[PATCH v2 1/3] perf record: Fix per-thread option.

[PATCH v2 1/3] perf record: Fix per-thread option.

[Ian Rogers]

From: Alexey Bayduraev <alexey.bayduraev@gmail.com>

Per-thread mode doesn't have specific CPUs for events, add checks for
this case.

Minor fix to a pr_debug by Ian Rogers <irogers@google.com> to avoid an
out of bound array access.

Reported-by: Ian Rogers <irogers@google.com>
Fixes: 7954f71689f9 ("perf record: Introduce thread affinity and mmap masks")
Signed-off-by: Ian Rogers <irogers@google.com>
Signed-off-by: Alexey Bayduraev <alexey.bayduraev@gmail.com>
---
 tools/perf/builtin-record.c | 22 +++++++++++++++++-----
 1 file changed, 17 insertions(+), 5 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index ba74fab02e62..069825c48d40 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -989,8 +989,11 @@ static int record__thread_data_init_maps(struct record_thread *thread_data, stru
 	struct mmap *overwrite_mmap = evlist->overwrite_mmap;
 	struct perf_cpu_map *cpus = evlist->core.user_requested_cpus;
 
-	thread_data->nr_mmaps = bitmap_weight(thread_data->mask->maps.bits,
-					      thread_data->mask->maps.nbits);
+	if (cpu_map__is_dummy(cpus))
+		thread_data->nr_mmaps = nr_mmaps;
+	else
+		thread_data->nr_mmaps = bitmap_weight(thread_data->mask->maps.bits,
+						      thread_data->mask->maps.nbits);
 	if (mmap) {
 		thread_data->maps = zalloc(thread_data->nr_mmaps * sizeof(struct mmap *));
 		if (!thread_data->maps)
@@ -1007,16 +1010,17 @@ static int record__thread_data_init_maps(struct record_thread *thread_data, stru
 		 thread_data->nr_mmaps, thread_data->maps, thread_data->overwrite_maps);
 
 	for (m = 0, tm = 0; m < nr_mmaps && tm < thread_data->nr_mmaps; m++) {
-		if (test_bit(cpus->map[m].cpu, thread_data->mask->maps.bits)) {
+		if (cpu_map__is_dummy(cpus) ||
+		    test_bit(cpus->map[m].cpu, thread_data->mask->maps.bits)) {
 			if (thread_data->maps) {
 				thread_data->maps[tm] = &mmap[m];
 				pr_debug2("thread_data[%p]: cpu%d: maps[%d] -> mmap[%d]\n",
-					  thread_data, cpus->map[m].cpu, tm, m);
+					  thread_data, perf_cpu_map__cpu(cpus, m).cpu, tm, m);
 			}
 			if (thread_data->overwrite_maps) {
 				thread_data->overwrite_maps[tm] = &overwrite_mmap[m];
 				pr_debug2("thread_data[%p]: cpu%d: ow_maps[%d] -> ow_mmap[%d]\n",
-					  thread_data, cpus->map[m].cpu, tm, m);
+					  thread_data, perf_cpu_map__cpu(cpus, m).cpu, tm, m);
 			}
 			tm++;
 		}
@@ -3329,6 +3333,9 @@ static void record__mmap_cpu_mask_init(struct mmap_cpu_mask *mask, struct perf_c
 {
 	int c;
 
+	if (cpu_map__is_dummy(cpus))
+		return;
+
 	for (c = 0; c < cpus->nr; c++)
 		set_bit(cpus->map[c].cpu, mask->bits);
 }
@@ -3680,6 +3687,11 @@ static int record__init_thread_masks(struct record *rec)
 	if (!record__threads_enabled(rec))
 		return record__init_thread_default_masks(rec, cpus);
 
+	if (cpu_map__is_dummy(cpus)) {
+		pr_err("--per-thread option is mutually exclusive to parallel streaming mode.\n");
+		return -EINVAL;
+	}
+
 	switch (rec->opts.threads_spec) {
 	case THREAD_SPEC__CPU:
 		ret = record__init_thread_cpu_masks(rec, cpus);
-- 
2.36.0.rc0.470.gd361397f0d-goog

[PATCH v2 2/3] perf cpumap: Switch to using perf_cpu_map API

[Ian Rogers]

Switch some raw accesses to the cpu map to using the library API. This
can help with reference count checking.

Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/builtin-record.c          | 13 ++++++------
 tools/perf/util/bpf_counter_cgroup.c | 31 ++++++++++++++--------------
 2 files changed, 22 insertions(+), 22 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 069825c48d40..a5cf6a99d67f 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1011,7 +1011,7 @@ static int record__thread_data_init_maps(struct record_thread *thread_data, stru
 
 	for (m = 0, tm = 0; m < nr_mmaps && tm < thread_data->nr_mmaps; m++) {
 		if (cpu_map__is_dummy(cpus) ||
-		    test_bit(cpus->map[m].cpu, thread_data->mask->maps.bits)) {
+		    test_bit(perf_cpu_map__cpu(cpus, m).cpu, thread_data->mask->maps.bits)) {
 			if (thread_data->maps) {
 				thread_data->maps[tm] = &mmap[m];
 				pr_debug2("thread_data[%p]: cpu%d: maps[%d] -> mmap[%d]\n",
@@ -3331,13 +3331,14 @@ struct option *record_options = __record_options;
 
 static void record__mmap_cpu_mask_init(struct mmap_cpu_mask *mask, struct perf_cpu_map *cpus)
 {
-	int c;
+	struct perf_cpu cpu;
+	int idx;
 
 	if (cpu_map__is_dummy(cpus))
 		return;
 
-	for (c = 0; c < cpus->nr; c++)
-		set_bit(cpus->map[c].cpu, mask->bits);
+	perf_cpu_map__for_each_cpu(cpu, idx, cpus)
+		set_bit(cpu.cpu, mask->bits);
 }
 
 static int record__mmap_cpu_mask_init_spec(struct mmap_cpu_mask *mask, const char *mask_spec)
@@ -3404,8 +3405,8 @@ static int record__init_thread_cpu_masks(struct record *rec, struct perf_cpu_map
 	pr_debug("nr_threads: %d\n", rec->nr_threads);
 
 	for (t = 0; t < rec->nr_threads; t++) {
-		set_bit(cpus->map[t].cpu, rec->thread_masks[t].maps.bits);
-		set_bit(cpus->map[t].cpu, rec->thread_masks[t].affinity.bits);
+		set_bit(perf_cpu_map__cpu(cpus, t).cpu, rec->thread_masks[t].maps.bits);
+		set_bit(perf_cpu_map__cpu(cpus, t).cpu, rec->thread_masks[t].affinity.bits);
 		if (verbose) {
 			pr_debug("thread_masks[%d]: ", t);
 			mmap_cpu_mask__scnprintf(&rec->thread_masks[t].maps, "maps");
diff --git a/tools/perf/util/bpf_counter_cgroup.c b/tools/perf/util/bpf_counter_cgroup.c
index ac60c08e8e2a..a4b676920da0 100644
--- a/tools/perf/util/bpf_counter_cgroup.c
+++ b/tools/perf/util/bpf_counter_cgroup.c
@@ -47,7 +47,7 @@ static int bperf_load_program(struct evlist *evlist)
 	struct evsel *evsel;
 	struct cgroup *cgrp, *leader_cgrp;
 	__u32 i, cpu;
-	__u32 nr_cpus = evlist->core.all_cpus->nr;
+	__u32 nr_cpus = perf_cpu_map__nr(evlist->core.all_cpus);
 	int total_cpus = cpu__max_cpu().cpu;
 	int map_size, map_fd;
 	int prog_fd, err;
@@ -125,7 +125,7 @@ static int bperf_load_program(struct evlist *evlist)
 			for (cpu = 0; cpu < nr_cpus; cpu++) {
 				int fd = FD(evsel, cpu);
 				__u32 idx = evsel->core.idx * total_cpus +
-					evlist->core.all_cpus->map[cpu].cpu;
+					perf_cpu_map__cpu(evlist->core.all_cpus, cpu).cpu;
 
