perf record: Extend --threads command line option

Extend --threads option in perf record command line interface.
The option can have a value in the form of masks that specify
CPUs to be monitored with data streaming threads and its layout
in system topology. The masks can be filtered using CPU mask
provided via -C option.

The specification value can be user defined list of masks. Masks
separated by colon define CPUs to be monitored by one thread and
affinity mask of that thread is separated by slash. For example:
<cpus mask 1>/<affinity mask 1>:<cpu mask 2>/<affinity mask 2>
specifies parallel threads layout that consists of two threads
with corresponding assigned CPUs to be monitored.

The specification value can be a string e.g. "cpu", "core" or
"package" meaning creation of data streaming thread for every
CPU or core or package to monitor distinct CPUs or CPUs grouped
by core or package.

The option provided with no or empty value defaults to per-cpu
parallel threads layout creating data streaming thread for every
CPU being monitored.

Document --threads option syntax and parallel data streaming modes
in Documentation/perf-record.txt.

Suggested-by: Jiri Olsa <[e-mail]>
Suggested-by: Namhyung Kim <[e-mail]>
Reviewed-by: Riccardo Mancini <[e-mail]>
Signed-off-by: Alexey Bayduraev <[e-mail]>
Tested-by: Jiri Olsa <[e-mail]>
Tested-by: Riccardo Mancini <[e-mail]>
Acked-by: Andi Kleen <[e-mail]>
Acked-by: Namhyung Kim <[e-mail]>
Cc: Adrian Hunter <[e-mail]>
Cc: Alexander Antonov <[e-mail]>
Cc: Alexander Shishkin <[e-mail]>
Cc: Alexei Budankov <[e-mail]>
Cc: Ingo Molnar <[e-mail]>
Cc: Peter Zijlstra <[e-mail]>
Link: https://lore.kernel.org/r/079e2619be[e-mail]
Signed-off-by: Arnaldo Carvalho de Melo <[e-mail]>

tools/perf/Documentation/perf-record.txt

···
  wait -n ${perf_pid}
  exit $?
 
---threads::
+--threads=<spec>::
 Write collected trace data into several data files using parallel threads.
-The option creates a data streaming thread for each CPU in the system.
+<spec> value can be user defined list of masks. Masks separated by colon
+define CPUs to be monitored by a thread and affinity mask of that thread
+is separated by slash:
+
+    <cpus mask 1>/<affinity mask 1>:<cpus mask 2>/<affinity mask 2>:...
+
+CPUs or affinity masks must not overlap with other corresponding masks.
+Invalid CPUs are ignored, but masks containing only invalid CPUs are not
+allowed.
+
+For example user specification like the following:
+
+    0,2-4/2-4:1,5-7/5-7
+
+specifies parallel threads layout that consists of two threads,
+the first thread monitors CPUs 0 and 2-4 with the affinity mask 2-4,
+the second monitors CPUs 1 and 5-7 with the affinity mask 5-7.
+
+<spec> value can also be a string meaning predefined parallel threads
+layout:
+
+    cpu    - create new data streaming thread for every monitored cpu
+    core   - create new thread to monitor CPUs grouped by a core
+    package - create new thread to monitor CPUs grouped by a package
+    numa   - create new threed to monitor CPUs grouped by a NUMA domain
+
+Predefined layouts can be used on systems with large number of CPUs in
+order not to spawn multiple per-cpu streaming threads but still avoid LOST
+events in data directory files. Option specified with no or empty value
+defaults to CPU layout. Masks defined or provided by the option value are
+filtered through the mask provided by -C option.
 
 include::intel-hybrid.txt[]
 

tools/perf/builtin-record.c

···
 #include "util/evlist-hybrid.h"
 #include "asm/bug.h"
 #include "perf.h"
+#include "cputopo.h"
 
 #include <errno.h>
 #include <inttypes.h>
···
 enum thread_spec {
 	THREAD_SPEC__UNDEFINED = 0,
 	THREAD_SPEC__CPU,
+	THREAD_SPEC__CORE,
+	THREAD_SPEC__PACKAGE,
+	THREAD_SPEC__NUMA,
+	THREAD_SPEC__USER,
+	THREAD_SPEC__MAX,
+};
+
+static const char *thread_spec_tags[THREAD_SPEC__MAX] = {
+	"undefined", "cpu", "core", "package", "numa", "user"
 };
 
 struct record {
···
 
 static int record__parse_threads(const struct option *opt, const char *str, int unset)
 {
+	int s;
 	struct record_opts *opts = opt->value;
 
