[v2] perf evlist: Fix evlist__new_default() for > 1 core PMU

Message ID 20240123102728.239147-1-james.clark@arm.com
State New
Headers
Series [v2] perf evlist: Fix evlist__new_default() for > 1 core PMU |

Commit Message

James Clark Jan. 23, 2024, 10:27 a.m. UTC
  The 'Session topology' test currently fails with this message when
evlist__new_default() opens more than one event:

  32: Session topology                                                :
  --- start ---
  templ file: /tmp/perf-test-vv5YzZ
  Using CPUID 0x00000000410fd070
  Opening: unknown-hardware:HG
  ------------------------------------------------------------
  perf_event_attr:
    type                             0 (PERF_TYPE_HARDWARE)
    config                           0xb00000000
    disabled                         1
  ------------------------------------------------------------
  sys_perf_event_open: pid 0  cpu -1  group_fd -1  flags 0x8 = 4
  Opening: unknown-hardware:HG
  ------------------------------------------------------------
  perf_event_attr:
    type                             0 (PERF_TYPE_HARDWARE)
    config                           0xa00000000
    disabled                         1
  ------------------------------------------------------------
  sys_perf_event_open: pid 0  cpu -1  group_fd -1  flags 0x8 = 5
  non matching sample_type
  FAILED tests/topology.c:73 can't get session
  ---- end ----
  Session topology: FAILED!

This is because when re-opening the file and parsing the header, Perf
expects that any file that has more than one event has the sample ID
flag set. Perf record already sets the flag in a similar way when there
is more than one event, so add the same logic to evlist__new_default().

evlist__new_default() is only currently used in tests, so I don't
expect this change to have any other side effects. The other tests that
use it don't save and re-open the file so don't hit this issue.

The session topology test has been failing on Arm big.LITTLE platforms
since commit 251aa040244a ("perf parse-events: Wildcard most
"numeric" events") when evlist__new_default() started opening multiple
events for 'cycles'.

Fixes: 251aa040244a ("perf parse-events: Wildcard most "numeric" events")
Closes: https://lore.kernel.org/lkml/CAP-5=fWVQ-7ijjK3-w1q+k2WYVNHbAcejb-xY0ptbjRw476VKA@mail.gmail.com/
Tested-by: Ian Rogers <irogers@google.com>
Reviewed-by: Ian Rogers <irogers@google.com>
Tested-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: James Clark <james.clark@arm.com>
---
 tools/perf/util/evlist.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

Changes since v1:

  * Reduce scope of evsel variable
  * Add argument label
  * Change summary to be less specific about the failing test
  * Add the closes: tag
  

