| Message ID | 20200416221457.46710-1-irogers@google.com |
|---|---|
| Headers | show |
| Series | perf tools: add support for libpfm4 | expand |
On Thu, Apr 16, 2020 at 03:14:53PM -0700, Ian Rogers wrote: > This patch links perf with the libpfm4 library if it is available > and NO_LIBPFM4 isn't passed to the build. The libpfm4 library > contains hardware event tables for all processors supported by > perf_events. It is a helper library that helps convert from a > symbolic event name to the event encoding required by the > underlying kernel interface. This library is open-source and > available from: http://perfmon2.sf.net. > > With this patch, it is possible to specify full hardware events > by name. Hardware filters are also supported. Events must be > specified via the --pfm-events and not -e option. Both options > are active at the same time and it is possible to mix and match: > > $ perf stat --pfm-events inst_retired:any_p:c=1:i -e cycles .... > > v11 reformats the perf list output to be: > List of pre-defined events (to be used in -e): > > branch-instructions OR branches [Hardware event] > branch-misses [Hardware event] > ... > > List of pre-defined events (to be used in --pfm-events): > > ix86arch: > UNHALTED_CORE_CYCLES > [count core clock cycles whenever the clock signal on the specific core is running (not halted)] > INSTRUCTION_RETIRED > [count the number of instructions at retirement. For instructions that consists of multiple mic> > ... > skx: > UNHALTED_CORE_CYCLES > [Count core clock cycles whenever the clock signal on the specific core is running (not halted)] > ... > BACLEARS > [Branch re-steered] > BACLEARS:ANY > [Number of front-end re-steers due to BPU misprediction] > ... Acked-by: Jiri Olsa <jolsa@redhat.com> thanks, jirka
This patch links perf with the libpfm4 library if it is available and NO_LIBPFM4 isn't passed to the build. The libpfm4 library contains hardware event tables for all processors supported by perf_events. It is a helper library that helps convert from a symbolic event name to the event encoding required by the underlying kernel interface. This library is open-source and available from: http://perfmon2.sf.net. With this patch, it is possible to specify full hardware events by name. Hardware filters are also supported. Events must be specified via the --pfm-events and not -e option. Both options are active at the same time and it is possible to mix and match: $ perf stat --pfm-events inst_retired:any_p:c=1:i -e cycles .... v11 reformats the perf list output to be: List of pre-defined events (to be used in -e): branch-instructions OR branches [Hardware event] branch-misses [Hardware event] ... List of pre-defined events (to be used in --pfm-events): ix86arch: UNHALTED_CORE_CYCLES [count core clock cycles whenever the clock signal on the specific core is running (not halted)] INSTRUCTION_RETIRED [count the number of instructions at retirement. For instructions that consists of multiple mic> ... skx: UNHALTED_CORE_CYCLES [Count core clock cycles whenever the clock signal on the specific core is running (not halted)] ... BACLEARS [Branch re-steered] BACLEARS:ANY [Number of front-end re-steers due to BPU misprediction] ... v10 addresses review comments from jolsa@redhat.com. v9 removes some unnecessary #ifs. v8 addresses review comments from jolsa@redhat.com. Breaks the patch into 4, adds a test and moves the libpfm code into its own file. perf list encoding tries to be closer to existing. v7 rebases and adds fallback code for libpfm4 events. The fallback code is to force user only priv level in case the perf_event_open() syscall failed for permissions reason. the fallback forces a user privilege level restriction on the event string, so depending on the syntax either u or :u is needed. But libpfm4 can use a : or . as the separator, so simply searching for ':' vs. '/' is not good enough to determine the syntax needed. Therefore, this patch introduces a new evsel boolean field to mark events coming from libpfm4. The field is then used to adjust the fallback string. v6 was a rebase. v5 was a rebase. v4 was a rebase on git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git branch perf/core and re-adds the tools/build/feature/test-libpfm4.c missed in v3. v3 is against acme/perf/core and removes a diagnostic warning. v2 of this patch makes the --pfm-events man page documentation conditional on libpfm4 behing configured. It tidies some of the documentation and adds the feature test missed in the v1 patch. Ian Rogers (1): perf doc: allow ASCIIDOC_EXTRA to be an argument Stephane Eranian (3): tools feature: add support for detecting libpfm4 perf pmu: add perf_pmu__find_by_type helper perf tools: add support for libpfm4 tools/build/Makefile.feature | 3 +- tools/build/feature/Makefile | 6 +- tools/build/feature/test-libpfm4.c | 9 + tools/perf/Documentation/Makefile | 4 +- tools/perf/Documentation/perf-record.txt | 11 + tools/perf/Documentation/perf-stat.txt | 10 + tools/perf/Documentation/perf-top.txt | 11 + tools/perf/Makefile.config | 13 ++ tools/perf/Makefile.perf | 6 +- tools/perf/builtin-list.c | 3 + tools/perf/builtin-record.c | 8 + tools/perf/builtin-stat.c | 8 + tools/perf/builtin-top.c | 8 + tools/perf/tests/Build | 1 + tools/perf/tests/builtin-test.c | 9 + tools/perf/tests/pfm.c | 207 +++++++++++++++++ tools/perf/tests/tests.h | 3 + tools/perf/util/Build | 2 + tools/perf/util/evsel.c | 2 +- tools/perf/util/evsel.h | 1 + tools/perf/util/parse-events.c | 30 ++- tools/perf/util/parse-events.h | 4 + tools/perf/util/pfm.c | 278 +++++++++++++++++++++++ tools/perf/util/pfm.h | 43 ++++ tools/perf/util/pmu.c | 11 + tools/perf/util/pmu.h | 1 + 26 files changed, 678 insertions(+), 14 deletions(-) create mode 100644 tools/build/feature/test-libpfm4.c create mode 100644 tools/perf/tests/pfm.c create mode 100644 tools/perf/util/pfm.c create mode 100644 tools/perf/util/pfm.h