[PATCH v2] perf evlist: Don't restrict uncore events to PMU CPUs

[PATCH v2] perf evlist: Don't restrict uncore events to PMU CPUs

[Nia Su]

The PMU CPU map has different semantics for core and uncore PMUs.

For core PMUs, pmu_cpus contains the CPUs on which the event can be
opened. Intersect the user-requested CPU map with the PMU CPU map to
reject CPUs unsupported by the PMU.

For uncore PMUs, pmu_cpus is read from the PMU's cpumask sysfs attribute,
which lists the default CPU to open the event on per socket/die, not the
set of CPUs it can only be opened on. The user-requested CPU may override
these defaults, and the driver can redirect the event to the appropriate
control CPU.

Currently, __perf_evlist__propagate_maps() intersects the user-requested
CPU map with pmu_cpus for all PMUs. If the user-requested CPU is not in
pmu_cpus, the intersection becomes empty and the evsel is removed before
perf_event_open() is called.

Add evsel->is_pmu_uncore, set once from pmu->is_uncore at parse time,
to skip the intersection for uncore PMUs. Core PMUs, and other
non-core PMUs such as software events, breakpoints, or tracepoints,
continue to be restricted to their PMU CPU map.

Fixes: 811082e4b668 ("perf parse-events: Support user CPUs mixed with threads/processes")

Assisted-by: Claude Sonnet 5
Signed-off-by: Nia Su <nia.su@sifive.com>
---
Changes in v2:
- Add a dedicated is_pmu_uncore flag instead of reusing requires_cpu,
  since it more precisely represents whether the PMU is uncore.
- Link to v1: https://lore.kernel.org/r/20260904-perf-evlist-uncore-cpu-b4-v1-1-604b55c025ee@sifive.com
---
 tools/lib/perf/evlist.c                 | 5 +++--
 tools/lib/perf/include/internal/evsel.h | 2 ++
 tools/perf/util/evsel.c                 | 1 +
 tools/perf/util/parse-events.c          | 1 +
 4 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/tools/lib/perf/evlist.c b/tools/lib/perf/evlist.c
index 1f210dadd666ff818a23712ab95acaa659ccf8dc..c1c5b0684eedb7a6ab8f3df85f326fbd5ef809a2 100644
--- a/tools/lib/perf/evlist.c
+++ b/tools/lib/perf/evlist.c
@@ -79,8 +79,9 @@ static void __perf_evlist__propagate_maps(struct perf_evlist *evlist,
 		}
 	}
 
-	/* Ensure cpus only references valid PMU CPUs. */
-	if (!perf_cpu_map__has_any_cpu(evsel->cpus) &&
+	/* Ensure cpus only references valid PMU CPUs, except for uncore PMUs. */
+	if (!evsel->is_pmu_uncore &&
+	    !perf_cpu_map__has_any_cpu(evsel->cpus) &&
 	    !perf_cpu_map__is_subset(evsel->pmu_cpus, evsel->cpus)) {
 		struct perf_cpu_map *tmp = perf_cpu_map__intersect(evsel->pmu_cpus, evsel->cpus);
 
diff --git a/tools/lib/perf/include/internal/evsel.h b/tools/lib/perf/include/internal/evsel.h
index b988034f13710ec4d49d942d486c4b9afe21a1f0..35447bed0f9586b538dad40c223e70d50486c5f6 100644
--- a/tools/lib/perf/include/internal/evsel.h
+++ b/tools/lib/perf/include/internal/evsel.h
@@ -128,6 +128,8 @@ struct perf_evsel {
 	bool			 requires_cpu;
 	/** Is the PMU for the event a core one? Effects the handling of own_cpus. */
 	bool			 is_pmu_core;
+	/** Is the PMU for the event an uncore one? */
+	bool			 is_pmu_uncore;
 	/** Does the evsel on read on the first CPU index such as tool time events? */
 	bool			 reads_only_on_cpu_idx0;
 	int			 idx;
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index d4cb455f4a7d45ae2f9133709ff3248b54163ce0..5acec3f6345099ef9e46e81614b9df2595925b1c 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -551,6 +551,7 @@ struct evsel *evsel__clone(struct evsel *orig)
 	evsel->core.system_wide = orig->core.system_wide;
 	evsel->core.requires_cpu = orig->core.requires_cpu;
 	evsel->core.is_pmu_core = orig->core.is_pmu_core;
+	evsel->core.is_pmu_uncore = orig->core.is_pmu_uncore;
 
 	if (orig->name) {
 		evsel->name = strdup(orig->name);
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index cc7ad331a49f34536f66724370b8bb94ab65bacd..8302be207c1d6ecf8ae6fd67f54d66d08d50e01c 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -269,6 +269,7 @@ __add_event(struct list_head *list, int *idx,
 	evsel->core.pmu_cpus = pmu_cpus;
 	evsel->core.requires_cpu = pmu ? pmu->is_uncore : false;
 	evsel->core.is_pmu_core = is_pmu_core;
+	evsel->core.is_pmu_uncore = pmu ? pmu->is_uncore : false;
 	evsel->core.reads_only_on_cpu_idx0 = perf_pmu__reads_only_on_cpu_idx0(attr);
 	evsel->pmu = pmu;
 	evsel->alternate_hw_config = alternate_hw_config;

---
base-commit: 73e3f0710014fe6d4ed98cfc02292f6121db7558
change-id: 20260904-perf-evlist-uncore-cpu-b4-418af0f17c80

--

Re: [PATCH v2] perf evlist: Don't restrict uncore events to PMU CPUs

[sashiko-bot]

> The PMU CPU map has different semantics for core and uncore PMUs.
> 
> For core PMUs, pmu_cpus contains the CPUs on which the event can be
> opened. Intersect the user-requested CPU map with the PMU CPU map to
> reject CPUs unsupported by the PMU.
> 
> For uncore PMUs, pmu_cpus is read from the PMU's cpumask sysfs attribute,
> which lists the default CPU to open the event on per socket/die, not the
> set of CPUs it can only be opened on. The user-requested CPU may override
> these defaults, and the driver can redirect the event to the appropriate
> control CPU.
> 
> Currently, __perf_evlist__propagate_maps() intersects the user-requested
> CPU map with pmu_cpus for all PMUs. If the user-requested CPU is not in
> pmu_cpus, the intersection becomes empty and the evsel is removed before
> perf_event_open() is called.
> 
> Add evsel->is_pmu_uncore, set once from pmu->is_uncore at parse time,
> to skip the intersection for uncore PMUs. Core PMUs, and other
> non-core PMUs such as software events, breakpoints, or tracepoints,
> continue to be restricted to their PMU CPU map.
> 
> Fixes: 811082e4b668 ("perf parse-events: Support user CPUs mixed with threads/processes")
> 
> Assisted-by: Claude Sonnet 5
> Signed-off-by: Nia Su <nia.su@sifive.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260909-perf-evlist-uncore-cpu-b4-v2-1-dc673a1f4592@sifive.com?part=1