| Message ID | 1473978296-20712-17-git-send-email-sukadev@linux.vnet.ibm.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Headers | show |
Em Thu, Sep 15, 2016 at 03:24:53PM -0700, Sukadev Bhattiprolu escreveu: > From: Andi Kleen <ak@linux.intel.com> > > Make alias matching the events parser case-insensitive. This is useful > with the JSON events. perf uses lower case events, but the CPU manuals > generally use upper case event names. The JSON files use lower > case by default too. But if we search case insensitively then > users can cut-n-paste the upper case event names. > > So the following works: > > % perf stat -e BR_INST_EXEC.TAKEN_INDIRECT_NEAR_CALL true > > Performance counter stats for 'true': > > 305 BR_INST_EXEC.TAKEN_INDIRECT_NEAR_CALL > > 0.000492799 seconds time elapsed So now trying to figure this out: [acme@jouet linux]$ perf stat -e br_inst_exec.all_direct_near_call true event syntax error: 'br_inst_exec.all_direct_near_call' \___ 'period' is not usable in 'perf stat' Run 'perf list' for a list of valid events Usage: perf stat [<options>] [<command>] -e, --event <event> event selector. use 'perf list' to list available events [acme@jouet linux]$ [acme@jouet linux]$ perf stat -e cycles true Performance counter stats for 'true': 228,093 cycles:u 0.002678362 seconds time elapsed [acme@jouet linux]$ Who is setting period in that case? For cycles we get in verbose mode: [acme@jouet linux]$ perf stat -vv -e cycles true Using CPUID GenuineIntel-6-3D intel_pt default config: tsc ------------------------------------------------------------ perf_event_attr: size 112 sample_type IDENTIFIER read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING disabled 1 inherit 1 enable_on_exec 1 exclude_guest 1 ------------------------------------------------------------ sys_perf_event_open: pid 13346 cpu -1 group_fd -1 flags 0x8 sys_perf_event_open failed, error -13 Warning: kernel.perf_event_paranoid=2, trying to fall back to excluding kernel samples ------------------------------------------------------------ perf_event_attr: size 112 sample_type IDENTIFIER read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING disabled 1 inherit 1 exclude_kernel 1 enable_on_exec 1 exclude_guest 1 ------------------------------------------------------------ sys_perf_event_open: pid 13346 cpu -1 group_fd -1 flags 0x8 cycles:u: 0: 322539 536424 536424 cycles:u: 322539 536424 536424 Performance counter stats for 'true': 322,539 cycles:u 0.001061349 seconds time elapsed And for br_inst_exec.all_direct_near_call: Strange: [acme@jouet linux]$ perf stat -vvv -e 'br_inst_exec.all_direct_near_call' true Using CPUID GenuineIntel-6-3D event syntax error: 'br_inst_exec.all_direct_near_call' \___ 'period' is not usable in 'perf stat' Run 'perf list' for a list of valid events Usage: perf stat [<options>] [<command>] -e, --event <event> event selector. use 'perf list' to list available events [acme@jouet linux]$ [acme@jouet linux]$ perf record -e 'br_inst_exec.all_direct_near_call' true [ perf record: Woken up 1 times to write data ] [ perf record: Captured and wrote 0.020 MB perf.data ] [acme@jouet linux]$ perf evlist -v br_inst_exec.all_direct_near_call:u: type: 4, size: 112, config: 0xd088, { sample_period, sample_freq }: 200003, sample_type: IP|TID|TIME|PERIOD, disabled: 1, inherit: 1, exclude_kernel: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1 [acme@jouet linux]$ I've put what I have on a tmp.perf/json branch, will run it thru my tests and continue tomorrow, when I'll try to investigate this problem with 'perf stat' and these events, unless someone does it while I'm asleep. :-) - Arnaldo > Signed-off-by: Andi Kleen <ak@linux.intel.com> > Acked-by: Ingo Molnar <mingo@kernel.org> > --- > tools/perf/util/parse-events.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c > index 9abd60d..1abda10 100644 > --- a/tools/perf/util/parse-events.c > +++ b/tools/perf/util/parse-events.c > @@ -1453,7 +1453,7 @@ comp_pmu(const void *p1, const void *p2) > struct perf_pmu_event_symbol *pmu1 = (struct perf_pmu_event_symbol *) p1; > struct perf_pmu_event_symbol *pmu2 = (struct perf_pmu_event_symbol *) p2; > > - return strcmp(pmu1->symbol, pmu2->symbol); > + return strcasecmp(pmu1->symbol, pmu2->symbol); > } > > static void perf_pmu__parse_cleanup(void) > -- > 1.8.3.1
