perf evsel: Set attr.exclude_kernel when probing max attr.precise_ip
authorArnaldo Carvalho de Melo <acme@redhat.com>
Mon, 3 Jul 2017 20:01:41 +0000 (17:01 -0300)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 4 Jul 2017 14:42:21 +0000 (11:42 -0300)
We should set attr.exclude_kernel when probing for attr.precise_ip
level, otherwise !CAP_SYS_ADMIN users will not default to skidless
samples in capable hardware.

The increase in the paranoid level in commit 0161028b7c8a ("perf/core:
Change the default paranoia level to 2") broke this, fix it by excluding
kernel samples when probing.

Before:

  $ perf record usleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.018 MB perf.data (6 samples) ]
  $ perf evlist -v
  cycles:u: sample_freq: 4000, sample_type: IP|TID|TIME|PERIOD, exclude_kernel: 1

After:

  $ perf record usleep 1
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.018 MB perf.data (8 samples) ]
  $ perf evlist -v
  cycles:ppp: sample_freq: 4000, sample_type: IP|TID|TIME|PERIOD, exclude_kernel: 1, precise_ip: 3
                                                                                     ^^^^^^^^^^^^^
                                                                                     ^^^^^^^^^^^^^
                                                                                     ^^^^^^^^^^^^^
  $

To further clarify: we always set .exclude_kernel when non !CAP_SYS_ADMIN
users profile, its just on the attr.precise_ip probing that we weren't doing
so, fix it.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: 7f8d1ade1b19 ("perf tools: By default use the most precise "cycles" hw counter available")
Link: http://lkml.kernel.org/n/tip-t2qttwhbnua62o5gt75cueml@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/evsel.c

index cda44b0e821c63baea1f9da52123184768fdd2f4..91192415f38d58ab03a3122aec0c0d0eb4990b7f 100644 (file)
@@ -269,6 +269,7 @@ struct perf_evsel *perf_evsel__new_cycles(void)
        struct perf_event_attr attr = {
                .type   = PERF_TYPE_HARDWARE,
                .config = PERF_COUNT_HW_CPU_CYCLES,
+               .exclude_kernel = 1,
        };
        struct perf_evsel *evsel;