-	if (unset || !str || !strlen(str))
+	if (unset || !str || !strlen(str)) {
 		opts->threads_spec = THREAD_SPEC__CPU;
+	} else {
+		for (s = 1; s < THREAD_SPEC__MAX; s++) {
+			if (s == THREAD_SPEC__USER) {
+				opts->threads_user_spec = strdup(str);
+				if (!opts->threads_user_spec)
+					return -ENOMEM;
+				opts->threads_spec = THREAD_SPEC__USER;
+				break;
+			}
+			if (!strncasecmp(str, thread_spec_tags[s], strlen(thread_spec_tags[s]))) {
+				opts->threads_spec = s;
+				break;
+			}
+		}
+	}
+
+	if (opts->threads_spec == THREAD_SPEC__USER)
+		pr_debug("threads_spec: %s\n", opts->threads_user_spec);
+	else
+		pr_debug("threads_spec: %s\n", thread_spec_tags[opts->threads_spec]);
 
 	return 0;
 }
···
 		set_bit(cpus->map[c].cpu, mask->bits);
 }
 
+static int record__mmap_cpu_mask_init_spec(struct mmap_cpu_mask *mask, const char *mask_spec)
+{
+	struct perf_cpu_map *cpus;
+
+	cpus = perf_cpu_map__new(mask_spec);
+	if (!cpus)
+		return -ENOMEM;
+
+	bitmap_zero(mask->bits, mask->nbits);
+	record__mmap_cpu_mask_init(mask, cpus);
+	perf_cpu_map__put(cpus);
+
+	return 0;
+}
+
 static void record__free_thread_masks(struct record *rec, int nr_threads)
 {
 	int t;
···
 	return 0;
 }
 
