[PATCH v3 3/3] perf record: adapt affinity to machines with #CPUs > 1K

[Alexey Budankov]

Use struct mmap_cpu_mask type for tool's thread and mmap data
buffers to overcome current 1024 CPUs mask size limitation of
cpu_set_t type.

Currently glibc cpu_set_t type has internal mask size limit
of 1024 CPUs. Moving to struct mmap_cpu_mask type allows
overcoming that limit. tools bitmap API is used to manipulate
objects of struct mmap_cpu_mask type.

Reported-by: Andi Kleen <ak@xxxxxxxxxxxxxxx>
Signed-off-by: Alexey Budankov <alexey.budankov@xxxxxxxxxxxxxxx>
---
Changes in v3:
- use perf_mmap__print_cpu_mask() to log thread and mmap cpus masks
  when verbose level is equal to 2
Changes in v2:
- returned -1 from perf_mmap__setup_affinity_mask()
- moved debug printing under -vv option
- implemented releasing of masks using bitmap_free()
---
 tools/perf/builtin-record.c | 27 +++++++++++++++++++++------
 tools/perf/util/mmap.c      | 28 ++++++++++++++++++++++------
 tools/perf/util/mmap.h      |  2 +-
 3 files changed, 44 insertions(+), 13 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index b95c000c1ed9..b36699bf296b 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -62,6 +62,7 @@
 #include <linux/string.h>
 #include <linux/time64.h>
 #include <linux/zalloc.h>
+#include <linux/bitmap.h>
 
 struct switch_output {
 	bool		 enabled;
@@ -93,7 +94,7 @@ struct record {
 	bool			timestamp_boundary;
 	struct switch_output	switch_output;
 	unsigned long long	samples;
-	cpu_set_t		affinity_mask;
+	struct mmap_cpu_mask	affinity_mask;
 	unsigned long		output_max_size;	/* = 0: unlimited */
 };
 
@@ -954,10 +955,15 @@ static struct perf_event_header finished_round_event = {
 static void record__adjust_affinity(struct record *rec, struct mmap *map)
 {
 	if (rec->opts.affinity != PERF_AFFINITY_SYS &&
-	    !CPU_EQUAL(&rec->affinity_mask, &map->affinity_mask)) {
-		CPU_ZERO(&rec->affinity_mask);
-		CPU_OR(&rec->affinity_mask, &rec->affinity_mask, &map->affinity_mask);
-		sched_setaffinity(0, sizeof(rec->affinity_mask), &rec->affinity_mask);
+	    !bitmap_equal(rec->affinity_mask.bits, map->affinity_mask.bits,
+			  rec->affinity_mask.nbits)) {
+		bitmap_zero(rec->affinity_mask.bits, rec->affinity_mask.nbits);
+		bitmap_or(rec->affinity_mask.bits, rec->affinity_mask.bits,
+			  map->affinity_mask.bits, rec->affinity_mask.nbits);
+		sched_setaffinity(0, MMAP_CPU_MASK_BYTES(&rec->affinity_mask),
+				  (cpu_set_t *)rec->affinity_mask.bits);
+		if (verbose == 2)
+			perf_mmap__print_cpu_mask(&map->affinity_mask, "thread");
 	}
 }
 
@@ -2389,7 +2395,6 @@ int cmd_record(int argc, const char **argv)
 # undef REASON
 #endif
 
-	CPU_ZERO(&rec->affinity_mask);
 	rec->opts.affinity = PERF_AFFINITY_SYS;
 
 	rec->evlist = evlist__new();
@@ -2455,6 +2460,14 @@ int cmd_record(int argc, const char **argv)
 
 	symbol__init(NULL);
 
