perf sharded_mutex: Introduce sharded_mutex

Per object mutexes may come with significant memory cost while a
global mutex can suffer from unnecessary contention. A sharded mutex
is a compromise where objects are hashed and then a particular mutex
for the hash of the object used. Contention can be controlled by the
number of shards.

v2. Use hashmap.h's hash_bits in case of contention from alignment of
    objects.

Signed-off-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Cc: Andres Freund <andres@anarazel.de>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Yuan Can <yuancan@huawei.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Link: https://lore.kernel.org/r/20230615040715.2064350-1-irogers@google.com
Signed-off-by: Namhyung Kim <namhyung@kernel.org>

tools/perf/util/Build

@@ -145,6 +145,7 @@ perf-y += mem2node.o
 perf-y += clockid.o
 perf-y += list_sort.o
 perf-y += mutex.o
+perf-y += sharded_mutex.o
 
 perf-$(CONFIG_LIBBPF) += bpf-loader.o
 perf-$(CONFIG_LIBBPF) += bpf_map.o

tools/perf/util/sharded_mutex.c

@@ -0,0 +1,33 @@
+// SPDX-License-Identifier: GPL-2.0
+#include "sharded_mutex.h"
+
+#include <stdlib.h>
+
+struct sharded_mutex *sharded_mutex__new(size_t num_shards)
+{
+	struct sharded_mutex *result;
+	size_t size;
+	unsigned int bits;
+
+	for (bits = 0; ((size_t)1 << bits) < num_shards; bits++)
+		;
+
+	size = sizeof(*result) + sizeof(struct mutex) * (1 << bits);
+	result = malloc(size);
+	if (!result)
+		return NULL;
+
+	result->cap_bits = bits;
+	for (size_t i = 0; i < ((size_t)1 << bits); i++)
+		mutex_init(&result->mutexes[i]);
+
+	return result;
+}
+
+void sharded_mutex__delete(struct sharded_mutex *sm)
+{
+	for (size_t i = 0; i < ((size_t)1 << sm->cap_bits); i++)
+		mutex_destroy(&sm->mutexes[i]);
+
+	free(sm);
+}

tools/perf/util/sharded_mutex.h

@@ -0,0 +1,29 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef PERF_SHARDED_MUTEX_H
+#define PERF_SHARDED_MUTEX_H
+
+#include "mutex.h"
+#include "hashmap.h"
+
+/*
+ * In a situation where a lock is needed per object, having a mutex can be
+ * relatively memory expensive (40 bytes on x86-64). If the object can be
+ * constantly hashed, a sharded mutex is an alternative global pool of mutexes
+ * where the mutex is looked up from a hash value. This can lead to collisions
+ * if the number of shards isn't large enough.
+ */
+struct sharded_mutex {
+	/* mutexes array is 1<<cap_bits in size. */
+	unsigned int cap_bits;
+	struct mutex mutexes[];
+};
+
+struct sharded_mutex *sharded_mutex__new(size_t num_shards);
+void sharded_mutex__delete(struct sharded_mutex *sm);
+
+static inline struct mutex *sharded_mutex__get_mutex(struct sharded_mutex *sm, size_t hash)
+{
+	return &sm->mutexes[hash_bits(hash, sm->cap_bits)];
+}
+
+#endif  /* PERF_SHARDED_MUTEX_H */