diff options
| author | Arnaldo Carvalho de Melo <acme@redhat.com> | 2010-12-25 17:12:25 +0300 |
|---|---|---|
| committer | Arnaldo Carvalho de Melo <acme@redhat.com> | 2010-12-25 17:12:25 +0300 |
| commit | a43d3f08c64125edbdfdc3d3aa512d3e37321f37 (patch) | |
| tree | 3d8a2e82045f13c0b1e1fe67601c632002882e93 /tools | |
| parent | 765532c8aaac624b5f8687af6d319c6a1138a257 (diff) | |
| download | linux-a43d3f08c64125edbdfdc3d3aa512d3e37321f37.tar.xz | |
perf record: Fix use of sample_id_all userspace with !sample_id_all kernels
Check if parse_single_tracepoint_event has already asked for PERF_SAMPLE_TIME.
This is kludgy but short term fix for problems introduced by eac23d1c that
broke 'perf script' by having different sample_types when using multiple
tracepoint events when we use a perf binary that tries to use sample_id_all on
an older kernel.
We need to move counter creation to perf_session, support different
sample_types, etc.
Ongoing work on the perf test infrastructure needs this so that we can create
counters to monitor threads generating specific events, etc.
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Tom Zanussi <tzanussi@gmail.com>
Cc: Torok Edwin <edwintorok@gmail.com>
Cc: Ian Munsie <imunsie@au1.ibm.com>
LKML-Reference: <new-submission>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Diffstat (limited to 'tools')
| -rw-r--r-- | tools/perf/builtin-record.c | 24 |
1 files changed, 19 insertions, 5 deletions
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 5149e3deb7bc..50efbd509b8f 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -243,6 +243,19 @@ static void create_counter(int counter, int cpu) u64 time_running; u64 id; } read_data; + /* + * Check if parse_single_tracepoint_event has already asked for + * PERF_SAMPLE_TIME. + * + * XXX this is kludgy but short term fix for problems introduced by + * eac23d1c that broke 'perf script' by having different sample_types + * when using multiple tracepoint events when we use a perf binary + * that tries to use sample_id_all on an older kernel. + * + * We need to move counter creation to perf_session, support + * different sample_types, etc. + */ + bool time_needed = attr->sample_type & PERF_SAMPLE_TIME; attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | PERF_FORMAT_TOTAL_TIME_RUNNING | @@ -285,7 +298,8 @@ static void create_counter(int counter, int cpu) if (system_wide) attr->sample_type |= PERF_SAMPLE_CPU; - if (sample_time || system_wide || !no_inherit || cpu_list) + if (sample_id_all_avail && + (sample_time || system_wide || !no_inherit || cpu_list)) attr->sample_type |= PERF_SAMPLE_TIME; if (raw_samples) { @@ -294,9 +308,6 @@ static void create_counter(int counter, int cpu) attr->sample_type |= PERF_SAMPLE_CPU; } - if (!sample_type) - sample_type = attr->sample_type; - attr->mmap = track; attr->comm = track; attr->inherit = !no_inherit; @@ -327,7 +338,7 @@ try_again: * Old kernel, no attr->sample_id_type_all field */ sample_id_all_avail = false; - if (!sample_time && !raw_samples) + if (!sample_time && !raw_samples && !time_needed) attr->sample_type &= ~PERF_SAMPLE_TIME; goto retry_sample_id; @@ -428,6 +439,9 @@ try_again: } } } + + if (!sample_type) + sample_type = attr->sample_type; } static void open_counters(int cpu) |