perf: Fix "Track with sched_switch" test by not printing warnings in quiet mode
PATCH 1/1 perf: Fix "Track with sched_switch" test by not printing warnings in quiet mode
Especially when CONFIG_LOCKDEP and other debug configs are enabled, Perf can print the following warning when running the "Track with sched_switch" test:
Warning:
Processed 1378918 events and lost 4 chunks!
Check IO/CPU overload!
Warning:
Processed 4593325 samples and lost 70.00%!
The test already supplies -q to run in quiet mode, so extend quiet mode to perf_stdio__warning() and also ui__warning() for consistency.
This fixes the following failure due to the extra lines counted:
perf test "lock cont" -vvv
82: kernel lock contention analysis test :
--- start ---
test child forked, pid 3125
Testing perf lock record and perf lock contention [Fail] Recorded result count is not 1: 9 test child finished with -1
---- end ----
kernel lock contention analysis test: FAILED!
--- tools/perf/ui/util.c | 5 +++++ 1 file changed, 5 insertions(+)
tools/perf/ui/util.c +5 -0
@@ -15,6 +15,9 @@ static int perf_stdio__error(const char *format, va_list args)static int perf_stdio__warning (const char *format,va_list args) { +if (quiet) +return 0 ; +fprintf (stderr,"Warning:\n" );vfprintf (stderr, format, args);return 0 ;
@@ -45,6 +48,8 @@ int ui__warning(const char *format, ...) {int ret;va_list args; +if (quiet) +return 0 ;va_start (args, format); ret = perf_eops->warning (format, args);
Signature · 1 line
2.28.0
Quoted 3 lines · James Clark
Especially when
CONFIG_LOCKDEPand other debug configs are enabled, Perf can print the following warning when running the "Track withsched_switch" test:
Oops got the wrong test name here and in the title. Should be "kernel lock contention analysis test"
Quoted 55 lines
Warning:
Processed 1378918 events and lost 4 chunks!Check IO/CPU overload!
Warning:
Processed 4593325 samples and lost 70.00%!The test already supplies -q to run in quiet mode, so extend quiet mode to
perf_stdio__warning()and alsoui__warning()for consistency.This fixes the following failure due to the extra lines counted:
perf test "lock cont" -vvv
82: kernel lock contention analysis test :
--- start ---
test child forked, pid 3125
Testing perf lock record and perf lock contention [Fail] Recorded result count is not 1: 9 test child finished with -1
---- end ----
kernel lock contention analysis test: FAILED!Fixes ec685de25b67 ("perf test: Add kernel lock contention test")Cc Namhyung Kim <[email protected]>Signed-off-by James Clark <[email protected]>--- tools/perf/ui/util.c | 5 +++++ 1 file changed, 5 insertions(+)tools/perf/ui/util.c +5
-0@@ -15,6 +15,9 @@ static int perf_stdio__error(const char *format, va_list args)static int perf_stdio__warning (const char *format,va_list args) { +if (quiet) +return 0 ; +fprintf (stderr,"Warning:\n" );vfprintf (stderr, format, args);return 0 ;@@ -45,6 +48,8 @@ int ui__warning(const char *format, ...) {int ret;va_list args; +if (quiet) +return 0 ;va_start (args, format); ret = perf_eops->warning (format, args);
Namhyung Kim
maintainerQuoted 10 lines · James Clark
Quoted 3 lines · James Clark
Especially when
CONFIG_LOCKDEPand other debug configs are enabled, Perf can print the following warning when running the "Track withsched_switch" test:Oops got the wrong test name here and in the title. Should be "kernel lock contention analysis test"
Could you please resend?
Quoted 5 lines
Quoted 5 lines
Warning:
Processed 1378918 events and lost 4 chunks!
Quoted 7 lines
Quoted 7 lines
Check IO/CPU overload!
Warning:
Processed 4593325 samples and lost 70.00%!The test already supplies -q to run in quiet mode, so extend quiet mode to
perf_stdio__warning()and alsoui__warning()for consistency.