Em Mon, Oct 03, 2016 at 09:47:06PM -0300, Arnaldo Carvalho de Melo escreveu: > Em Thu, Sep 15, 2016 at 03:24:53PM -0700, Sukadev Bhattiprolu escreveu: > > From: Andi Kleen <ak@linux.intel.com> > > > > Make alias matching the events parser case-insensitive. This is useful > > with the JSON events. perf uses lower case events, but the CPU manuals > > generally use upper case event names. The JSON files use lower > > case by default too. But if we search case insensitively then > > users can cut-n-paste the upper case event names. > > > > So the following works: > > > > % perf stat -e BR_INST_EXEC.TAKEN_INDIRECT_NEAR_CALL true > > > > Performance counter stats for 'true': > > > > 305 BR_INST_EXEC.TAKEN_INDIRECT_NEAR_CALL > > > > 0.000492799 seconds time elapsed > > So now trying to figure this out: Ok, so this is just another case of bad patch ordering: this uses 'perf stat' as the example, but it doesn't work at this point, one needs to go to the last patch, apply it and then test it again, when it will work. So just moved the last patch to before this one, sigh. - Arnaldo > [acme@jouet linux]$ perf stat -e br_inst_exec.all_direct_near_call true > event syntax error: 'br_inst_exec.all_direct_near_call' > \___ 'period' is not usable in 'perf stat' > Run 'perf list' for a list of valid events > > Usage: perf stat [<options>] [<command>] > > -e, --event <event> event selector. use 'perf list' to list available events > [acme@jouet linux]$ > > [acme@jouet linux]$ perf stat -e cycles true > > Performance counter stats for 'true': > > 228,093 cycles:u > > 0.002678362 seconds time elapsed > > [acme@jouet linux]$ > > > Who is setting period in that case? > > For cycles we get in verbose mode: > > [acme@jouet linux]$ perf stat -vv -e cycles true > Using CPUID GenuineIntel-6-3D > intel_pt default config: tsc > ------------------------------------------------------------ > perf_event_attr: > size 112 > sample_type IDENTIFIER > read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING > disabled 1 > inherit 1 > enable_on_exec 1 > exclude_guest 1 > ------------------------------------------------------------ > sys_perf_event_open: pid 13346 cpu -1 group_fd -1 flags 0x8 > sys_perf_event_open failed, error -13 > Warning: > kernel.perf_event_paranoid=2, trying to fall back to excluding kernel samples > ------------------------------------------------------------ > perf_event_attr: > size 112 > sample_type IDENTIFIER > read_format TOTAL_TIME_ENABLED|TOTAL_TIME_RUNNING > disabled 1 > inherit 1 > exclude_kernel 1 > enable_on_exec 1 > exclude_guest 1 > ------------------------------------------------------------ > sys_perf_event_open: pid 13346 cpu -1 group_fd -1 flags 0x8 > cycles:u: 0: 322539 536424 536424 > cycles:u: 322539 536424 536424 > > Performance counter stats for 'true': > > 322,539 cycles:u > > 0.001061349 seconds time elapsed > > And for br_inst_exec.all_direct_near_call: > > Strange: > > [acme@jouet linux]$ perf stat -vvv -e 'br_inst_exec.all_direct_near_call' true > Using CPUID GenuineIntel-6-3D > event syntax error: 'br_inst_exec.all_direct_near_call' > \___ 'period' is not usable in 'perf stat' > Run 'perf list' for a list of valid events > > Usage: perf stat [<options>] [<command>] > > -e, --event <event> event selector. use 'perf list' to list available events > [acme@jouet linux]$ > [acme@jouet linux]$ perf record -e 'br_inst_exec.all_direct_near_call' true > [ perf record: Woken up 1 times to write data ] > [ perf record: Captured and wrote 0.020 MB perf.data ] > [acme@jouet linux]$ perf evlist -v > br_inst_exec.all_direct_near_call:u: type: 4, size: 112, config: 0xd088, { sample_period, sample_freq }: 200003, sample_type: IP|TID|TIME|PERIOD, disabled: 1, inherit: 1, exclude_kernel: 1, mmap: 1, comm: 1, enable_on_exec: 1, task: 1, sample_id_all: 1, exclude_guest: 1, mmap2: 1, comm_exec: 1 > [acme@jouet linux]$ > > I've put what I have on a tmp.perf/json branch, will run it thru my tests and > continue tomorrow, when I'll try to investigate this problem with 'perf stat' and > these events, unless someone does it while I'm asleep. :-) > > - Arnaldo > > > Signed-off-by: Andi Kleen <ak@linux.intel.com> > > Acked-by: Ingo Molnar <mingo@kernel.org> > > --- > > tools/perf/util/parse-events.