2023-07-22 09:40:00

by Yang Jihong

[permalink] [raw]
Subject: [PATCH v3 0/7] perf record: Track sideband events for all CPUs when tracing selected CPUs

User space tasks can migrate between CPUs, track sideband events for all
CPUs.

The specific scenarios are as follows:

CPU0 CPU1
perf record -C 0 start
taskA starts to be created and executed
-> PERF_RECORD_COMM and PERF_RECORD_MMAP
events only deliver to CPU1
......
|
migrate to CPU0
|
Running on CPU0 <----------/
...

perf record -C 0 stop

Now perf samples the PC of taskA. However, perf does not record the
PERF_RECORD_COMM and PERF_RECORD_COMM events of taskA.
Therefore, the comm and symbols of taskA cannot be parsed.

The sys_perf_event_open invoked is as follows:

# perf --debug verbose=3 record -e cpu-clock -C 1 true
<SNIP>
Opening: cpu-clock
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0 (PERF_COUNT_SW_CPU_CLOCK)
{ sample_period, sample_freq } 4000
sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
read_format ID|LOST
disabled 1
inherit 1
freq 1
sample_id_all 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
Opening: dummy:u
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0x9 (PERF_COUNT_SW_DUMMY)
{ sample_period, sample_freq } 1
sample_type IP|TID|TIME|CPU|IDENTIFIER
read_format ID|LOST
inherit 1
exclude_kernel 1
exclude_hv 1
mmap 1
comm 1
task 1
sample_id_all 1
exclude_guest 1
mmap2 1
comm_exec 1
ksymbol 1
bpf_event 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
<SNIP>

Changes since_v2:
- Rename record_tracking.sh to record_sideband.sh in tools/perf/tests/shell.
- Remove "perf evlist: Skip dummy event sample_type check for evlist_config" patch.
- Add opts->all_kernel check in record__config_tracking_events().
- Add perf_event_attr test for record selected CPUs exclude_user.
- Update base-record & system-wide-dummy sample_type attr expected values for test-record-C0.

Changes since v1:
- Add perf_evlist__go_system_wide() via internal/evlist.h instead of
exporting perf_evlist__propagate_maps().
- Use evlist__add_aux_dummy() instead of evlist__add_dummy() in
evlist__findnew_tracking_event().
- Add a parameter in evlist__findnew_tracking_event() to deal with
system_wide inside.
- Add sideband for all CPUs when tracing selected CPUs comments on
the perf record man page.
- Use "sideband events" instead of "tracking events".
- Adjust the patches Sequence.
- Add patch5 to skip dummy event sample_type check for evlist_config.
- Add patch6 to update system-wide-dummy attr values for perf test.

Yang Jihong (7):
perf evlist: Add perf_evlist__go_system_wide() helper
perf evlist: Add evlist__findnew_tracking_event() helper
perf record: Move setting dummy tracking before
record__init_thread_masks()
perf record: Track sideband events for all CPUs when tracing selected
CPUs
perf test: Update base-record & system-wide-dummy attr expected values
for test-record-C0
perf test: Add test case for record sideband events
perf test: Add perf_event_attr test for record selected CPUs
exclude_user

tools/lib/perf/evlist.c | 9 +++
tools/lib/perf/include/internal/evlist.h | 2 +
tools/perf/Documentation/perf-record.txt | 3 +
tools/perf/builtin-record.c | 76 ++++++++++++-------
tools/perf/tests/attr/system-wide-dummy | 14 ++--
tools/perf/tests/attr/test-record-C0 | 4 +-
.../tests/attr/test-record-C0-exclude_user | 32 ++++++++
tools/perf/tests/shell/record_sideband.sh | 44 +++++++++++
tools/perf/util/evlist.c | 18 +++++
tools/perf/util/evlist.h | 1 +
10 files changed, 168 insertions(+), 35 deletions(-)
create mode 100644 tools/perf/tests/attr/test-record-C0-exclude_user
create mode 100755 tools/perf/tests/shell/record_sideband.sh

--
2.30.GIT



2023-07-22 09:40:57

by Yang Jihong

[permalink] [raw]
Subject: [PATCH v3 3/7] perf record: Move setting dummy tracking before record__init_thread_masks()

When dummy tracking go system wide, the mmap cpu mask is changed.
Therefore, needs to be placed before record__init_thread_masks().
Dummy tracking has been set in record__open(), move it before
record__init_thread_masks() and add a helper for unified processing.

The sys_perf_event_open invoked is as follows:

# perf --debug verbose=3 record -e cpu-clock -D 100 true
<SNIP>
Opening: cpu-clock
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0 (PERF_COUNT_SW_CPU_CLOCK)
{ sample_period, sample_freq } 4000
sample_type IP|TID|TIME|PERIOD|IDENTIFIER
read_format ID|LOST
disabled 1
inherit 1
freq 1
sample_id_all 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid 10318 cpu 0 group_fd -1 flags 0x8 = 5
sys_perf_event_open: pid 10318 cpu 1 group_fd -1 flags 0x8 = 6
sys_perf_event_open: pid 10318 cpu 2 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid 10318 cpu 3 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid 10318 cpu 4 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid 10318 cpu 5 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid 10318 cpu 6 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid 10318 cpu 7 group_fd -1 flags 0x8 = 13
Opening: dummy:u
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0x9 (PERF_COUNT_SW_DUMMY)
{ sample_period, sample_freq } 1
sample_type IP|TID|TIME|IDENTIFIER
read_format ID|LOST
disabled 1
inherit 1
exclude_kernel 1
exclude_hv 1
mmap 1
comm 1
enable_on_exec 1
task 1
sample_id_all 1
exclude_guest 1
mmap2 1
comm_exec 1
ksymbol 1
bpf_event 1
------------------------------------------------------------
sys_perf_event_open: pid 10318 cpu 0 group_fd -1 flags 0x8 = 14
sys_perf_event_open: pid 10318 cpu 1 group_fd -1 flags 0x8 = 15
sys_perf_event_open: pid 10318 cpu 2 group_fd -1 flags 0x8 = 16
sys_perf_event_open: pid 10318 cpu 3 group_fd -1 flags 0x8 = 17
sys_perf_event_open: pid 10318 cpu 4 group_fd -1 flags 0x8 = 18
sys_perf_event_open: pid 10318 cpu 5 group_fd -1 flags 0x8 = 19
sys_perf_event_open: pid 10318 cpu 6 group_fd -1 flags 0x8 = 20
sys_perf_event_open: pid 10318 cpu 7 group_fd -1 flags 0x8 = 21
<SNIP>

