improve handling of termination by signalQuoted 5 lines
tools/lib/perf/cpumap.c | 10 +++++---- tools/perf/builtin-stat.c | 29 ++++++++++++++++++------- tools/perf/tests/shell/stat.sh | 39 ++++++++++++++++++++++++++++++++++ tools/perf/util/cpumap.c | 9 ++++++-- 4 files changed, 73 insertions(+), 14 deletions(-)Quoted 5 lines
A belated:
Tested-by Ingo Molnar <[email protected]>And thank you a lot for doing these QoL fixes!
There's one more perf stat QoL bug I'd like to report - I frequently do repeated runs of perf stat --repeat and grep the output, to get a feel for the run-to-run stability of a particular benchmark:
starship:~/tip> while :; do perf stat --null --repeat 3 sleep 0.1 2>&1 | grep elapsed; done 0.1017997 +- 0.0000771 seconds time elapsed ( +- 0.08% ) 0.1017627 +- 0.0000795 seconds time elapsed ( +- 0.08% ) 0.1018106 +- 0.0000650 seconds time elapsed ( +- 0.06% ) 0.1017844 +- 0.0000601 seconds time elapsed ( +- 0.06% ) 0.101883 +- 0.000169 seconds time elapsed ( +- 0.17% ) <==== 0.1017757 +- 0.0000532 seconds time elapsed ( +- 0.05% ) 0.1017991 +- 0.0000720 seconds time elapsed ( +- 0.07% ) 0.1018024 +- 0.0000704 seconds time elapsed ( +- 0.07% ) 0.1018074 +- 0.0000946 seconds time elapsed ( +- 0.09% ) 0.1019797 +- 0.0000524 seconds time elapsed ( +- 0.05% ) 0.1018407 +- 0.0000658 seconds time elapsed ( +- 0.06% ) 0.1017907 +- 0.0000605 seconds time elapsed ( +- 0.06% ) 0.1018328 +- 0.0000868 seconds time elapsed ( +- 0.09% ) 0.1017469 +- 0.0000285 seconds time elapsed ( +- 0.03% ) 0.1019589 +- 0.0000549 seconds time elapsed ( +- 0.05% ) 0.1018465 +- 0.0000891 seconds time elapsed ( +- 0.09% ) 0.101868 +- 0.000117 seconds time elapsed ( +- 0.12% ) <==== 0.1017705 +- 0.0000590 seconds time elapsed ( +- 0.06% ) 0.1017728 +- 0.0000718 seconds time elapsed ( +- 0.07% ) 0.1017821 +- 0.0000419 seconds time elapsed ( +- 0.04% ) 0.1018328 +- 0.0000581 seconds time elapsed ( +- 0.06% ) 0.1017836 +- 0.0000853 seconds time elapsed ( +- 0.08% ) 0.1018124 +- 0.0000765 seconds time elapsed ( +- 0.08% ) 0.1018706 +- 0.0000639 seconds time elapsed ( +- 0.06% )Note the two outliers, which happen due to some misguided output optimization feature in perf shortening zero-ended numbers unnecessarily, and adding noise to the grepped output's vertical alignment.
Those two lines should be:
0.1017844 +- 0.0000601 seconds time elapsed ( +- 0.06% ) 0.1018830 +- 0.0001690 seconds time elapsed ( +- 0.17% ) <==== 0.1017757 +- 0.0000532 seconds time elapsed ( +- 0.05% )
Quoted 8 lines
0.1018465 +- 0.0000891 seconds time elapsed ( +- 0.09% ) 0.1018680 +- 0.0001170 seconds time elapsed ( +- 0.12% ) <==== 0.1017705 +- 0.0000590 seconds time elapsed ( +- 0.06% )(The zeroes are printed fully, to full precision.)
Basically random chance causing an apparent lack of significant numbers doesn't mean the tool should strip them from the output.
Ugh. I'm not sure why the output is like that. Searching back through history it seems you are to blame :-)
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/commit/tools/perf/builtin-stat.c?h=perf-tools-next&id=bc22de9bcdb2249150fb5b3c48fdc4f6bedd3ad7
Perhaps we can drop the precision printf flag and just make the flags fixed. I think this output is obscure enough that nobody cares about minor changes.