+	rec->affinity_mask.nbits = cpu__max_cpu();
+	rec->affinity_mask.bits = bitmap_alloc(rec->affinity_mask.nbits);
+	if (!rec->affinity_mask.bits) {
+		pr_err("Failed to allocate thread mask for %ld cpus\n", rec->affinity_mask.nbits);
+		return -ENOMEM;
+	}
+	pr_debug2("thread mask[%ld]: empty\n", rec->affinity_mask.nbits);
+
 	err = record__auxtrace_init(rec);
 	if (err)
 		goto out;
@@ -2569,6 +2582,8 @@ int cmd_record(int argc, const char **argv)
 
 	err = __cmd_record(&record, argc, argv);
 out:
+	if (rec->affinity_mask.bits)
+		bitmap_free(rec->affinity_mask.bits);
 	evlist__delete(rec->evlist);
 	symbol__exit();
 	auxtrace_record__free(rec->itr);
diff --git a/tools/perf/util/mmap.c b/tools/perf/util/mmap.c
index 30ff7aef06f2..615d05870849 100644
--- a/tools/perf/util/mmap.c
+++ b/tools/perf/util/mmap.c
@@ -219,6 +219,9 @@ static void perf_mmap__aio_munmap(struct mmap *map __maybe_unused)
 
 void mmap__munmap(struct mmap *map)
 {
+	if (map->affinity_mask.bits)
+		bitmap_free(map->affinity_mask.bits);
+
 	perf_mmap__aio_munmap(map);
 	if (map->data != NULL) {
 		munmap(map->data, mmap__mmap_len(map));
@@ -227,7 +230,7 @@ void mmap__munmap(struct mmap *map)
 	auxtrace_mmap__munmap(&map->auxtrace_mmap);
 }
 
-static void build_node_mask(int node, cpu_set_t *mask)
+static void build_node_mask(int node, struct mmap_cpu_mask *mask)
 {
 	int c, cpu, nr_cpus;
 	const struct perf_cpu_map *cpu_map = NULL;
@@ -240,17 +243,23 @@ static void build_node_mask(int node, cpu_set_t *mask)
 	for (c = 0; c < nr_cpus; c++) {
 		cpu = cpu_map->map[c]; /* map c index to online cpu index */
 		if (cpu__get_node(cpu) == node)
-			CPU_SET(cpu, mask);
+			set_bit(cpu, mask->bits);
 	}
 }
 
-static void perf_mmap__setup_affinity_mask(struct mmap *map, struct mmap_params *mp)
+static int perf_mmap__setup_affinity_mask(struct mmap *map, struct mmap_params *mp)
 {
-	CPU_ZERO(&map->affinity_mask);
+	map->affinity_mask.nbits = cpu__max_cpu();
+	map->affinity_mask.bits = bitmap_alloc(map->affinity_mask.nbits);
+	if (!map->affinity_mask.bits)
+		return -1;
+
 	if (mp->affinity == PERF_AFFINITY_NODE && cpu__max_node() > 1)
 		build_node_mask(cpu__get_node(map->core.cpu), &map->affinity_mask);
 	else if (mp->affinity == PERF_AFFINITY_CPU)
-		CPU_SET(map->core.cpu, &map->affinity_mask);
+		set_bit(map->core.cpu, map->affinity_mask.bits);
+
+	return 0;
 }
 
 int mmap__mmap(struct mmap *map, struct mmap_params *mp, int fd, int cpu)
@@ -261,7 +270,14 @@ int mmap__mmap(struct mmap *map, struct mmap_params *mp, int fd, int cpu)
 		return -1;
 	}
 
-	perf_mmap__setup_affinity_mask(map, mp);
+	if (perf_mmap__setup_affinity_mask(map, mp)) {
+		pr_debug2("failed to alloc mmap affinity mask, error %d\n",
+			  errno);
+		return -1;
+	}
+
+	if (verbose == 2)
+		perf_mmap__print_cpu_mask(&map->affinity_mask, "mmap");
 
 	map->core.flush = mp->flush;
 
diff --git a/tools/perf/util/mmap.h b/tools/perf/util/mmap.h
index 598e2def8a48..ab6449675d5b 100644
--- a/tools/perf/util/mmap.h
+++ b/tools/perf/util/mmap.h
@@ -40,7 +40,7 @@ struct mmap {
 		int		 nr_cblocks;
 	} aio;
 #endif
-	cpu_set_t	affinity_mask;
+	struct mmap_cpu_mask	affinity_mask;
 	void		*data;
 	int		comp_level;
 };
-- 
2.20.1