Signed-off-by: Yang Jihong <[email protected]>
---
tools/perf/builtin-record.c | 59 +++++++++++++++++++++++--------------
1 file changed, 37 insertions(+), 22 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index ca83599cc50c..3ff9d972225e 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -908,6 +908,37 @@ static int record__config_off_cpu(struct record *rec)
return off_cpu_prepare(rec->evlist, &rec->opts.target, &rec->opts);
}

+static int record__config_tracking_events(struct record *rec)
+{
+ struct record_opts *opts = &rec->opts;
+ struct evlist *evlist = rec->evlist;
+ struct evsel *evsel;
+
+ /*
+ * For initial_delay, system wide or a hybrid system, we need to add a
+ * dummy event so that we can track PERF_RECORD_MMAP to cover the delay
+ * of waiting or event synthesis.
+ */
+ if (opts->target.initial_delay || target__has_cpu(&opts->target) ||
+ perf_pmus__num_core_pmus() > 1) {
+ evsel = evlist__findnew_tracking_event(evlist, false);
+ if (!evsel)
+ return -ENOMEM;
+
+ /*
+ * Enable the dummy event when the process is forked for
+ * initial_delay, immediately for system wide.
+ */
+ if (opts->target.initial_delay && !evsel->immediate &&
+ !target__has_cpu(&opts->target))
+ evsel->core.attr.enable_on_exec = 1;
+ else
+ evsel->immediate = 1;
+ }
+
+ return 0;
+}
+
static bool record__kcore_readable(struct machine *machine)
{
char kcore[PATH_MAX];
@@ -1288,28 +1319,6 @@ static int record__open(struct record *rec)
struct record_opts *opts = &rec->opts;
int rc = 0;

- /*
- * For initial_delay, system wide or a hybrid system, we need to add a
- * dummy event so that we can track PERF_RECORD_MMAP to cover the delay
- * of waiting or event synthesis.
- */
- if (opts->target.initial_delay || target__has_cpu(&opts->target) ||
- perf_pmus__num_core_pmus() > 1) {
- pos = evlist__findnew_tracking_event(evlist, false);
- if (!pos)
- return -ENOMEM;
-
- /*
- * Enable the dummy event when the process is forked for
- * initial_delay, immediately for system wide.
- */
- if (opts->target.initial_delay && !pos->immediate &&
- !target__has_cpu(&opts->target))
- pos->core.attr.enable_on_exec = 1;
- else
- pos->immediate = 1;
- }
-
evlist__config(evlist, opts, &callchain_param);

evlist__for_each_entry(evlist, pos) {
@@ -4235,6 +4244,12 @@ int cmd_record(int argc, const char **argv)
goto out;
}

+ err = record__config_tracking_events(rec);
+ if (err) {
+ pr_err("record__config_tracking_events failed, error %d\n", err);
+ goto out;
+ }
+
err = record__init_thread_masks(rec);
if (err) {
pr_err("Failed to initialize parallel data streaming masks\n");
--
2.30.GIT


2023-07-22 09:52:54

by Yang Jihong

[permalink] [raw]
Subject: [PATCH v3 7/7] perf test: Add perf_event_attr test for record selected CPUs exclude_user

If all (non-dummy) evsel have exclude_user, system_wide sideband is not
needed. Add this test scenario.

Signed-off-by: Yang Jihong <[email protected]>
---
.../tests/attr/test-record-C0-exclude_user | 32 +++++++++++++++++++
1 file changed, 32 insertions(+)
create mode 100644 tools/perf/tests/attr/test-record-C0-exclude_user

diff --git a/tools/perf/tests/attr/test-record-C0-exclude_user b/tools/perf/tests/attr/test-record-C0-exclude_user
new file mode 100644
index 000000000000..2d7549277c1e
--- /dev/null
+++ b/tools/perf/tests/attr/test-record-C0-exclude_user
@@ -0,0 +1,32 @@
+[config]
+command = record
+args = --no-bpf-event --all-kernel -C 0 kill >/dev/null 2>&1
+ret = 1
+
+[event:base-record]
+cpu=0
+
+# no enable on exec for CPU attached
+enable_on_exec=0
+
+# PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_TIME |
+# PERF_SAMPLE_PERIOD | PERF_SAMPLE_IDENTIFIER
+# + PERF_SAMPLE_CPU added by -C 0
+sample_type=65927
+
+# Dummy event handles mmaps, comm and task.
+mmap=0
+comm=0
+task=0
+
+# exclude_user for all-kernel option
+exclude_user=1
+
+[event:system-wide-dummy]
+
+# system_wide is not need for all (non-dummy) events have exclude_user
+cpu=0
+
+# exclude_user for all-kernel option
+exclude_user=1
+exclude_kernel=0
--
2.30.GIT


2023-07-22 10:24:06

by Yang Jihong

[permalink] [raw]
Subject: [PATCH v3 2/7] perf evlist: Add evlist__findnew_tracking_event() helper

Currently, intel-bts, intel-pt, and arm-spe may add a dummy event for
tracking to the evlist. We may need to search for the dummy event for
some settings. Therefore, add evlist__findnew_tracking_event() helper.

evlist__findnew_tracking_event() also deal with system_wide maps if
system_wide is true.

Signed-off-by: Yang Jihong <[email protected]>
---
tools/perf/builtin-record.c | 11 +++--------
tools/perf/util/evlist.c | 18 ++++++++++++++++++
tools/perf/util/evlist.h | 1 +
3 files changed, 22 insertions(+), 8 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index aec18db7ff23..ca83599cc50c 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -1295,14 +1295,9 @@ static int record__open(struct record *rec)
*/
if (opts->target.initial_delay || target__has_cpu(&opts->target) ||
perf_pmus__num_core_pmus() > 1) {
- pos = evlist__get_tracking_event(evlist);
- if (!evsel__is_dummy_event(pos)) {
- /* Set up dummy event. */
- if (evlist__add_dummy(evlist))
- return -ENOMEM;
- pos = evlist__last(evlist);
- evlist__set_tracking_event(evlist, pos);
- }
+ pos = evlist__findnew_tracking_event(evlist, false);
+ if (!pos)
+ return -ENOMEM;

/*
* Enable the dummy event when the process is forked for
diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
index 7ef43f72098e..25c3ebe2c2f5 100644
--- a/tools/perf/util/evlist.c
+++ b/tools/perf/util/evlist.c
@@ -1694,6 +1694,24 @@ void evlist__set_tracking_event(struct evlist *evlist, struct evsel *tracking_ev
tracking_evsel->tracking = true;
}

