Re: [PATCH v7 4/6] perf record: Track sideband events for all CPUs when tracing selected CPUs
From: Yang Jihong
Date: Mon Aug 28 2023 - 21:43:01 EST
Hello,
On 2023/8/28 19:25, Adrian Hunter wrote:
On 28/08/23 06:03, Yang Jihong wrote:
Hello,
On 2023/8/26 22:59, Namhyung Kim wrote:
Hello,
On Fri, Aug 25, 2023 at 8:29 PM Yang Jihong <yangjihong1@xxxxxxxxxx> 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 <yangjihong1@xxxxxxxxxx>
Acked-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
---
tools/perf/Documentation/perf-record.txt | 3 ++
tools/perf/builtin-record.c | 44 +++++++++++++++++++++++-
2 files changed, 46 insertions(+), 1 deletion(-)
diff --git a/tools/perf/Documentation/perf-record.txt b/tools/perf/Documentation/perf-record.txt
index d5217be012d7..1889f66addf2 100644
--- a/tools/perf/Documentation/perf-record.txt
+++ b/tools/perf/Documentation/perf-record.txt
@@ -374,6 +374,9 @@ 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 83bd1f117191..21c571018148 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -906,10 +906,44 @@ static int record__config_off_cpu(struct record *rec)
return off_cpu_prepare(rec->evlist, &rec->opts.target, &rec->opts);
}
+static bool record__tracking_system_wide(struct record *rec)
+{
+ struct record_opts *opts = &rec->opts;
+ struct evlist *evlist = rec->evlist;
+ struct evsel *evsel;
+
+ /*
+ * If all (non-dummy) evsel have exclude_user,
+ * system_wide is not needed.
Maybe I missed some earlier discussion but why is it not
needed when exclude_user is set? I think it still needs
FORK or COMM at least..
This is Adrian's suggestion earlier, I think it's probably because if exclude_user is set, MMAP information is not needed, that's my guess.
However, as you said, even if exclude_user is set, at least FORK and COMM are required.
Therefore, the conditions here need to be changed to:
"system_wide is need as long as there is a non-dummy event."
@Adrian, is this change okay?
If you wish. I think we use FORK to get mappings, so not sure it
would be needed. There is PID, TID so COMM is not essential.
There are FORK, COMM etc from the CPUs being traced of course.
I think COMM is also necessary. If there is no COMM, the perf report
cannot display the comm of the process migrated from other core.
The test result is as follows:
# perf report --stdio
# To display the perf.data header info, please use
--header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 33 of event 'cpu-clock:khppp'
# Event count (approx.): 8250000
#
# Overhead Command Shared Object Symbol
# ........ ............... .................
.......................................
#
15.15% test3 [kernel.kallsyms] [k] finish_task_switch
9.09% kworker/1:0-eve [kernel.kallsyms] [k] _raw_spin_unlock_irq
3.03% :934 [kernel.kallsyms] [k] blk_done_softirq
3.03% :934 [kernel.kallsyms] [k] finish_task_switch
<SNIP>
Thanks,
Yang