 				err = bpf_map_update_elem(map_fd, &idx, &fd,
 							  BPF_ANY);
@@ -207,13 +207,13 @@ static int bperf_cgrp__install_pe(struct evsel *evsel __maybe_unused,
  */
 static int bperf_cgrp__sync_counters(struct evlist *evlist)
 {
-	int i, cpu;
-	int nr_cpus = evlist->core.all_cpus->nr;
+	struct perf_cpu cpu;
+	int idx;
 	int prog_fd = bpf_program__fd(skel->progs.trigger_read);
 
-	for (i = 0; i < nr_cpus; i++) {
-		cpu = evlist->core.all_cpus->map[i].cpu;
-		bperf_trigger_reading(prog_fd, cpu);
+	perf_cpu_map__for_each_cpu(cpu, idx, cpus) {
+		cpu = perf_cpu_map__cpu(evlist->core.all_cpus, i);
+		bperf_trigger_reading(prog_fd, cpu.cpu);
 	}
 
 	return 0;
@@ -244,12 +244,10 @@ static int bperf_cgrp__disable(struct evsel *evsel)
 static int bperf_cgrp__read(struct evsel *evsel)
 {
 	struct evlist *evlist = evsel->evlist;
-	int i, cpu, nr_cpus = evlist->core.all_cpus->nr;
 	int total_cpus = cpu__max_cpu().cpu;
 	struct perf_counts_values *counts;
 	struct bpf_perf_event_value *values;
 	int reading_map_fd, err = 0;
-	__u32 idx;
 
 	if (evsel->core.idx)
 		return 0;
@@ -263,7 +261,10 @@ static int bperf_cgrp__read(struct evsel *evsel)
 	reading_map_fd = bpf_map__fd(skel->maps.cgrp_readings);
 
 	evlist__for_each_entry(evlist, evsel) {
-		idx = evsel->core.idx;
+		__u32 idx = evsel->core.idx;
+		int i;
+		struct perf_cpu_map cpu;
+
 		err = bpf_map_lookup_elem(reading_map_fd, &idx, values);
 		if (err) {
 			pr_err("bpf map lookup failed: idx=%u, event=%s, cgrp=%s\n",
@@ -271,13 +272,11 @@ static int bperf_cgrp__read(struct evsel *evsel)
 			goto out;
 		}
 
-		for (i = 0; i < nr_cpus; i++) {
-			cpu = evlist->core.all_cpus->map[i].cpu;
-
+		perf_cpu_map__for_each_cpu(cpu, i, evlist->core.all_cpu) {
 			counts = perf_counts(evsel->counts, i, 0);
-			counts->val = values[cpu].counter;
-			counts->ena = values[cpu].enabled;
-			counts->run = values[cpu].running;
+			counts->val = values[cpu.cpu].counter;
+			counts->ena = values[cpu.cpu].enabled;
+			counts->run = values[cpu.cpu].running;
 		}
 	}
 
-- 
2.36.0.rc0.470.gd361397f0d-goog

Re: [PATCH v2 2/3] perf cpumap: Switch to using perf_cpu_map API

[Arnaldo Carvalho de Melo]

Em Wed, Apr 13, 2022 at 06:46:41PM -0700, Ian Rogers escreveu:
> Switch some raw accesses to the cpu map to using the library API. This
> can help with reference count checking.

Thanks, applying to perf/core after what is in perf/urgent, which this
patch depends on, gets merged there.