+struct evsel *evlist__findnew_tracking_event(struct evlist *evlist, bool system_wide)
+{
+ struct evsel *evsel;
+
+ evsel = evlist__get_tracking_event(evlist);
+ if (!evsel__is_dummy_event(evsel)) {
+ evsel = evlist__add_aux_dummy(evlist, system_wide);
+ if (!evsel)
+ return NULL;
+
+ evlist__set_tracking_event(evlist, evsel);
+ } else if (system_wide) {
+ perf_evlist__go_system_wide(&evlist->core, &evsel->core);
+ }
+
+ return evsel;
+}
+
struct evsel *evlist__find_evsel_by_str(struct evlist *evlist, const char *str)
{
struct evsel *evsel;
diff --git a/tools/perf/util/evlist.h b/tools/perf/util/evlist.h
index 664c6bf7b3e0..98e7ddb2bd30 100644
--- a/tools/perf/util/evlist.h
+++ b/tools/perf/util/evlist.h
@@ -387,6 +387,7 @@ bool evlist_cpu_iterator__end(const struct evlist_cpu_iterator *evlist_cpu_itr);

struct evsel *evlist__get_tracking_event(struct evlist *evlist);
void evlist__set_tracking_event(struct evlist *evlist, struct evsel *tracking_evsel);
+struct evsel *evlist__findnew_tracking_event(struct evlist *evlist, bool system_wide);

struct evsel *evlist__find_evsel_by_str(struct evlist *evlist, const char *str);

--
2.30.GIT


2023-07-22 10:28:21

by Yang Jihong

[permalink] [raw]
Subject: [PATCH v3 6/7] perf test: Add test case for record sideband events

Add a new test case to record sideband events for all CPUs when tracing
selected CPUs

Test result:

# ./perf test list 2>&1 | grep 'perf record sideband tests'
95: perf record sideband tests
# ./perf test 95
95: perf record sideband tests : Ok

Signed-off-by: Yang Jihong <[email protected]>
---
tools/perf/tests/shell/record_sideband.sh | 44 +++++++++++++++++++++++
1 file changed, 44 insertions(+)
create mode 100755 tools/perf/tests/shell/record_sideband.sh

diff --git a/tools/perf/tests/shell/record_sideband.sh b/tools/perf/tests/shell/record_sideband.sh
new file mode 100755
index 000000000000..2ecf00011cb1
--- /dev/null
+++ b/tools/perf/tests/shell/record_sideband.sh
@@ -0,0 +1,44 @@
+#!/bin/sh
+# perf record sideband tests
+# SPDX-License-Identifier: GPL-2.0
+
+set -e
+
+err=0
+perfdata=$(mktemp /tmp/__perf_test.perf.data.XXXXX)
+
+can_cpu_wide()
+{
+ if ! perf record -o ${perfdata} -BN --no-bpf-event -C $1 true 2>&1 >/dev/null
+ then
+ echo "record sideband test [Skipped cannot record cpu$1]"
+ err=2
+ fi
+
+ rm -f ${perfdata}
+ return $err
+}
+
+test_system_wide_tracking()
+{
+ # Need CPU 0 and CPU 1
+ can_cpu_wide 0 || return 0
+ can_cpu_wide 1 || return 0
+
+ # Record on CPU 0 a task running on CPU 1
+ perf record -BN --no-bpf-event -o ${perfdata} -C 0 -- taskset --cpu-list 1 true
+
+ # Should get MMAP events from CPU 1
+ mmap_cnt=`perf script -i ${perfdata} --show-mmap-events -C 1 2>/dev/null | grep MMAP | wc -l`
+
+ rm -f ${perfdata}
+
+ if [ ${mmap_cnt} -gt 0 ] ; then
+ return 0
+ fi
+
+ echo "Failed to record MMAP events on CPU 1 when tracing CPU 0"
+ return 1
+}
+
+test_system_wide_tracking
--
2.30.GIT


2023-07-22 10:29:02

by Yang Jihong

[permalink] [raw]
Subject: [PATCH v3 1/7] perf evlist: Add perf_evlist__go_system_wide() helper

For dummy events that keep tracking, we may need to modify its cpu_maps.
For example, change the cpu_maps to record sideband events for all CPUS.
Add perf_evlist__go_system_wide() helper to support this scenario.

Signed-off-by: Yang Jihong <[email protected]>
---
tools/lib/perf/evlist.c | 9 +++++++++
tools/lib/perf/include/internal/evlist.h | 2 ++
2 files changed, 11 insertions(+)

diff --git a/tools/lib/perf/evlist.c b/tools/lib/perf/evlist.c
index b8b066d0dc5e..3acbbccc1901 100644
--- a/tools/lib/perf/evlist.c
+++ b/tools/lib/perf/evlist.c
@@ -738,3 +738,12 @@ int perf_evlist__nr_groups(struct perf_evlist *evlist)
}
return nr_groups;
}
+
+void perf_evlist__go_system_wide(struct perf_evlist *evlist, struct perf_evsel *evsel)
+{
+ if (!evsel->system_wide) {
+ evsel->system_wide = true;
+ if (evlist->needs_map_propagation)
+ __perf_evlist__propagate_maps(evlist, evsel);
+ }
+}
diff --git a/tools/lib/perf/include/internal/evlist.h b/tools/lib/perf/include/internal/evlist.h
index 3339bc2f1765..d86ffe8ed483 100644
--- a/tools/lib/perf/include/internal/evlist.h
+++ b/tools/lib/perf/include/internal/evlist.h
@@ -135,4 +135,6 @@ int perf_evlist__id_add_fd(struct perf_evlist *evlist,
void perf_evlist__reset_id_hash(struct perf_evlist *evlist);

void __perf_evlist__set_leader(struct list_head *list, struct perf_evsel *leader);
+
+void perf_evlist__go_system_wide(struct perf_evlist *evlist, struct perf_evsel *evsel);
#endif /* __LIBPERF_INTERNAL_EVLIST_H */
--
2.30.GIT


2023-07-22 10:30:50

by Yang Jihong

[permalink] [raw]
Subject: [PATCH v3 4/7] perf record: Track sideband events for all CPUs when tracing selected CPUs

User space tasks can migrate between CPUs, we need to track side-band
events for all CPUs.