+static int record__init_thread_masks_spec(struct record *rec, struct perf_cpu_map *cpus,
+					  const char **maps_spec, const char **affinity_spec,
+					  u32 nr_spec)
+{
+	u32 s;
+	int ret = 0, t = 0;
+	struct mmap_cpu_mask cpus_mask;
+	struct thread_mask thread_mask, full_mask, *thread_masks;
+
+	ret = record__mmap_cpu_mask_alloc(&cpus_mask, cpu__max_cpu().cpu);
+	if (ret) {
+		pr_err("Failed to allocate CPUs mask\n");
+		return ret;
+	}
+	record__mmap_cpu_mask_init(&cpus_mask, cpus);
+
+	ret = record__thread_mask_alloc(&full_mask, cpu__max_cpu().cpu);
+	if (ret) {
+		pr_err("Failed to allocate full mask\n");
+		goto out_free_cpu_mask;
+	}
+
+	ret = record__thread_mask_alloc(&thread_mask, cpu__max_cpu().cpu);
+	if (ret) {
+		pr_err("Failed to allocate thread mask\n");
+		goto out_free_full_and_cpu_masks;
+	}
+
+	for (s = 0; s < nr_spec; s++) {
+		ret = record__mmap_cpu_mask_init_spec(&thread_mask.maps, maps_spec[s]);
+		if (ret) {
+			pr_err("Failed to initialize maps thread mask\n");
+			goto out_free;
+		}
+		ret = record__mmap_cpu_mask_init_spec(&thread_mask.affinity, affinity_spec[s]);
+		if (ret) {
+			pr_err("Failed to initialize affinity thread mask\n");
+			goto out_free;
+		}
+
+		/* ignore invalid CPUs but do not allow empty masks */
+		if (!bitmap_and(thread_mask.maps.bits, thread_mask.maps.bits,
+				cpus_mask.bits, thread_mask.maps.nbits)) {
+			pr_err("Empty maps mask: %s\n", maps_spec[s]);
+			ret = -EINVAL;
+			goto out_free;
+		}
+		if (!bitmap_and(thread_mask.affinity.bits, thread_mask.affinity.bits,
+				cpus_mask.bits, thread_mask.affinity.nbits)) {
+			pr_err("Empty affinity mask: %s\n", affinity_spec[s]);
+			ret = -EINVAL;
+			goto out_free;
+		}
+
+		/* do not allow intersection with other masks (full_mask) */
+		if (bitmap_intersects(thread_mask.maps.bits, full_mask.maps.bits,
+				      thread_mask.maps.nbits)) {
+			pr_err("Intersecting maps mask: %s\n", maps_spec[s]);
+			ret = -EINVAL;
+			goto out_free;
+		}
+		if (bitmap_intersects(thread_mask.affinity.bits, full_mask.affinity.bits,
+				      thread_mask.affinity.nbits)) {
+			pr_err("Intersecting affinity mask: %s\n", affinity_spec[s]);
+			ret = -EINVAL;
+			goto out_free;
+		}
+
+		bitmap_or(full_mask.maps.bits, full_mask.maps.bits,
+			  thread_mask.maps.bits, full_mask.maps.nbits);
+		bitmap_or(full_mask.affinity.bits, full_mask.affinity.bits,
+			  thread_mask.affinity.bits, full_mask.maps.nbits);
+
+		thread_masks = realloc(rec->thread_masks, (t + 1) * sizeof(struct thread_mask));
+		if (!thread_masks) {
+			pr_err("Failed to reallocate thread masks\n");
+			ret = -ENOMEM;
+			goto out_free;
+		}
+		rec->thread_masks = thread_masks;
+		rec->thread_masks[t] = thread_mask;
+		if (verbose) {
+			pr_debug("thread_masks[%d]: ", t);
+			mmap_cpu_mask__scnprintf(&rec->thread_masks[t].maps, "maps");
+			pr_debug("thread_masks[%d]: ", t);
+			mmap_cpu_mask__scnprintf(&rec->thread_masks[t].affinity, "affinity");
+		}
+		t++;
+		ret = record__thread_mask_alloc(&thread_mask, cpu__max_cpu().cpu);
+		if (ret) {
+			pr_err("Failed to allocate thread mask\n");
+			goto out_free_full_and_cpu_masks;
+		}
+	}
+	rec->nr_threads = t;
+	pr_debug("nr_threads: %d\n", rec->nr_threads);
+	if (!rec->nr_threads)
+		ret = -EINVAL;
+
+out_free:
+	record__thread_mask_free(&thread_mask);
+out_free_full_and_cpu_masks:
+	record__thread_mask_free(&full_mask);
+out_free_cpu_mask:
+	record__mmap_cpu_mask_free(&cpus_mask);
+
+	return ret;
+}
+
+static int record__init_thread_core_masks(struct record *rec, struct perf_cpu_map *cpus)
+{
+	int ret;
+	struct cpu_topology *topo;
+
+	topo = cpu_topology__new();
+	if (!topo) {
+		pr_err("Failed to allocate CPU topology\n");
+		return -ENOMEM;
+	}
+
+	ret = record__init_thread_masks_spec(rec, cpus, topo->core_cpus_list,
+					     topo->core_cpus_list, topo->core_cpus_lists);
+	cpu_topology__delete(topo);
+
+	return ret;
+}
+
+static int record__init_thread_package_masks(struct record *rec, struct perf_cpu_map *cpus)
+{
+	int ret;
+	struct cpu_topology *topo;
+
+	topo = cpu_topology__new();
+	if (!topo) {
+		pr_err("Failed to allocate CPU topology\n");
+		return -ENOMEM;
+	}
+
+	ret = record__init_thread_masks_spec(rec, cpus, topo->package_cpus_list,
+					     topo->package_cpus_list, topo->package_cpus_lists);
+	cpu_topology__delete(topo);
+
+	return ret;
+}
+
+static int record__init_thread_numa_masks(struct record *rec, struct perf_cpu_map *cpus)
+{
+	u32 s;
+	int ret;
+	const char **spec;
+	struct numa_topology *topo;
+
+	topo = numa_topology__new();
+	if (!topo) {
+		pr_err("Failed to allocate NUMA topology\n");
+		return -ENOMEM;
+	}
+
+	spec = zalloc(topo->nr * sizeof(char *));
+	if (!spec) {
+		pr_err("Failed to allocate NUMA spec\n");
+		ret = -ENOMEM;
+		goto out_delete_topo;
+	}
+	for (s = 0; s < topo->nr; s++)
+		spec[s] = topo->nodes[s].cpus;
+
+	ret = record__init_thread_masks_spec(rec, cpus, spec, spec, topo->nr);
+
+	zfree(&spec);
+
+out_delete_topo:
+	numa_topology__delete(topo);
+
+	return ret;
+}
+
+static int record__init_thread_user_masks(struct record *rec, struct perf_cpu_map *cpus)
+{
+	int t, ret;
+	u32 s, nr_spec = 0;
+	char **maps_spec = NULL, **affinity_spec = NULL, **tmp_spec;
+	char *user_spec, *spec, *spec_ptr, *mask, *mask_ptr, *dup_mask = NULL;
+
+	for (t = 0, user_spec = (char *)rec->opts.threads_user_spec; ; t++, user_spec = NULL) {
+		spec = strtok_r(user_spec, ":", &spec_ptr);
+		if (spec == NULL)
+			break;
+		pr_debug2("threads_spec[%d]: %s\n", t, spec);
+		mask = strtok_r(spec, "/", &mask_ptr);
+		if (mask == NULL)
+			break;
+		pr_debug2("  maps mask: %s\n", mask);
+		tmp_spec = realloc(maps_spec, (nr_spec + 1) * sizeof(char *));
+		if (!tmp_spec) {
+			pr_err("Failed to reallocate maps spec\n");
+			ret = -ENOMEM;
+			goto out_free;
+		}
+		maps_spec = tmp_spec;
+		maps_spec[nr_spec] = dup_mask = strdup(mask);
+		if (!maps_spec[nr_spec]) {
+			pr_err("Failed to allocate maps spec[%d]\n", nr_spec);
+			ret = -ENOMEM;
+			goto out_free;
+		}
+		mask = strtok_r(NULL, "/", &mask_ptr);
+		if (mask == NULL) {
+			pr_err("Invalid thread maps or affinity specs\n");
+			ret = -EINVAL;
+			goto out_free;
+		}
+		pr_debug2("  affinity mask: %s\n", mask);
+		tmp_spec = realloc(affinity_spec, (nr_spec + 1) * sizeof(char *));
+		if (!tmp_spec) {
+			pr_err("Failed to reallocate affinity spec\n");
+			ret = -ENOMEM;
+			goto out_free;
+		}
+		affinity_spec = tmp_spec;
+		affinity_spec[nr_spec] = strdup(mask);
+		if (!affinity_spec[nr_spec]) {
+			pr_err("Failed to allocate affinity spec[%d]\n", nr_spec);
+			ret = -ENOMEM;
+			goto out_free;
+		}
+		dup_mask = NULL;
+		nr_spec++;
+	}
+
+	ret = record__init_thread_masks_spec(rec, cpus, (const char **)maps_spec,
+					     (const char **)affinity_spec, nr_spec);
+
+out_free:
+	free(dup_mask);
+	for (s = 0; s < nr_spec; s++) {
+		if (maps_spec)
+			free(maps_spec[s]);
+		if (affinity_spec)
+			free(affinity_spec[s]);
+	}
+	free(affinity_spec);
+	free(maps_spec);
+
+	return ret;
+}
+
 static int record__init_thread_default_masks(struct record *rec, struct perf_cpu_map *cpus)
 {
 	int ret;
···
 