- Arnaldo
 
> Signed-off-by: Ian Rogers <irogers@google.com>
> ---
>  tools/perf/builtin-record.c          | 13 ++++++------
>  tools/perf/util/bpf_counter_cgroup.c | 31 ++++++++++++++--------------
>  2 files changed, 22 insertions(+), 22 deletions(-)
> 
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 069825c48d40..a5cf6a99d67f 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -1011,7 +1011,7 @@ static int record__thread_data_init_maps(struct record_thread *thread_data, stru
>  
>  	for (m = 0, tm = 0; m < nr_mmaps && tm < thread_data->nr_mmaps; m++) {
>  		if (cpu_map__is_dummy(cpus) ||
> -		    test_bit(cpus->map[m].cpu, thread_data->mask->maps.bits)) {
> +		    test_bit(perf_cpu_map__cpu(cpus, m).cpu, thread_data->mask->maps.bits)) {
>  			if (thread_data->maps) {
>  				thread_data->maps[tm] = &mmap[m];
>  				pr_debug2("thread_data[%p]: cpu%d: maps[%d] -> mmap[%d]\n",
> @@ -3331,13 +3331,14 @@ struct option *record_options = __record_options;
>  
>  static void record__mmap_cpu_mask_init(struct mmap_cpu_mask *mask, struct perf_cpu_map *cpus)
>  {
> -	int c;
> +	struct perf_cpu cpu;
> +	int idx;
>  
>  	if (cpu_map__is_dummy(cpus))
>  		return;
>  
> -	for (c = 0; c < cpus->nr; c++)
> -		set_bit(cpus->map[c].cpu, mask->bits);
> +	perf_cpu_map__for_each_cpu(cpu, idx, cpus)
> +		set_bit(cpu.cpu, mask->bits);
>  }
>  
>  static int record__mmap_cpu_mask_init_spec(struct mmap_cpu_mask *mask, const char *mask_spec)
> @@ -3404,8 +3405,8 @@ static int record__init_thread_cpu_masks(struct record *rec, struct perf_cpu_map
>  	pr_debug("nr_threads: %d\n", rec->nr_threads);
>  
>  	for (t = 0; t < rec->nr_threads; t++) {
> -		set_bit(cpus->map[t].cpu, rec->thread_masks[t].maps.bits);
> -		set_bit(cpus->map[t].cpu, rec->thread_masks[t].affinity.bits);
> +		set_bit(perf_cpu_map__cpu(cpus, t).cpu, rec->thread_masks[t].maps.bits);
> +		set_bit(perf_cpu_map__cpu(cpus, t).cpu, rec->thread_masks[t].affinity.bits);
>  		if (verbose) {
>  			pr_debug("thread_masks[%d]: ", t);
>  			mmap_cpu_mask__scnprintf(&rec->thread_masks[t].maps, "maps");
> diff --git a/tools/perf/util/bpf_counter_cgroup.c b/tools/perf/util/bpf_counter_cgroup.c
> index ac60c08e8e2a..a4b676920da0 100644
> --- a/tools/perf/util/bpf_counter_cgroup.c
> +++ b/tools/perf/util/bpf_counter_cgroup.c
> @@ -47,7 +47,7 @@ static int bperf_load_program(struct evlist *evlist)
>  	struct evsel *evsel;
>  	struct cgroup *cgrp, *leader_cgrp;
>  	__u32 i, cpu;
> -	__u32 nr_cpus = evlist->core.all_cpus->nr;
> +	__u32 nr_cpus = perf_cpu_map__nr(evlist->core.all_cpus);
>  	int total_cpus = cpu__max_cpu().cpu;
>  	int map_size, map_fd;
>  	int prog_fd, err;
> @@ -125,7 +125,7 @@ static int bperf_load_program(struct evlist *evlist)
>  			for (cpu = 0; cpu < nr_cpus; cpu++) {
>  				int fd = FD(evsel, cpu);
>  				__u32 idx = evsel->core.idx * total_cpus +
> -					evlist->core.all_cpus->map[cpu].cpu;
> +					perf_cpu_map__cpu(evlist->core.all_cpus, cpu).cpu;
>  
>  				err = bpf_map_update_elem(map_fd, &idx, &fd,
>  							  BPF_ANY);
> @@ -207,13 +207,13 @@ static int bperf_cgrp__install_pe(struct evsel *evsel __maybe_unused,
>   */
>  static int bperf_cgrp__sync_counters(struct evlist *evlist)
>  {
> -	int i, cpu;
> -	int nr_cpus = evlist->core.all_cpus->nr;
> +	struct perf_cpu cpu;
> +	int idx;
>  	int prog_fd = bpf_program__fd(skel->progs.trigger_read);
>  
> -	for (i = 0; i < nr_cpus; i++) {
> -		cpu = evlist->core.all_cpus->map[i].cpu;
> -		bperf_trigger_reading(prog_fd, cpu);
> +	perf_cpu_map__for_each_cpu(cpu, idx, cpus) {
> +		cpu = perf_cpu_map__cpu(evlist->core.all_cpus, i);
> +		bperf_trigger_reading(prog_fd, cpu.cpu);
>  	}
>  
>  	return 0;
> @@ -244,12 +244,10 @@ static int bperf_cgrp__disable(struct evsel *evsel)
>  static int bperf_cgrp__read(struct evsel *evsel)
>  {
>  	struct evlist *evlist = evsel->evlist;
> -	int i, cpu, nr_cpus = evlist->core.all_cpus->nr;
>  	int total_cpus = cpu__max_cpu().cpu;
>  	struct perf_counts_values *counts;
>  	struct bpf_perf_event_value *values;
>  	int reading_map_fd, err = 0;
> -	__u32 idx;
>  
>  	if (evsel->core.idx)
>  		return 0;
> @@ -263,7 +261,10 @@ static int bperf_cgrp__read(struct evsel *evsel)
>  	reading_map_fd = bpf_map__fd(skel->maps.cgrp_readings);
>  
>  	evlist__for_each_entry(evlist, evsel) {
> -		idx = evsel->core.idx;
> +		__u32 idx = evsel->core.idx;
> +		int i;
> +		struct perf_cpu_map cpu;
> +
>  		err = bpf_map_lookup_elem(reading_map_fd, &idx, values);
>  		if (err) {
>  			pr_err("bpf map lookup failed: idx=%u, event=%s, cgrp=%s\n",
> @@ -271,13 +272,11 @@ static int bperf_cgrp__read(struct evsel *evsel)
>  			goto out;
>  		}
>  
> -		for (i = 0; i < nr_cpus; i++) {
> -			cpu = evlist->core.all_cpus->map[i].cpu;
> -
> +		perf_cpu_map__for_each_cpu(cpu, i, evlist->core.all_cpu) {
>  			counts = perf_counts(evsel->counts, i, 0);
> -			counts->val = values[cpu].counter;
> -			counts->ena = values[cpu].enabled;
> -			counts->run = values[cpu].running;
> +			counts->val = values[cpu.cpu].counter;
> +			counts->ena = values[cpu.cpu].enabled;
> +			counts->run = values[cpu.cpu].running;
>  		}
>  	}
>  
> -- 
> 2.36.0.rc0.470.gd361397f0d-goog