The specific scenarios are as follows:

CPU0 CPU1
perf record -C 0 start
taskA starts to be created and executed
-> PERF_RECORD_COMM and PERF_RECORD_MMAP
events only deliver to CPU1
......
|
migrate to CPU0
|
Running on CPU0 <----------/
...

perf record -C 0 stop

Now perf samples the PC of taskA. However, perf does not record the
PERF_RECORD_COMM and PERF_RECORD_MMAP events of taskA.
Therefore, the comm and symbols of taskA cannot be parsed.

The solution is to record sideband events for all CPUs when tracing
selected CPUs. Because this modifies the default behavior, add related
comments to the perf record man page.

The sys_perf_event_open invoked is as follows:

# perf --debug verbose=3 record -e cpu-clock -C 1 true
<SNIP>
Opening: cpu-clock
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0 (PERF_COUNT_SW_CPU_CLOCK)
{ sample_period, sample_freq } 4000
sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
read_format ID|LOST
disabled 1
inherit 1
freq 1
sample_id_all 1
exclude_guest 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
Opening: dummy:u
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0x9 (PERF_COUNT_SW_DUMMY)
{ sample_period, sample_freq } 1
sample_type IP|TID|TIME|CPU|IDENTIFIER
read_format ID|LOST
inherit 1
exclude_kernel 1
exclude_hv 1
mmap 1
comm 1
task 1
sample_id_all 1
exclude_guest 1
mmap2 1
comm_exec 1
ksymbol 1
bpf_event 1
------------------------------------------------------------
sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
<SNIP>

Signed-off-by: Yang Jihong <[email protected]>
---
tools/perf/Documentation/perf-record.txt | 3 +++
tools/perf/builtin-record.c | 14 +++++++++++++-
2 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index 680396c56bd1..dac53ece51ab 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -388,6 +388,9 @@ comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-
In per-thread mode with inheritance mode on (default), samples are captured only when
the thread executes on the designated CPUs. Default is to monitor all CPUs.

+User space tasks can migrate between CPUs, so when tracing selected CPUs,
+a dummy event is created to track sideband for all CPUs.
+
-B::
--no-buildid::
Do not save the build ids of binaries in the perf.data files. This skips
diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 3ff9d972225e..4e8e97928f05 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -912,6 +912,7 @@ static int record__config_tracking_events(struct record *rec)
{
struct record_opts *opts = &rec->opts;
struct evlist *evlist = rec->evlist;
+ bool system_wide = false;
struct evsel *evsel;

/*
@@ -921,7 +922,18 @@ static int record__config_tracking_events(struct record *rec)
*/
if (opts->target.initial_delay || target__has_cpu(&opts->target) ||
perf_pmus__num_core_pmus() > 1) {
- evsel = evlist__findnew_tracking_event(evlist, false);
+
+ /*
+ * User space tasks can migrate between CPUs, so when tracing
+ * selected CPUs, sideband for all CPUs is still needed.
+ *
+ * If all (non-dummy) evsel have exclude_user,
+ * system_wide is not needed.
+ */
+ if (!!opts->target.cpu_list && !opts->all_kernel)
+ system_wide = true;
+
+ evsel = evlist__findnew_tracking_event(evlist, system_wide);
if (!evsel)
return -ENOMEM;

--
2.30.GIT


2023-07-22 10:32:48

by Yang Jihong

[permalink] [raw]
Subject: [PATCH v3 5/7] perf test: Update base-record & system-wide-dummy attr expected values for test-record-C0

1. Because a dummy sideband event is added to the sampling of specified
CPUs. When evlist contains evsel of different sample_type,
evlist__config() will change the default PERF_SAMPLE_ID bit to
PERF_SAMPLE_IDENTIFICATION bit.
The attr sample_type expected value of base-record and system-wide-dummy
in test-record-C0 needs to be updated.

2. The perf record uses evlist__add_aux_dummy() instead of
evlist__add_dummy() to add a dummy event.
The expected value of system-wide-dummy attr needs to be updated.

The perf test result is as follows:

# ./perf test list 2>&1 | grep 'Setup struct perf_event_attr'
17: Setup struct perf_event_attr
# ./perf test 17
17: Setup struct perf_event_attr : Ok

Signed-off-by: Yang Jihong <[email protected]>
---
tools/perf/tests/attr/system-wide-dummy | 14 ++++++++------
tools/perf/tests/attr/test-record-C0 | 4 ++--
2 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/tools/perf/tests/attr/system-wide-dummy b/tools/perf/tests/attr/system-wide-dummy
index 2f3e3eb728eb..a1e1d6a263bf 100644
--- a/tools/perf/tests/attr/system-wide-dummy
+++ b/tools/perf/tests/attr/system-wide-dummy
@@ -9,8 +9,10 @@ flags=8
type=1
size=136
config=9
-sample_period=4000
-sample_type=455
+sample_period=1
+# PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_TIME |
+# PERF_SAMPLE_CPU | PERF_SAMPLE_IDENTIFIER
+sample_type=65671
read_format=4|20
# Event will be enabled right away.
disabled=0
@@ -18,12 +20,12 @@ inherit=1
pinned=0
exclusive=0
exclude_user=0
-exclude_kernel=0
-exclude_hv=0
+exclude_kernel=1
+exclude_hv=1
exclude_idle=0
mmap=1
comm=1
-freq=1
+freq=0
inherit_stat=0
enable_on_exec=0
task=1
@@ -32,7 +34,7 @@ precise_ip=0
mmap_data=0
sample_id_all=1
exclude_host=0
-exclude_guest=0
+exclude_guest=1
exclude_callchain_kernel=0
exclude_callchain_user=0
mmap2=1
diff --git a/tools/perf/tests/attr/test-record-C0 b/tools/perf/tests/attr/test-record-C0
index 317730b906dd..198e8429a1bf 100644
--- a/tools/perf/tests/attr/test-record-C0
+++ b/tools/perf/tests/attr/test-record-C0
@@ -10,9 +10,9 @@ cpu=0
enable_on_exec=0

# PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_TIME |
-# PERF_SAMPLE_ID | PERF_SAMPLE_PERIOD
+# PERF_SAMPLE_PERIOD | PERF_SAMPLE_IDENTIFIER
# + PERF_SAMPLE_CPU added by -C 0
-sample_type=455
+sample_type=65927

# Dummy event handles mmaps, comm and task.
mmap=0
--
2.30.GIT


2023-07-31 11:35:39

by Adrian Hunter

[permalink] [raw]
Subject: Re: [PATCH v3 4/7] perf record: Track sideband events for all CPUs when tracing selected CPUs