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c > > index 9abd60d..1abda10 100644 > > --- a/tools/perf/util/parse-events.c > > +++ b/tools/perf/util/parse-events.c > > @@ -1453,7 +1453,7 @@ comp_pmu(const void *p1, const void *p2) > > struct perf_pmu_event_symbol *pmu1 = (struct perf_pmu_event_symbol *) p1; > > struct perf_pmu_event_symbol *pmu2 = (struct perf_pmu_event_symbol *) p2; > > > > - return strcmp(pmu1->symbol, pmu2->symbol); > > + return strcasecmp(pmu1->symbol, pmu2->symbol); > > } > > > > static void perf_pmu__parse_cleanup(void) > > -- > > 1.8.3.1
On Mon, Oct 03, 2016 at 09:47:06PM -0300, Arnaldo Carvalho de Melo wrote: > Em Thu, Sep 15, 2016 at 03:24:53PM -0700, Sukadev Bhattiprolu escreveu: > > From: Andi Kleen <ak@linux.intel.com> > > > > Make alias matching the events parser case-insensitive. This is useful > > with the JSON events. perf uses lower case events, but the CPU manuals > > generally use upper case event names. The JSON files use lower > > case by default too. But if we search case insensitively then > > users can cut-n-paste the upper case event names. > > > > So the following works: > > > > % perf stat -e BR_INST_EXEC.TAKEN_INDIRECT_NEAR_CALL true > > > > Performance counter stats for 'true': > > > > 305 BR_INST_EXEC.TAKEN_INDIRECT_NEAR_CALL > > > > 0.000492799 seconds time elapsed > > So now trying to figure this out: > > [acme@jouet linux]$ perf stat -e br_inst_exec.all_direct_near_call true > event syntax error: 'br_inst_exec.all_direct_near_call' > \___ 'period' is not usable in 'perf stat' > Run 'perf list' for a list of valid events looks like you need also this one: 155128eaa62f Allow period= in perf stat CPU event descriptions. jirka
On Mon, Oct 03, 2016 at 09:54:48PM -0300, Arnaldo Carvalho de Melo wrote: > Em Mon, Oct 03, 2016 at 09:47:06PM -0300, Arnaldo Carvalho de Melo escreveu: > > Em Thu, Sep 15, 2016 at 03:24:53PM -0700, Sukadev Bhattiprolu escreveu: > > > From: Andi Kleen <ak@linux.intel.com> > > > > > > Make alias matching the events parser case-insensitive. This is useful > > > with the JSON events. perf uses lower case events, but the CPU manuals > > > generally use upper case event names. The JSON files use lower > > > case by default too. But if we search case insensitively then > > > users can cut-n-paste the upper case event names. > > > > > > So the following works: > > > > > > % perf stat -e BR_INST_EXEC.TAKEN_INDIRECT_NEAR_CALL true > > > > > > Performance counter stats for 'true': > > > > > > 305 BR_INST_EXEC.TAKEN_INDIRECT_NEAR_CALL > > > > > > 0.000492799 seconds time elapsed > > > > So now trying to figure this out: > > Ok, so this is just another case of bad patch ordering: this uses 'perf > stat' as the example, but it doesn't work at this point, one needs to go > to the last patch, apply it and then test it again, when it will work. > So just moved the last patch to before this one, sigh. ok, you found it.. sry ;-) jirka
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 9abd60d..1abda10 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -1453,7 +1453,7 @@ comp_pmu(const void *p1, const void *p2) struct perf_pmu_event_symbol *pmu1 = (struct perf_pmu_event_symbol *) p1; struct perf_pmu_event_symbol *pmu2 = (struct perf_pmu_event_symbol *) p2; - return strcmp(pmu1->symbol, pmu2->symbol); + return strcasecmp(pmu1->symbol, pmu2->symbol); } static void perf_pmu__parse_cleanup(void)