-- 

- Arnaldo

[PATCH v2 3/3] perf test: Add basic perf record tests

[Ian Rogers]

Test the --per-thread flag.
Test Intel machine state capturing.

Suggested-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 tools/perf/tests/shell/record.sh | 42 ++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)
 create mode 100755 tools/perf/tests/shell/record.sh

diff --git a/tools/perf/tests/shell/record.sh b/tools/perf/tests/shell/record.sh
new file mode 100755
index 000000000000..cd1cf14259b8
--- /dev/null
+++ b/tools/perf/tests/shell/record.sh
@@ -0,0 +1,42 @@
+#!/bin/sh
+# perf record tests
+# SPDX-License-Identifier: GPL-2.0
+
+set -e
+
+err=0
+test_per_thread() {
+  echo "Basic --per-thread mode test"
+  perf record -e instructions:u --per-thread -o- true 2> /dev/null \
+    | perf report -i- -q \
+    | egrep -q true
+  echo "Basic --per-thread mode test [Success]"
+}
+
+test_register_capture() {
+  echo "Register capture test"
+  if ! perf list | egrep -q 'br_inst_retired.near_call'
+  then
+    echo "Register capture test [Skipped missing instruction]"
+    return
+  fi
+  if ! perf record --intr-regs=\? 2>&1 | egrep -q 'available registers: AX BX CX DX SI DI BP SP IP FLAGS CS SS R8 R9 R10 R11 R12 R13 R14 R15'
+  then
+    echo "Register capture test [Skipped missing registers]"
+    return
+  fi
+  if ! perf record -o - --intr-regs=di,r8,dx,cx -e cpu/br_inst_retired.near_call/p \
+    -c 1000 --per-thread true 2> /dev/null \
+    | perf script -F ip,sym,iregs -i - 2> /dev/null \
+    | egrep -q "DI:"
+  then
+    echo "Register capture test [Failed missing output]"
+    err=1
+    return
+  fi
+  echo "Register capture test [Success]"
+}
+
+test_per_thread
+test_register_capture
+exit $err
-- 
2.36.0.rc0.470.gd361397f0d-goog