Thanks,
Ian
Quoted 3 lines
Thanks,
Ingo
Quoted 45 lines · Ian Rogers
Quoted 45 lines
There's one more perf stat QoL bug I'd like to report - I frequently do repeated runs of perf stat --repeat and grep the output, to get a feel for the run-to-run stability of a particular benchmark:
starship:~/tip> while :; do perf stat --null --repeat 3 sleep 0.1 2>&1 | grep elapsed; done 0.1017997 +- 0.0000771 seconds time elapsed ( +- 0.08% ) 0.1017627 +- 0.0000795 seconds time elapsed ( +- 0.08% ) 0.1018106 +- 0.0000650 seconds time elapsed ( +- 0.06% ) 0.1017844 +- 0.0000601 seconds time elapsed ( +- 0.06% ) 0.101883 +- 0.000169 seconds time elapsed ( +- 0.17% ) <==== 0.1017757 +- 0.0000532 seconds time elapsed ( +- 0.05% ) 0.1017991 +- 0.0000720 seconds time elapsed ( +- 0.07% ) 0.1018024 +- 0.0000704 seconds time elapsed ( +- 0.07% ) 0.1018074 +- 0.0000946 seconds time elapsed ( +- 0.09% ) 0.1019797 +- 0.0000524 seconds time elapsed ( +- 0.05% ) 0.1018407 +- 0.0000658 seconds time elapsed ( +- 0.06% ) 0.1017907 +- 0.0000605 seconds time elapsed ( +- 0.06% ) 0.1018328 +- 0.0000868 seconds time elapsed ( +- 0.09% ) 0.1017469 +- 0.0000285 seconds time elapsed ( +- 0.03% ) 0.1019589 +- 0.0000549 seconds time elapsed ( +- 0.05% ) 0.1018465 +- 0.0000891 seconds time elapsed ( +- 0.09% ) 0.101868 +- 0.000117 seconds time elapsed ( +- 0.12% ) <==== 0.1017705 +- 0.0000590 seconds time elapsed ( +- 0.06% ) 0.1017728 +- 0.0000718 seconds time elapsed ( +- 0.07% ) 0.1017821 +- 0.0000419 seconds time elapsed ( +- 0.04% ) 0.1018328 +- 0.0000581 seconds time elapsed ( +- 0.06% ) 0.1017836 +- 0.0000853 seconds time elapsed ( +- 0.08% ) 0.1018124 +- 0.0000765 seconds time elapsed ( +- 0.08% ) 0.1018706 +- 0.0000639 seconds time elapsed ( +- 0.06% )Note the two outliers, which happen due to some misguided output optimization feature in perf shortening zero-ended numbers unnecessarily, and adding noise to the grepped output's vertical alignment.
Those two lines should be:
0.1017844 +- 0.0000601 seconds time elapsed ( +- 0.06% ) 0.1018830 +- 0.0001690 seconds time elapsed ( +- 0.17% ) <==== 0.1017757 +- 0.0000532 seconds time elapsed ( +- 0.05% )0.1018465 +- 0.0000891 seconds time elapsed ( +- 0.09% ) 0.1018680 +- 0.0001170 seconds time elapsed ( +- 0.12% ) <==== 0.1017705 +- 0.0000590 seconds time elapsed ( +- 0.06% )
Quoted 6 lines
Quoted 4 lines
(The zeroes are printed fully, to full precision.)
Basically random chance causing an apparent lack of significant numbers doesn't mean the tool should strip them from the output.
Ugh. I'm not sure why the output is like that.
Yeah, I'm sure some dim-witted kernel developer requested it, without thinking through all the consequences.
Quoted 2 lines
[...] Searching back through history it seems you are to blame :-)
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/commit/tools/perf/builtin-stat.c?h=perf-tools-next&id=bc22de9bcdb2249150fb5b3c48fdc4f6bedd3ad7
Uhm, never mind, what a brilliant feature! ;-)
Quoted 3 lines
Perhaps we can drop the precision printf flag and just make the flags fixed. I think this output is obscure enough that nobody cares about minor changes.
Yeah, sounds good.
