[PATCH v1] perf cpumap: Increment reference count for online cpumap

[PATCH v1] perf cpumap: Increment reference count for online cpumap

[Ian Rogers]

Thomas Richter <tmricht@linux.ibm.com> reported a double put on the
cpumap for the placeholder core PMU:
https://lore.kernel.org/lkml/20250318095132.1502654-3-tmricht@linux.ibm.com/
Requiring the caller to get the cpumap is not how these things are
usually done, switch cpu_map__online to do the get and then fix up any
use cases where a put is needed.

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/arch/arm/util/pmu.c | 5 +++--
 tools/perf/util/cpumap.c       | 2 +-
 tools/perf/util/evlist.c       | 5 +++--
 tools/perf/util/mem-events.c   | 5 ++++-
 tools/perf/util/mmap.c         | 4 ++--
 tools/perf/util/pmu.c          | 2 +-
 tools/perf/util/tool_pmu.c     | 1 +
 7 files changed, 15 insertions(+), 9 deletions(-)

diff --git a/tools/perf/arch/arm/util/pmu.c b/tools/perf/arch/arm/util/pmu.c
index 57dc94a6e38c..f70075c89aa0 100644
--- a/tools/perf/arch/arm/util/pmu.c
+++ b/tools/perf/arch/arm/util/pmu.c
@@ -18,7 +18,7 @@
 
 void perf_pmu__arch_init(struct perf_pmu *pmu)
 {
-	struct perf_cpu_map *intersect;
+	struct perf_cpu_map *intersect, *online = cpu_map__online();
 
 #ifdef HAVE_AUXTRACE_SUPPORT
 	if (!strcmp(pmu->name, CORESIGHT_ETM_PMU_NAME)) {
@@ -41,7 +41,8 @@ void perf_pmu__arch_init(struct perf_pmu *pmu)
 	}
 #endif
 	/* Workaround some ARM PMU's failing to correctly set CPU maps for online processors. */
-	intersect = perf_cpu_map__intersect(cpu_map__online(), pmu->cpus);
+	intersect = perf_cpu_map__intersect(online, pmu->cpus);
+	perf_cpu_map__put(online);
 	perf_cpu_map__put(pmu->cpus);
 	pmu->cpus = intersect;
 }
diff --git a/tools/perf/util/cpumap.c b/tools/perf/util/cpumap.c
index 9bc5e0370234..89570397a4b3 100644
--- a/tools/perf/util/cpumap.c
+++ b/tools/perf/util/cpumap.c
@@ -722,7 +722,7 @@ struct perf_cpu_map *cpu_map__online(void) /* thread unsafe */
 	if (!online)
 		online = perf_cpu_map__new_online_cpus(); /* from /sys/devices/system/cpu/online */
 
-	return online;
+	return perf_cpu_map__get(online);
 }
 
 bool aggr_cpu_id__equal(const struct aggr_cpu_id *a, const struct aggr_cpu_id *b)
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 49e10d6981ad..c1a04141aed0 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -2534,10 +2534,10 @@ void evlist__warn_user_requested_cpus(struct evlist *evlist, const char *cpu_lis
 		return;
 
 	evlist__for_each_entry(evlist, pos) {
-		struct perf_cpu_map *intersect, *to_test;
+		struct perf_cpu_map *intersect, *to_test, *online = cpu_map__online();
 		const struct perf_pmu *pmu = evsel__find_pmu(pos);
 
-		to_test = pmu && pmu->is_core ? pmu->cpus : cpu_map__online();
+		to_test = pmu && pmu->is_core ? pmu->cpus : online;
 		intersect = perf_cpu_map__intersect(to_test, user_requested_cpus);
 		if (!perf_cpu_map__equal(intersect, user_requested_cpus)) {
 			char buf[128];
@@ -2547,6 +2547,7 @@ void evlist__warn_user_requested_cpus(struct evlist *evlist, const char *cpu_lis
 				cpu_list, pmu ? pmu->name : "cpu", buf, evsel__name(pos));
 		}
 		perf_cpu_map__put(intersect);
+		perf_cpu_map__put(online);
 	}
 	perf_cpu_map__put(user_requested_cpus);
 }
diff --git a/tools/perf/util/mem-events.c b/tools/perf/util/mem-events.c
index 9011784b950d..884d9aebce91 100644
--- a/tools/perf/util/mem-events.c
+++ b/tools/perf/util/mem-events.c
@@ -303,12 +303,15 @@ int perf_mem_events__record_args(const char **rec_argv, int *argv_nr, char **eve
 	}
 
 	if (cpu_map) {
-		if (!perf_cpu_map__equal(cpu_map, cpu_map__online())) {
+		struct perf_cpu_map *online = cpu_map__online();
+
+		if (!perf_cpu_map__equal(cpu_map, online)) {
 			char buf[200];
 