Re: [PATCH v2 3/3] perf test: Add basic perf record tests

[Arnaldo Carvalho de Melo]

Em Wed, Apr 13, 2022 at 06:46:42PM -0700, Ian Rogers escreveu:
> Test the --per-thread flag.
> Test Intel machine state capturing.

Thanks, applied to perf/core.

- Arnaldo

 
> Suggested-by: Namhyung Kim <namhyung@kernel.org>
> Signed-off-by: Ian Rogers <irogers@google.com>
> ---
>  tools/perf/tests/shell/record.sh | 42 ++++++++++++++++++++++++++++++++
>  1 file changed, 42 insertions(+)
>  create mode 100755 tools/perf/tests/shell/record.sh
> 
> diff --git a/tools/perf/tests/shell/record.sh b/tools/perf/tests/shell/record.sh
> new file mode 100755
> index 000000000000..cd1cf14259b8
> --- /dev/null
> +++ b/tools/perf/tests/shell/record.sh
> @@ -0,0 +1,42 @@
> +#!/bin/sh
> +# perf record tests
> +# SPDX-License-Identifier: GPL-2.0
> +
> +set -e
> +
> +err=0
> +test_per_thread() {
> +  echo "Basic --per-thread mode test"
> +  perf record -e instructions:u --per-thread -o- true 2> /dev/null \
> +    | perf report -i- -q \
> +    | egrep -q true
> +  echo "Basic --per-thread mode test [Success]"
> +}
> +
> +test_register_capture() {
> +  echo "Register capture test"
> +  if ! perf list | egrep -q 'br_inst_retired.near_call'
> +  then
> +    echo "Register capture test [Skipped missing instruction]"
> +    return
> +  fi
> +  if ! perf record --intr-regs=\? 2>&1 | egrep -q 'available registers: AX BX CX DX SI DI BP SP IP FLAGS CS SS R8 R9 R10 R11 R12 R13 R14 R15'
> +  then
> +    echo "Register capture test [Skipped missing registers]"
> +    return
> +  fi
> +  if ! perf record -o - --intr-regs=di,r8,dx,cx -e cpu/br_inst_retired.near_call/p \
> +    -c 1000 --per-thread true 2> /dev/null \
> +    | perf script -F ip,sym,iregs -i - 2> /dev/null \
> +    | egrep -q "DI:"
> +  then
> +    echo "Register capture test [Failed missing output]"
> +    err=1
> +    return
> +  fi
> +  echo "Register capture test [Success]"
> +}
> +
> +test_per_thread
> +test_register_capture
> +exit $err
> -- 
> 2.36.0.rc0.470.gd361397f0d-goog

-- 

- Arnaldo

Re: [PATCH v2 1/3] perf record: Fix per-thread option.

[Arnaldo Carvalho de Melo]

Em Wed, Apr 13, 2022 at 06:46:40PM -0700, Ian Rogers escreveu:
> From: Alexey Bayduraev <alexey.bayduraev@gmail.com>
> 
> Per-thread mode doesn't have specific CPUs for events, add checks for
> this case.
> 
> Minor fix to a pr_debug by Ian Rogers <irogers@google.com> to avoid an
> out of bound array access.