 static int record__init_thread_masks(struct record *rec)
 {
+	int ret = 0;
 	struct perf_cpu_map *cpus = rec->evlist->core.cpus;
 
 	if (!record__threads_enabled(rec))
 		return record__init_thread_default_masks(rec, cpus);
 
-	return record__init_thread_cpu_masks(rec, cpus);
+	switch (rec->opts.threads_spec) {
+	case THREAD_SPEC__CPU:
+		ret = record__init_thread_cpu_masks(rec, cpus);
+		break;
+	case THREAD_SPEC__CORE:
+		ret = record__init_thread_core_masks(rec, cpus);
+		break;
+	case THREAD_SPEC__PACKAGE:
+		ret = record__init_thread_package_masks(rec, cpus);
+		break;
+	case THREAD_SPEC__NUMA:
+		ret = record__init_thread_numa_masks(rec, cpus);
+		break;
+	case THREAD_SPEC__USER:
+		ret = record__init_thread_user_masks(rec, cpus);
+		break;
+	default:
+		break;
+	}
+
+	return ret;
 }
 
 int cmd_record(int argc, const char **argv)

tools/perf/util/record.h

···
 	bool	      ctl_fd_close;
 	int	      synth;
 	int	      threads_spec;
+	const char    *threads_user_spec;
 };
 
 extern const char * const *record_usage;