 			cpu_map__snprint(cpu_map, buf, sizeof(buf));
 			pr_warning("Memory events are enabled on a subset of CPUs: %s\n", buf);
 		}
+		perf_cpu_map__put(online);
 		perf_cpu_map__put(cpu_map);
 	}
 
diff --git a/tools/perf/util/mmap.c b/tools/perf/util/mmap.c
index a7ef4d8d57d8..a34726219af3 100644
--- a/tools/perf/util/mmap.c
+++ b/tools/perf/util/mmap.c
@@ -244,9 +244,8 @@ static void build_node_mask(int node, struct mmap_cpu_mask *mask)
 {
 	int idx, nr_cpus;
 	struct perf_cpu cpu;
-	const struct perf_cpu_map *cpu_map = NULL;
+	struct perf_cpu_map *cpu_map = cpu_map__online();
 
-	cpu_map = cpu_map__online();
 	if (!cpu_map)
 		return;
 
@@ -256,6 +255,7 @@ static void build_node_mask(int node, struct mmap_cpu_mask *mask)
 		if (cpu__get_node(cpu) == node)
 			__set_bit(cpu.cpu, mask->bits);
 	}
+	perf_cpu_map__put(cpu_map);
 }
 
 static int perf_mmap__setup_affinity_mask(struct mmap *map, struct mmap_params *mp)
diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
index 57450c73fb63..b7ebac5ab1d1 100644
--- a/tools/perf/util/pmu.c
+++ b/tools/perf/util/pmu.c
@@ -779,7 +779,7 @@ static struct perf_cpu_map *pmu_cpumask(int dirfd, const char *pmu_name, bool is
 	}
 
 	/* Nothing found, for core PMUs assume this means all CPUs. */
-	return is_core ? perf_cpu_map__get(cpu_map__online()) : NULL;
+	return is_core ? cpu_map__online() : NULL;
 }
 
 static bool pmu_is_uncore(int dirfd, const char *name)
diff --git a/tools/perf/util/tool_pmu.c b/tools/perf/util/tool_pmu.c
index 9156745ea180..b60ac390d52d 100644
--- a/tools/perf/util/tool_pmu.c
+++ b/tools/perf/util/tool_pmu.c
@@ -355,6 +355,7 @@ bool tool_pmu__read_event(enum tool_pmu_event ev, u64 *result)
 
 		if (online) {
 			*result = perf_cpu_map__nr(online);
+			perf_cpu_map__put(online);
 			return true;
 		}
 		return false;
-- 
2.49.0.rc1.451.g8f38331e32-goog

Re: [PATCH v1] perf cpumap: Increment reference count for online cpumap

[Thomas Richter]

On 3/18/25 18:19, Ian Rogers wrote:
> Thomas Richter <tmricht@linux.ibm.com> reported a double put on the
> cpumap for the placeholder core PMU:
> https://lore.kernel.org/lkml/20250318095132.1502654-3-tmricht@linux.ibm.com/
> Requiring the caller to get the cpumap is not how these things are
> usually done, switch cpu_map__online to do the get and then fix up any
> use cases where a put is needed.
> 
> Signed-off-by: Ian Rogers <irogers@google.com>

Ian,

thanks for fixing this. Here are the results on s390 z/VM:

# ./perf bench internals pmu-scan
# Running 'internals/pmu-scan' benchmark:
Computing performance of sysfs PMU event scan for 100 times
  Average core PMU scanning took: 6.340 usec (+- 0.050 usec)
  Average PMU scanning took: 93.785 usec (+- 6.210 usec)
# 

Tested-by: Thomas Richter <tmricht@linux.ibm.com>

-- 
Thomas Richter, Dept 3303, IBM s390 Linux Development, Boeblingen, Germany
--
IBM Deutschland Research & Development GmbH

Vorsitzender des Aufsichtsrats: Wolfgang Wendt

Geschäftsführung: David Faller

Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294

Re: [PATCH v1] perf cpumap: Increment reference count for online cpumap

[Namhyung Kim]

On Tue, 18 Mar 2025 10:19:14 -0700, Ian Rogers wrote:
> Thomas Richter <tmricht@linux.ibm.com> reported a double put on the
> cpumap for the placeholder core PMU:
> https://lore.kernel.org/lkml/20250318095132.1502654-3-tmricht@linux.ibm.com/
> Requiring the caller to get the cpumap is not how these things are
> usually done, switch cpu_map__online to do the get and then fix up any
> use cases where a put is needed.
> 
> [...]
Applied to perf-tools-next, thanks!

Best regards,
Namhyung