Thanks, applied to perf/urgent.

- Arnaldo

 
> Reported-by: Ian Rogers <irogers@google.com>
> Fixes: 7954f71689f9 ("perf record: Introduce thread affinity and mmap masks")
> Signed-off-by: Ian Rogers <irogers@google.com>
> Signed-off-by: Alexey Bayduraev <alexey.bayduraev@gmail.com>
> ---
>  tools/perf/builtin-record.c | 22 +++++++++++++++++-----
>  1 file changed, 17 insertions(+), 5 deletions(-)
> 
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index ba74fab02e62..069825c48d40 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -989,8 +989,11 @@ static int record__thread_data_init_maps(struct record_thread *thread_data, stru
>  	struct mmap *overwrite_mmap = evlist->overwrite_mmap;
>  	struct perf_cpu_map *cpus = evlist->core.user_requested_cpus;
>  
> -	thread_data->nr_mmaps = bitmap_weight(thread_data->mask->maps.bits,
> -					      thread_data->mask->maps.nbits);
> +	if (cpu_map__is_dummy(cpus))
> +		thread_data->nr_mmaps = nr_mmaps;
> +	else
> +		thread_data->nr_mmaps = bitmap_weight(thread_data->mask->maps.bits,
> +						      thread_data->mask->maps.nbits);
>  	if (mmap) {
>  		thread_data->maps = zalloc(thread_data->nr_mmaps * sizeof(struct mmap *));
>  		if (!thread_data->maps)
> @@ -1007,16 +1010,17 @@ static int record__thread_data_init_maps(struct record_thread *thread_data, stru
>  		 thread_data->nr_mmaps, thread_data->maps, thread_data->overwrite_maps);
>  
>  	for (m = 0, tm = 0; m < nr_mmaps && tm < thread_data->nr_mmaps; m++) {
> -		if (test_bit(cpus->map[m].cpu, thread_data->mask->maps.bits)) {
> +		if (cpu_map__is_dummy(cpus) ||
> +		    test_bit(cpus->map[m].cpu, thread_data->mask->maps.bits)) {
>  			if (thread_data->maps) {
>  				thread_data->maps[tm] = &mmap[m];
>  				pr_debug2("thread_data[%p]: cpu%d: maps[%d] -> mmap[%d]\n",
> -					  thread_data, cpus->map[m].cpu, tm, m);
> +					  thread_data, perf_cpu_map__cpu(cpus, m).cpu, tm, m);
>  			}
>  			if (thread_data->overwrite_maps) {
>  				thread_data->overwrite_maps[tm] = &overwrite_mmap[m];
>  				pr_debug2("thread_data[%p]: cpu%d: ow_maps[%d] -> ow_mmap[%d]\n",
> -					  thread_data, cpus->map[m].cpu, tm, m);
> +					  thread_data, perf_cpu_map__cpu(cpus, m).cpu, tm, m);
>  			}
>  			tm++;
>  		}
> @@ -3329,6 +3333,9 @@ static void record__mmap_cpu_mask_init(struct mmap_cpu_mask *mask, struct perf_c
>  {
>  	int c;
>  
> +	if (cpu_map__is_dummy(cpus))
> +		return;
> +
>  	for (c = 0; c < cpus->nr; c++)
>  		set_bit(cpus->map[c].cpu, mask->bits);
>  }
> @@ -3680,6 +3687,11 @@ static int record__init_thread_masks(struct record *rec)
>  	if (!record__threads_enabled(rec))
>  		return record__init_thread_default_masks(rec, cpus);
>  
> +	if (cpu_map__is_dummy(cpus)) {
> +		pr_err("--per-thread option is mutually exclusive to parallel streaming mode.\n");
> +		return -EINVAL;
> +	}
> +
>  	switch (rec->opts.threads_spec) {
>  	case THREAD_SPEC__CPU:
>  		ret = record__init_thread_cpu_masks(rec, cpus);
> -- 
> 2.36.0.rc0.470.gd361397f0d-goog

-- 

- Arnaldo