Thanks,
Ingo
Ian Rogers
reviewerQuoted 62 lines · Ingo Molnar
Quoted 45 lines · Ian Rogers
Quoted 45 lines
There's one more perf stat QoL bug I'd like to report - I frequently do repeated runs of perf stat --repeat and grep the output, to get a feel for the run-to-run stability of a particular benchmark:
starship:~/tip> while :; do perf stat --null --repeat 3 sleep 0.1 2>&1 | grep elapsed; done 0.1017997 +- 0.0000771 seconds time elapsed ( +- 0.08% ) 0.1017627 +- 0.0000795 seconds time elapsed ( +- 0.08% ) 0.1018106 +- 0.0000650 seconds time elapsed ( +- 0.06% ) 0.1017844 +- 0.0000601 seconds time elapsed ( +- 0.06% ) 0.101883 +- 0.000169 seconds time elapsed ( +- 0.17% ) <==== 0.1017757 +- 0.0000532 seconds time elapsed ( +- 0.05% ) 0.1017991 +- 0.0000720 seconds time elapsed ( +- 0.07% ) 0.1018024 +- 0.0000704 seconds time elapsed ( +- 0.07% ) 0.1018074 +- 0.0000946 seconds time elapsed ( +- 0.09% ) 0.1019797 +- 0.0000524 seconds time elapsed ( +- 0.05% ) 0.1018407 +- 0.0000658 seconds time elapsed ( +- 0.06% ) 0.1017907 +- 0.0000605 seconds time elapsed ( +- 0.06% ) 0.1018328 +- 0.0000868 seconds time elapsed ( +- 0.09% ) 0.1017469 +- 0.0000285 seconds time elapsed ( +- 0.03% ) 0.1019589 +- 0.0000549 seconds time elapsed ( +- 0.05% ) 0.1018465 +- 0.0000891 seconds time elapsed ( +- 0.09% ) 0.101868 +- 0.000117 seconds time elapsed ( +- 0.12% ) <==== 0.1017705 +- 0.0000590 seconds time elapsed ( +- 0.06% ) 0.1017728 +- 0.0000718 seconds time elapsed ( +- 0.07% ) 0.1017821 +- 0.0000419 seconds time elapsed ( +- 0.04% ) 0.1018328 +- 0.0000581 seconds time elapsed ( +- 0.06% ) 0.1017836 +- 0.0000853 seconds time elapsed ( +- 0.08% ) 0.1018124 +- 0.0000765 seconds time elapsed ( +- 0.08% ) 0.1018706 +- 0.0000639 seconds time elapsed ( +- 0.06% )Note the two outliers, which happen due to some misguided output optimization feature in perf shortening zero-ended numbers unnecessarily, and adding noise to the grepped output's vertical alignment.
Those two lines should be:
0.1017844 +- 0.0000601 seconds time elapsed ( +- 0.06% ) 0.1018830 +- 0.0001690 seconds time elapsed ( +- 0.17% ) <==== 0.1017757 +- 0.0000532 seconds time elapsed ( +- 0.05% )0.1018465 +- 0.0000891 seconds time elapsed ( +- 0.09% ) 0.1018680 +- 0.0001170 seconds time elapsed ( +- 0.12% ) <==== 0.1017705 +- 0.0000590 seconds time elapsed ( +- 0.06% )Quoted 6 lines
Quoted 4 lines
(The zeroes are printed fully, to full precision.)
Basically random chance causing an apparent lack of significant numbers doesn't mean the tool should strip them from the output.
Ugh. I'm not sure why the output is like that.
Yeah, I'm sure some dim-witted kernel developer requested it, without thinking through all the consequences.
Quoted 2 lines
[...] Searching back through history it seems you are to blame :-)
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/commit/tools/perf/builtin-stat.c?h=perf-tools-next&id=bc22de9bcdb2249150fb5b3c48fdc4f6bedd3ad7
Quoted 7 lines
Uhm, never mind, what a brilliant feature! ;-)
Quoted 3 lines
Perhaps we can drop the precision printf flag and just make the flags fixed. I think this output is obscure enough that nobody cares about minor changes.
Yeah, sounds good.
Cool. I sent:
https://lore.kernel.org/lkml/[email protected]/
Thanks,
Ian
Quoted 4 lines
Thanks,
Ingo