perf record: Fix "read LOST count failed" msg with sample read
PATCH 1/1 perf record: Fix "read LOST count failed" msg with sample read
From: Kan Liang <[email protected]>
Hundreds of "read LOST count failed" error messages may be displayed, when the below command is launched.
perf record -e '{cpu/mem-loads-aux/,cpu/event=0xcd,umask=0x1/}:S' -a
According to the commit 89e3106fa25f ("libperf: Handle read format in perf_evsel__read()"), the PERF_FORMAT_GROUP is only available for the leader. However, the record__read_lost_samples() goes through every entry of an evlist, which includes both leader and member. The member event errors out and triggers the error message. Since there may be hundreds of CPUs on a server, the message will be printed hundreds of times, which is very annoying.
The message itself is correct, but the pr_err is a overkill. Other error messages in the record__read_lost_samples() are all pr_debug. To make the output format consistent, change the pr_err("read LOST count failed\n"); to pr_debug("read LOST count failed\n");.
User can still get the message via -v option.
--- tools/perf/builtin-record.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/builtin-record.c +1 -1
@@ -1866,7 +1866,7 @@ static void __record__read_lost_samples(struct record *rec, struct evsel *evsel,int id_hdr_size;if (perf_evsel__read (&evsel->core, cpu_idx, thread_idx, &count) <0 ) { -pr_err ("read LOST count failed\n" ); +pr_debug ("read LOST count failed\n" );return ; }
Signature · 1 line
2.35.1
Hi Kan,
Quoted 16 lines · On Wed, Mar 1, 2023 at 7:04 AM <[email protected]> wrote:
From: Kan Liang <[email protected]>
Hundreds of "read LOST count failed" error messages may be displayed, when the below command is launched.
perf record -e '{cpu/mem-loads-aux/,cpu/event=0xcd,umask=0x1/}:S' -a
According to the commit
89e3106fa25f("libperf: Handle read format inperf_evsel__read()"), thePERF_FORMAT_GROUPis only available for the leader. However, therecord__read_lost_samples()goes through every entry of an evlist, which includes both leader and member. The member event errors out and triggers the error message. Since there may be hundreds of CPUs on a server, the message will be printed hundreds of times, which is very annoying.
Quoted 8 lines
The message itself is correct, but the
pr_erris a overkill. Other error messages in therecord__read_lost_samples()are allpr_debug. To make the output format consistent, change thepr_err("read LOST count failed\n"); topr_debug("read LOST count failed\n");.
User can still get the message via -v option.Fixes e3a23261ad06 ("perf record: Read and inject LOST_SAMPLES events")Signed-off-by Kan Liang <[email protected]>
Thanks,
Namhyung
Quoted 20 lines
--- tools/perf/builtin-record.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)tools/perf/builtin-record.c +1
-2@@ -1866,7 +1866,7 @@ static void __record__read_lost_samples(struct record *rec, struct evsel *evsel,int id_hdr_size;if (perf_evsel__read (&evsel->core, cpu_idx, thread_idx, &count) <0 ) { -pr_err ("read LOST count failed\n" ); +pr_debug ("read LOST count failed\n" );return ; } --2.35.1