Messages in this thread | | | Date | Mon, 16 May 2022 10:22:00 -0300 | From | Arnaldo Carvalho de Melo <> | Subject | Re: [PATCH v2 5/8] perf record ibs: Warn about sampling period skew |
| |
Em Mon, May 09, 2022 at 10:19:11AM +0530, Ravi Bangoria escreveu: > Samples without an L3 miss are discarded and counter is reset with > random value (between 1-15 for fetch pmu and 1-127 for op pmu) when > IBS L3 miss filtering is enabled. This causes a sampling period skew > but there is no way to reconstruct aggregated sampling period. So > print a warning at perf record if user sets l3missonly=1. > > Ex: > # perf record -c 10000 -C 0 -e ibs_op/l3missonly=1/ > WARNING: Hw internally resets sampling period when L3 Miss Filtering is enabled > and tagged operation does not cause L3 Miss. This causes sampling period skew. > > Signed-off-by: Ravi Bangoria <ravi.bangoria@amd.com> > --- > tools/perf/arch/x86/util/evsel.c | 34 ++++++++++++++++++++++++++++++++ > tools/perf/util/evsel.c | 7 +++++++ > tools/perf/util/evsel.h | 1 + > 3 files changed, 42 insertions(+) > > diff --git a/tools/perf/arch/x86/util/evsel.c b/tools/perf/arch/x86/util/evsel.c > index ac2899a25b7a..6399faa70a88 100644 > --- a/tools/perf/arch/x86/util/evsel.c > +++ b/tools/perf/arch/x86/util/evsel.c > @@ -4,6 +4,8 @@ > #include "util/evsel.h" > #include "util/env.h" > #include "linux/string.h" > +#include "util/pmu.h" > +#include "util/debug.h" > > void arch_evsel__set_sample_weight(struct evsel *evsel) > { > @@ -29,3 +31,35 @@ void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr) > > free(env.cpuid); > } > + > +static void ibs_l3miss_warn(void) > +{ > + pr_warning( > +"WARNING: Hw internally resets sampling period when L3 Miss Filtering is enabled\n" > +"and tagged operation does not cause L3 Miss. This causes sampling period skew.\n"); > +} > + > +void arch_evsel__warn_ambiguity(struct evsel *evsel, struct perf_event_attr *attr) > +{ > + struct perf_env *env = evsel__env(evsel); > + struct perf_pmu *evsel_pmu = evsel__find_pmu(evsel); > + struct perf_pmu *ibs_fetch_pmu = perf_pmu__find("ibs_fetch"); > + struct perf_pmu *ibs_op_pmu = perf_pmu__find("ibs_op"); > + static int warned_once;
Please check first if the warning was emitted (warned_once is true) before calling all the find routines above.
> + if (warned_once || !perf_env__cpuid(env) || !env->cpuid || > + !strstarts(env->cpuid, "AuthenticAMD") || !evsel_pmu) > + return; > + > + if (ibs_fetch_pmu && ibs_fetch_pmu->type == evsel_pmu->type) { > + if (attr->config & (1ULL << 59)) { > + ibs_l3miss_warn(); > + warned_once = 1; > + } > + } else if (ibs_op_pmu && ibs_op_pmu->type == evsel_pmu->type) { > + if (attr->config & (1ULL << 16)) { > + ibs_l3miss_warn(); > + warned_once = 1; > + } > + } > +} > diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c > index 2a1729e7aee4..4f8b72d4a521 100644 > --- a/tools/perf/util/evsel.c > +++ b/tools/perf/util/evsel.c > @@ -1064,6 +1064,11 @@ void __weak arch_evsel__fixup_new_cycles(struct perf_event_attr *attr __maybe_un > { > } > > +void __weak arch_evsel__warn_ambiguity(struct evsel *evsel __maybe_unused, > + struct perf_event_attr *attr __maybe_unused) > +{ > +} > + > static void evsel__set_default_freq_period(struct record_opts *opts, > struct perf_event_attr *attr) > { > @@ -1339,6 +1344,8 @@ void evsel__config(struct evsel *evsel, struct record_opts *opts, > */ > if (evsel__is_dummy_event(evsel)) > evsel__reset_sample_bit(evsel, BRANCH_STACK); > + > + arch_evsel__warn_ambiguity(evsel, attr);
Wouldn't this be better as a single arch__post_evsel_config() function that could do arch specific fixups or emit such warnings _after_ (thus the "post") the common code evsel__config() does its thing?
> } > > int evsel__set_filter(struct evsel *evsel, const char *filter) > diff --git a/tools/perf/util/evsel.h b/tools/perf/util/evsel.h > index 041b42d33bf5..195ae30ec45b 100644 > --- a/tools/perf/util/evsel.h > +++ b/tools/perf/util/evsel.h > @@ -281,6 +281,7 @@ void evsel__set_sample_id(struct evsel *evsel, bool use_sample_identifier); > > void arch_evsel__set_sample_weight(struct evsel *evsel); > void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr); > +void arch_evsel__warn_ambiguity(struct evsel *evsel, struct perf_event_attr *attr); > > int evsel__set_filter(struct evsel *evsel, const char *filter); > int evsel__append_tp_filter(struct evsel *evsel, const char *filter); > -- > 2.27.0
--
- Arnaldo
| |