Comments

Namhyung Kim Jan. 24, 2024, 12:46 a.m. UTC | #1
Hi James,

On Tue, Jan 23, 2024 at 2:39 AM James Clark <james.clark@arm.com> wrote:
>
> The 'Session topology' test currently fails with this message when
> evlist__new_default() opens more than one event:
>
>   32: Session topology                                                :
>   --- start ---
>   templ file: /tmp/perf-test-vv5YzZ
>   Using CPUID 0x00000000410fd070
>   Opening: unknown-hardware:HG
>   ------------------------------------------------------------
>   perf_event_attr:
>     type                             0 (PERF_TYPE_HARDWARE)
>     config                           0xb00000000
>     disabled                         1
>   ------------------------------------------------------------
>   sys_perf_event_open: pid 0  cpu -1  group_fd -1  flags 0x8 = 4
>   Opening: unknown-hardware:HG
>   ------------------------------------------------------------
>   perf_event_attr:
>     type                             0 (PERF_TYPE_HARDWARE)
>     config                           0xa00000000
>     disabled                         1
>   ------------------------------------------------------------
>   sys_perf_event_open: pid 0  cpu -1  group_fd -1  flags 0x8 = 5
>   non matching sample_type
>   FAILED tests/topology.c:73 can't get session
>   ---- end ----
>   Session topology: FAILED!
>
> This is because when re-opening the file and parsing the header, Perf
> expects that any file that has more than one event has the sample ID
> flag set. Perf record already sets the flag in a similar way when there
> is more than one event, so add the same logic to evlist__new_default().
>
> evlist__new_default() is only currently used in tests, so I don't
> expect this change to have any other side effects. The other tests that
> use it don't save and re-open the file so don't hit this issue.
>
> The session topology test has been failing on Arm big.LITTLE platforms
> since commit 251aa040244a ("perf parse-events: Wildcard most
> "numeric" events") when evlist__new_default() started opening multiple
> events for 'cycles'.
>
> Fixes: 251aa040244a ("perf parse-events: Wildcard most "numeric" events")
> Closes: https://lore.kernel.org/lkml/CAP-5=fWVQ-7ijjK3-w1q+k2WYVNHbAcejb-xY0ptbjRw476VKA@mail.gmail.com/
> Tested-by: Ian Rogers <irogers@google.com>
> Reviewed-by: Ian Rogers <irogers@google.com>
> Tested-by: Kan Liang <kan.liang@linux.intel.com>
> Signed-off-by: James Clark <james.clark@arm.com>
> ---
>  tools/perf/util/evlist.c | 7 +++++++
>  1 file changed, 7 insertions(+)
>
> Changes since v2:
>
>    * Undo the fact that v2 was accidentally based on v1 instead of
>      perf-tools
>
> Changes since v1:
>
>   * Reduce scope of evsel variable
>   * Add argument label
>   * Change summary to be less specific about the failing test
>   * Add the closes: tag
>
> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
> index 95f25e9fb994..979a6053a84d 100644
> --- a/tools/perf/util/evlist.c
> +++ b/tools/perf/util/evlist.c
> @@ -106,6 +106,13 @@ struct evlist *evlist__new_default(void)
>                 evlist = NULL;
>         }
>
> +       if (evlist->core.nr_entries > 1) {

I think you need a NULL check for evlist here.

Thanks,
Namhyung


> +               struct evsel *evsel;
> +
> +               evlist__for_each_entry(evlist, evsel)
> +                       evsel__set_sample_id(evsel, /*can_sample_identifier=*/false);
> +       }
> +
>         return evlist;
>  }
>
> --
> 2.34.1
>
  
Arnaldo Carvalho de Melo Jan. 26, 2024, 4:25 p.m. UTC | #2
Em Wed, Jan 24, 2024 at 09:03:57AM +0000, James Clark escreveu:
> On 24/01/2024 00:46, Namhyung Kim wrote:
> > On Tue, Jan 23, 2024 at 2:39 AM James Clark <james.clark@arm.com> wrote:
> >> +++ b/tools/perf/util/evlist.c
> >> @@ -106,6 +106,13 @@ struct evlist *evlist__new_default(void)
> >>                 evlist = NULL;
> >>         }

> >> +       if (evlist->core.nr_entries > 1) {

> > I think you need a NULL check for evlist here.
 
> Oops yes. Or just return on the error above.

Was there a v4? I'm assuming this is for perf-tools, i.e. for v6.8-rc,
right?

- Arnaldo
  

Patch

diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 56db37fac6f6..979a6053a84d 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -95,7 +95,6 @@  struct evlist *evlist__new_default(void)
 	struct evlist *evlist = evlist__new();
 	bool can_profile_kernel;
 	int err;
-	struct evsel *evsel;
 
 	if (!evlist)
 		return NULL;
@@ -107,9 +106,12 @@  struct evlist *evlist__new_default(void)
 		evlist = NULL;
 	}
 
-	if (evlist->core.nr_entries > 1)
+	if (evlist->core.nr_entries > 1) {
+		struct evsel *evsel;
+
 		evlist__for_each_entry(evlist, evsel)
-			evsel__set_sample_id(evsel, false);
+			evsel__set_sample_id(evsel, /*can_sample_identifier=*/false);
+	}
 
 	return evlist;
 }