On 22/07/23 12:32, Yang Jihong wrote:
> User space tasks can migrate between CPUs, we need to track side-band
> events for all CPUs.
>
> The specific scenarios are as follows:
>
> CPU0 CPU1
> perf record -C 0 start
> taskA starts to be created and executed
> -> PERF_RECORD_COMM and PERF_RECORD_MMAP
> events only deliver to CPU1
> ......
> |
> migrate to CPU0
> |
> Running on CPU0 <----------/
> ...
>
> perf record -C 0 stop
>
> Now perf samples the PC of taskA. However, perf does not record the
> PERF_RECORD_COMM and PERF_RECORD_MMAP events of taskA.
> Therefore, the comm and symbols of taskA cannot be parsed.
>
> The solution is to record sideband events for all CPUs when tracing
> selected CPUs. Because this modifies the default behavior, add related
> comments to the perf record man page.
>
> The sys_perf_event_open invoked is as follows:
>
> # perf --debug verbose=3 record -e cpu-clock -C 1 true
> <SNIP>
> Opening: cpu-clock
> ------------------------------------------------------------
> perf_event_attr:
> type 1 (PERF_TYPE_SOFTWARE)
> size 136
> config 0 (PERF_COUNT_SW_CPU_CLOCK)
> { sample_period, sample_freq } 4000
> sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
> read_format ID|LOST
> disabled 1
> inherit 1
> freq 1
> sample_id_all 1
> exclude_guest 1
> ------------------------------------------------------------
> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
> Opening: dummy:u
> ------------------------------------------------------------
> perf_event_attr:
> type 1 (PERF_TYPE_SOFTWARE)
> size 136
> config 0x9 (PERF_COUNT_SW_DUMMY)
> { sample_period, sample_freq } 1
> sample_type IP|TID|TIME|CPU|IDENTIFIER
> read_format ID|LOST
> inherit 1
> exclude_kernel 1
> exclude_hv 1
> mmap 1
> comm 1
> task 1
> sample_id_all 1
> exclude_guest 1
> mmap2 1
> comm_exec 1
> ksymbol 1
> bpf_event 1
> ------------------------------------------------------------
> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
> sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
> sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
> sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
> sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
> <SNIP>
>
> Signed-off-by: Yang Jihong <[email protected]>
> ---
> tools/perf/Documentation/perf-record.txt | 3 +++
> tools/perf/builtin-record.c | 14 +++++++++++++-
> 2 files changed, 16 insertions(+), 1 deletion(-)
>
> diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
> index 680396c56bd1..dac53ece51ab 100644
> --- a/tools/perf/Documentation/perf-record.txt
> +++ b/tools/perf/Documentation/perf-record.txt
> @@ -388,6 +388,9 @@ comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-
> In per-thread mode with inheritance mode on (default), samples are captured only when
> the thread executes on the designated CPUs. Default is to monitor all CPUs.
>
> +User space tasks can migrate between CPUs, so when tracing selected CPUs,
> +a dummy event is created to track sideband for all CPUs.
> +
> -B::
> --no-buildid::
> Do not save the build ids of binaries in the perf.data files. This skips
> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> index 3ff9d972225e..4e8e97928f05 100644
> --- a/tools/perf/builtin-record.c
> +++ b/tools/perf/builtin-record.c
> @@ -912,6 +912,7 @@ static int record__config_tracking_events(struct record *rec)
> {
> struct record_opts *opts = &rec->opts;
> struct evlist *evlist = rec->evlist;
> + bool system_wide = false;
> struct evsel *evsel;
>
> /*
> @@ -921,7 +922,18 @@ static int record__config_tracking_events(struct record *rec)
> */
> if (opts->target.initial_delay || target__has_cpu(&opts->target) ||
> perf_pmus__num_core_pmus() > 1) {
> - evsel = evlist__findnew_tracking_event(evlist, false);
> +
> + /*
> + * User space tasks can migrate between CPUs, so when tracing
> + * selected CPUs, sideband for all CPUs is still needed.
> + *
> + * If all (non-dummy) evsel have exclude_user,
> + * system_wide is not needed.
> + */
> + if (!!opts->target.cpu_list && !opts->all_kernel)

Not everyone uses all-kernel. Can we check the evsels are either dummy
or exclude_user?

> + system_wide = true;
> +
> + evsel = evlist__findnew_tracking_event(evlist, system_wide);
> if (!evsel)
> return -ENOMEM;
>


2023-07-31 13:06:02

by Yang Jihong

[permalink] [raw]
Subject: Re: [PATCH v3 4/7] perf record: Track sideband events for all CPUs when tracing selected CPUs

Hello,

On 2023/7/31 19:08, Adrian Hunter wrote:
> On 22/07/23 12:32, Yang Jihong wrote:
>> User space tasks can migrate between CPUs, we need to track side-band
>> events for all CPUs.
>>
>> The specific scenarios are as follows:
>>
>> CPU0 CPU1
>> perf record -C 0 start
>> taskA starts to be created and executed
>> -> PERF_RECORD_COMM and PERF_RECORD_MMAP
>> events only deliver to CPU1
>> ......
>> |
>> migrate to CPU0
>> |
>> Running on CPU0 <----------/
>> ...
>>
>> perf record -C 0 stop
>>
>> Now perf samples the PC of taskA. However, perf does not record the
>> PERF_RECORD_COMM and PERF_RECORD_MMAP events of taskA.
>> Therefore, the comm and symbols of taskA cannot be parsed.
>>
>> The solution is to record sideband events for all CPUs when tracing
>> selected CPUs. Because this modifies the default behavior, add related
>> comments to the perf record man page.
>>
>> The sys_perf_event_open invoked is as follows:
>>
>> # perf --debug verbose=3 record -e cpu-clock -C 1 true
>> <SNIP>
>> Opening: cpu-clock
>> ------------------------------------------------------------
>> perf_event_attr:
>> type 1 (PERF_TYPE_SOFTWARE)
>> size 136
>> config 0 (PERF_COUNT_SW_CPU_CLOCK)
>> { sample_period, sample_freq } 4000
>> sample_type IP|TID|TIME|CPU|PERIOD|IDENTIFIER
>> read_format ID|LOST
>> disabled 1
>> inherit 1
>> freq 1
>> sample_id_all 1
>> exclude_guest 1
>> ------------------------------------------------------------
>> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 5
>> Opening: dummy:u
>> ------------------------------------------------------------
>> perf_event_attr:
>> type 1 (PERF_TYPE_SOFTWARE)
>> size 136
>> config 0x9 (PERF_COUNT_SW_DUMMY)
>> { sample_period, sample_freq } 1
>> sample_type IP|TID|TIME|CPU|IDENTIFIER
>> read_format ID|LOST
>> inherit 1
>> exclude_kernel 1
>> exclude_hv 1
>> mmap 1
>> comm 1
>> task 1
>> sample_id_all 1
>> exclude_guest 1
>> mmap2 1
>> comm_exec 1
>> ksymbol 1
>> bpf_event 1
>> ------------------------------------------------------------
>> sys_perf_event_open: pid -1 cpu 0 group_fd -1 flags 0x8 = 6
>> sys_perf_event_open: pid -1 cpu 1 group_fd -1 flags 0x8 = 7
>> sys_perf_event_open: pid -1 cpu 2 group_fd -1 flags 0x8 = 9
>> sys_perf_event_open: pid -1 cpu 3 group_fd -1 flags 0x8 = 10
>> sys_perf_event_open: pid -1 cpu 4 group_fd -1 flags 0x8 = 11
>> sys_perf_event_open: pid -1 cpu 5 group_fd -1 flags 0x8 = 12
>> sys_perf_event_open: pid -1 cpu 6 group_fd -1 flags 0x8 = 13
>> sys_perf_event_open: pid -1 cpu 7 group_fd -1 flags 0x8 = 14
>> <SNIP>
>>
>> Signed-off-by: Yang Jihong <[email protected]>
>> ---
>> tools/perf/Documentation/perf-record.txt | 3 +++
>> tools/perf/builtin-record.c | 14 +++++++++++++-
>> 2 files changed, 16 insertions(+), 1 deletion(-)
>>
>> diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
>> index 680396c56bd1..dac53ece51ab 100644
>> --- a/tools/perf/Documentation/perf-record.txt
>> +++ b/tools/perf/Documentation/perf-record.txt
>> @@ -388,6 +388,9 @@ comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-
>> In per-thread mode with inheritance mode on (default), samples are captured only when
>> the thread executes on the designated CPUs. Default is to monitor all CPUs.
>>
>> +User space tasks can migrate between CPUs, so when tracing selected CPUs,
>> +a dummy event is created to track sideband for all CPUs.
>> +
>> -B::
>> --no-buildid::
>> Do not save the build ids of binaries in the perf.data files. This skips
>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>> index 3ff9d972225e..4e8e97928f05 100644
>> --- a/tools/perf/builtin-record.c
>> +++ b/tools/perf/builtin-record.c
>> @@ -912,6 +912,7 @@ static int record__config_tracking_events(struct record *rec)
>> {
>> struct record_opts *opts = &rec->opts;
>> struct evlist *evlist = rec->evlist;
>> + bool system_wide = false;
>> struct evsel *evsel;
>>
>> /*
>> @@ -921,7 +922,18 @@ static int record__config_tracking_events(struct record *rec)
>> */
>> if (opts->target.initial_delay || target__has_cpu(&opts->target) ||
>> perf_pmus__num_core_pmus() > 1) {
>> - evsel = evlist__findnew_tracking_event(evlist, false);
>> +
>> + /*
>> + * User space tasks can migrate between CPUs, so when tracing
>> + * selected CPUs, sideband for all CPUs is still needed.
>> + *
>> + * If all (non-dummy) evsel have exclude_user,
>> + * system_wide is not needed.
>> + */
>> + if (!!opts->target.cpu_list && !opts->all_kernel)
>
> Not everyone uses all-kernel. Can we check the evsels are either dummy
> or exclude_user?
For perf_record, exclude_user of all evsels is set in evsel__config(),
and record__config_tracking_events() is before evsel__config().

Uh..., it seems that only opts->all_kernel can be used to check
exclude_user of evsels.

void evsel__config()
{
...
if (opts->all_kernel) {
attr->exclude_kernel = 0;
attr->exclude_user = 1;
}
...
}

Thanks,
Yang

2023-07-31 13:24:09

by Adrian Hunter

[permalink] [raw]
Subject: Re: [PATCH v3 4/7] perf record: Track sideband events for all CPUs when tracing selected CPUs

On 31/07/23 15:38, Yang Jihong wrote:
> Hello,
>
> On 2023/7/31 19:08, Adrian Hunter wrote:
>> On 22/07/23 12:32, Yang Jihong wrote:
>>> User space tasks can migrate between CPUs, we need to track side-band
>>> events for all CPUs.
>>>
>>> The specific scenarios are as follows:
>>>
>>>           CPU0                                 CPU1
>>>    perf record -C 0 start
>>>                                taskA starts to be created and executed
>>>                                  -> PERF_RECORD_COMM and PERF_RECORD_MMAP
>>>                                     events only deliver to CPU1
>>>                                ......
>>>                                  |
>>>                            migrate to CPU0
>>>                                  |
>>>    Running on CPU0    <----------/
>>>    ...
>>>
>>>    perf record -C 0 stop
>>>
>>> Now perf samples the PC of taskA. However, perf does not record the
>>> PERF_RECORD_COMM and PERF_RECORD_MMAP events of taskA.
>>> Therefore, the comm and symbols of taskA cannot be parsed.
>>>
>>> The solution is to record sideband events for all CPUs when tracing
>>> selected CPUs. Because this modifies the default behavior, add related
>>> comments to the perf record man page.
>>>
>>> The sys_perf_event_open invoked is as follows:
>>>
>>>    # perf --debug verbose=3 record -e cpu-clock -C 1 true
>>>    <SNIP>
>>>    Opening: cpu-clock
>>>    ------------------------------------------------------------
>>>    perf_event_attr:
>>>      type                             1 (PERF_TYPE_SOFTWARE)
>>>      size                             136
>>>      config                           0 (PERF_COUNT_SW_CPU_CLOCK)
>>>      { sample_period, sample_freq }   4000
>>>      sample_type                      IP|TID|TIME|CPU|PERIOD|IDENTIFIER
>>>      read_format                      ID|LOST
>>>      disabled                         1
>>>      inherit                          1
>>>      freq                             1
>>>      sample_id_all                    1
>>>      exclude_guest                    1
>>>    ------------------------------------------------------------
>>>    sys_perf_event_open: pid -1  cpu 1  group_fd -1  flags 0x8 = 5
>>>    Opening: dummy:u
>>>    ------------------------------------------------------------
>>>    perf_event_attr:
>>>      type                             1 (PERF_TYPE_SOFTWARE)
>>>      size                             136
>>>      config                           0x9 (PERF_COUNT_SW_DUMMY)
>>>      { sample_period, sample_freq }   1
>>>      sample_type                      IP|TID|TIME|CPU|IDENTIFIER
>>>      read_format                      ID|LOST
>>>      inherit                          1
>>>      exclude_kernel                   1
>>>      exclude_hv                       1
>>>      mmap                             1
>>>      comm                             1
>>>      task                             1
>>>      sample_id_all                    1
>>>      exclude_guest                    1
>>>      mmap2                            1
>>>      comm_exec                        1
>>>      ksymbol                          1
>>>      bpf_event                        1
>>>    ------------------------------------------------------------
>>>    sys_perf_event_open: pid -1  cpu 0  group_fd -1  flags 0x8 = 6
>>>    sys_perf_event_open: pid -1  cpu 1  group_fd -1  flags 0x8 = 7
>>>    sys_perf_event_open: pid -1  cpu 2  group_fd -1  flags 0x8 = 9
>>>    sys_perf_event_open: pid -1  cpu 3  group_fd -1  flags 0x8 = 10
>>>    sys_perf_event_open: pid -1  cpu 4  group_fd -1  flags 0x8 = 11
>>>    sys_perf_event_open: pid -1  cpu 5  group_fd -1  flags 0x8 = 12
>>>    sys_perf_event_open: pid -1  cpu 6  group_fd -1  flags 0x8 = 13
>>>    sys_perf_event_open: pid -1  cpu 7  group_fd -1  flags 0x8 = 14
>>>    <SNIP>
>>>
>>> Signed-off-by: Yang Jihong <[email protected]>
>>> ---
>>>   tools/perf/Documentation/perf-record.txt |  3 +++
>>>   tools/perf/builtin-record.c              | 14 +++++++++++++-
>>>   2 files changed, 16 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
>>> index 680396c56bd1..dac53ece51ab 100644
>>> --- a/tools/perf/Documentation/perf-record.txt
>>> +++ b/tools/perf/Documentation/perf-record.txt
>>> @@ -388,6 +388,9 @@ comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-
>>>   In per-thread mode with inheritance mode on (default), samples are captured only when
>>>   the thread executes on the designated CPUs. Default is to monitor all CPUs.
>>>   +User space tasks can migrate between CPUs, so when tracing selected CPUs,
>>> +a dummy event is created to track sideband for all CPUs.
>>> +
>>>   -B::
>>>   --no-buildid::
>>>   Do not save the build ids of binaries in the perf.data files. This skips
>>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>>> index 3ff9d972225e..4e8e97928f05 100644
>>> --- a/tools/perf/builtin-record.c
>>> +++ b/tools/perf/builtin-record.c
>>> @@ -912,6 +912,7 @@ static int record__config_tracking_events(struct record *rec)
>>>   {
>>>       struct record_opts *opts = &rec->opts;
>>>       struct evlist *evlist = rec->evlist;
>>> +    bool system_wide = false;
>>>       struct evsel *evsel;
>>>         /*
>>> @@ -921,7 +922,18 @@ static int record__config_tracking_events(struct record *rec)
>>>        */
>>>       if (opts->target.initial_delay || target__has_cpu(&opts->target) ||
>>>           perf_pmus__num_core_pmus() > 1) {
>>> -        evsel = evlist__findnew_tracking_event(evlist, false);
>>> +
>>> +        /*
>>> +         * User space tasks can migrate between CPUs, so when tracing
>>> +         * selected CPUs, sideband for all CPUs is still needed.
>>> +         *
>>> +         * If all (non-dummy) evsel have exclude_user,
>>> +         * system_wide is not needed.
>>> +         */
>>> +        if (!!opts->target.cpu_list && !opts->all_kernel)
>>
>> Not everyone uses all-kernel.  Can we check the evsels are either dummy
>> or exclude_user?
> For perf_record, exclude_user of all evsels is set in evsel__config(), and record__config_tracking_events() is before evsel__config().
>
> Uh..., it seems that only opts->all_kernel can be used to check exclude_user of evsels.
>
> void evsel__config()
> {
>   ...
>   if (opts->all_kernel) {
>     attr->exclude_kernel = 0;
>     attr->exclude_user   = 1;
>   }
>   ...
> }

The parser updates attr in accordance with ":k" etc. I guess
opts->all_kernel or opts->all_user override that as well.




2023-07-31 15:19:23

by Yang Jihong

[permalink] [raw]
Subject: Re: [PATCH v3 4/7] perf record: Track sideband events for all CPUs when tracing selected CPUs

Hello,

On 2023/7/31 21:01, Adrian Hunter wrote:
> On 31/07/23 15:38, Yang Jihong wrote:
>> Hello,
>>
>> On 2023/7/31 19:08, Adrian Hunter wrote:
>>> On 22/07/23 12:32, Yang Jihong wrote:
>>>> User space tasks can migrate between CPUs, we need to track side-band
>>>> events for all CPUs.
>>>>
>>>> The specific scenarios are as follows:
>>>>
>>>>           CPU0                                 CPU1
>>>>    perf record -C 0 start
>>>>                                taskA starts to be created and executed
>>>>                                  -> PERF_RECORD_COMM and PERF_RECORD_MMAP
>>>>                                     events only deliver to CPU1
>>>>                                ......
>>>>                                  |
>>>>                            migrate to CPU0
>>>>                                  |
>>>>    Running on CPU0    <----------/
>>>>    ...
>>>>
>>>>    perf record -C 0 stop
>>>>
>>>> Now perf samples the PC of taskA. However, perf does not record the
>>>> PERF_RECORD_COMM and PERF_RECORD_MMAP events of taskA.
>>>> Therefore, the comm and symbols of taskA cannot be parsed.
>>>>
>>>> The solution is to record sideband events for all CPUs when tracing
>>>> selected CPUs. Because this modifies the default behavior, add related
>>>> comments to the perf record man page.
>>>>
>>>> The sys_perf_event_open invoked is as follows:
>>>>
>>>>    # perf --debug verbose=3 record -e cpu-clock -C 1 true
>>>>    <SNIP>
>>>>    Opening: cpu-clock
>>>>    ------------------------------------------------------------
>>>>    perf_event_attr:
>>>>      type                             1 (PERF_TYPE_SOFTWARE)
>>>>      size                             136
>>>>      config                           0 (PERF_COUNT_SW_CPU_CLOCK)
>>>>      { sample_period, sample_freq }   4000
>>>>      sample_type                      IP|TID|TIME|CPU|PERIOD|IDENTIFIER
>>>>      read_format                      ID|LOST
>>>>      disabled                         1
>>>>      inherit                          1
>>>>      freq                             1
>>>>      sample_id_all                    1
>>>>      exclude_guest                    1
>>>>    ------------------------------------------------------------
>>>>    sys_perf_event_open: pid -1  cpu 1  group_fd -1  flags 0x8 = 5
>>>>    Opening: dummy:u
>>>>    ------------------------------------------------------------
>>>>    perf_event_attr:
>>>>      type                             1 (PERF_TYPE_SOFTWARE)
>>>>      size                             136
>>>>      config                           0x9 (PERF_COUNT_SW_DUMMY)
>>>>      { sample_period, sample_freq }   1
>>>>      sample_type                      IP|TID|TIME|CPU|IDENTIFIER
>>>>      read_format                      ID|LOST
>>>>      inherit                          1
>>>>      exclude_kernel                   1
>>>>      exclude_hv                       1
>>>>      mmap                             1
>>>>      comm                             1
>>>>      task                             1
>>>>      sample_id_all                    1
>>>>      exclude_guest                    1
>>>>      mmap2                            1
>>>>      comm_exec                        1
>>>>      ksymbol                          1
>>>>      bpf_event                        1
>>>>    ------------------------------------------------------------
>>>>    sys_perf_event_open: pid -1  cpu 0  group_fd -1  flags 0x8 = 6
>>>>    sys_perf_event_open: pid -1  cpu 1  group_fd -1  flags 0x8 = 7
>>>>    sys_perf_event_open: pid -1  cpu 2  group_fd -1  flags 0x8 = 9
>>>>    sys_perf_event_open: pid -1  cpu 3  group_fd -1  flags 0x8 = 10
>>>>    sys_perf_event_open: pid -1  cpu 4  group_fd -1  flags 0x8 = 11
>>>>    sys_perf_event_open: pid -1  cpu 5  group_fd -1  flags 0x8 = 12
>>>>    sys_perf_event_open: pid -1  cpu 6  group_fd -1  flags 0x8 = 13
>>>>    sys_perf_event_open: pid -1  cpu 7  group_fd -1  flags 0x8 = 14
>>>>    <SNIP>
>>>>
>>>> Signed-off-by: Yang Jihong <[email protected]>
>>>> ---
>>>>   tools/perf/Documentation/perf-record.txt |  3 +++
>>>>   tools/perf/builtin-record.c              | 14 +++++++++++++-
>>>>   2 files changed, 16 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
>>>> index 680396c56bd1..dac53ece51ab 100644
>>>> --- a/tools/perf/Documentation/perf-record.txt
>>>> +++ b/tools/perf/Documentation/perf-record.txt
>>>> @@ -388,6 +388,9 @@ comma-separated list with no space: 0,1. Ranges of CPUs are specified with -: 0-
>>>>   In per-thread mode with inheritance mode on (default), samples are captured only when
>>>>   the thread executes on the designated CPUs. Default is to monitor all CPUs.
>>>>   +User space tasks can migrate between CPUs, so when tracing selected CPUs,
>>>> +a dummy event is created to track sideband for all CPUs.
>>>> +
>>>>   -B::
>>>>   --no-buildid::
>>>>   Do not save the build ids of binaries in the perf.data files. This skips
>>>> diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
>>>> index 3ff9d972225e..4e8e97928f05 100644
>>>> --- a/tools/perf/builtin-record.c
>>>> +++ b/tools/perf/builtin-record.c
>>>> @@ -912,6 +912,7 @@ static int record__config_tracking_events(struct record *rec)
>>>>   {
>>>>       struct record_opts *opts = &rec->opts;
>>>>       struct evlist *evlist = rec->evlist;
>>>> +    bool system_wide = false;
>>>>       struct evsel *evsel;
>>>>         /*
>>>> @@ -921,7 +922,18 @@ static int record__config_tracking_events(struct record *rec)
>>>>        */
>>>>       if (opts->target.initial_delay || target__has_cpu(&opts->target) ||
>>>>           perf_pmus__num_core_pmus() > 1) {
>>>> -        evsel = evlist__findnew_tracking_event(evlist, false);
>>>> +
>>>> +        /*
>>>> +         * User space tasks can migrate between CPUs, so when tracing
>>>> +         * selected CPUs, sideband for all CPUs is still needed.
>>>> +         *
>>>> +         * If all (non-dummy) evsel have exclude_user,
>>>> +         * system_wide is not needed.
>>>> +         */
>>>> +        if (!!opts->target.cpu_list && !opts->all_kernel)
>>>
>>> Not everyone uses all-kernel.  Can we check the evsels are either dummy
>>> or exclude_user?
>> For perf_record, exclude_user of all evsels is set in evsel__config(), and record__config_tracking_events() is before evsel__config().
>>
>> Uh..., it seems that only opts->all_kernel can be used to check exclude_user of evsels.
>>
>> void evsel__config()
>> {
>>   ...
>>   if (opts->all_kernel) {
>>     attr->exclude_kernel = 0;
>>     attr->exclude_user   = 1;
>>   }
>>   ...
>> }
>
> The parser updates attr in accordance with ":k" etc. I guess
Yes, the ":k" situation also needs to be considered.

> opts->all_kernel or opts->all_user override that as well.
Yes, opts->all_kernel and opts->all_user will overwrite the original
attr, see [1].

may need to check all_user, all_kernel and non-dummy exclude_user at the
same time:

if ((all_user && one_non_dummy_exist) ||
(!all_user && !all_kernel && one_non_dummy_without_exclude_user)) {
system_wide = true;
}

[1]
# perf --debug verbose=2 record -e cpu-clock:u --all-kernel true
<SNIP>
------------------------------------------------------------
perf_event_attr:
type 1 (PERF_TYPE_SOFTWARE)
size 136
config 0 (PERF_COUNT_SW_CPU_CLOCK)
{ sample_period, sample_freq } 4000
sample_type IP|TID|TIME|PERIOD
read_format ID|LOST
disabled 1
inherit 1
exclude_user 1
exclude_hv 1
mmap 1
comm 1
freq 1
enable_on_exec 1
task 1
sample_id_all 1
exclude_guest 1
mmap2 1
comm_exec 1
ksymbol 1
bpf_event 1
------------------------------------------------------------
<SNIP>

Thanks,
Yang