I'm not sure if suppressing the warnings with -q is a good thing. Maybe we need to separate warning/debug messages from the output.
Thanks,
Namhyung
Quoted 44 lines
Quoted 44 lines
This fixes the following failure due to the extra lines counted:
perf test "lock cont" -vvv
82: kernel lock contention analysis test :
--- start ---
test child forked, pid 3125
Testing perf lock record and perf lock contention [Fail] Recorded result count is not 1: 9 test child finished with -1
---- end ----
kernel lock contention analysis test: FAILED!Fixes ec685de25b67 ("perf test: Add kernel lock contention test")Cc Namhyung Kim <[email protected]>Signed-off-by James Clark <[email protected]>--- tools/perf/ui/util.c | 5 +++++ 1 file changed, 5 insertions(+)tools/perf/ui/util.c +5
-0@@ -15,6 +15,9 @@ static int perf_stdio__error(const char *format, va_list args)static int perf_stdio__warning (const char *format,va_list args) { +if (quiet) +return 0 ; +fprintf (stderr,"Warning:\n" );vfprintf (stderr, format, args);return 0 ;@@ -45,6 +48,8 @@ int ui__warning(const char *format, ...) {int ret;va_list args; +if (quiet) +return 0 ;va_start (args, format); ret = perf_eops->warning (format, args);
Quoted 21 lines · Namhyung Kim
Quoted 10 lines · James Clark
Quoted 3 lines · James Clark
Especially when
CONFIG_LOCKDEPand other debug configs are enabled, Perf can print the following warning when running the "Track withsched_switch" test:Oops got the wrong test name here and in the title. Should be "kernel lock contention analysis test"
Could you please resend?
Quoted 7 lines
Quoted 7 lines
Warning:
Processed 1378918 events and lost 4 chunks!Check IO/CPU overload!
Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
Warning:
Processed 4593325 samples and lost 70.00%!The test already supplies -q to run in quiet mode, so extend quiet mode to
perf_stdio__warning()and alsoui__warning()for consistency.I'm not sure if suppressing the warnings with -q is a good thing. Maybe we need to separate warning/debug messages from the output.
I don't see the issue with warnings being suppressed in quiet mode as long as errors are still printed. In other cases warnings have already been suppressed by quiet mode and this site is the odd one out.
What use case are you thinking of where someone explicitly adds -q but wants to see non fatal warnings?
Thanks
James
Quoted 49 lines
Thanks,
NamhyungQuoted 44 lines
Quoted 44 lines
This fixes the following failure due to the extra lines counted:
perf test "lock cont" -vvv
82: kernel lock contention analysis test :
--- start ---
test child forked, pid 3125
Testing perf lock record and perf lock contention [Fail] Recorded result count is not 1: 9 test child finished with -1
---- end ----
kernel lock contention analysis test: FAILED!Fixes ec685de25b67 ("perf test: Add kernel lock contention test")Cc Namhyung Kim <[email protected]>Signed-off-by James Clark <[email protected]>--- tools/perf/ui/util.c | 5 +++++ 1 file changed, 5 insertions(+)tools/perf/ui/util.c +5
-0@@ -15,6 +15,9 @@ static int perf_stdio__error(const char *format, va_list args)static int perf_stdio__warning (const char *format,va_list args) { +if (quiet) +return 0 ; +fprintf (stderr,"Warning:\n" );vfprintf (stderr, format, args);return 0 ;@@ -45,6 +48,8 @@ int ui__warning(const char *format, ...) {int ret;va_list args; +if (quiet) +return 0 ;va_start (args, format); ret = perf_eops->warning (format, args);
Namhyung Kim
maintainerQuoted 11 lines · James Clark
Quoted 6 lines · Namhyung Kim
Quoted 2 lines · James Clark
Quoted 2 lines
The test already supplies -q to run in quiet mode, so extend quiet mode to
perf_stdio__warning()and alsoui__warning()for consistency.I'm not sure if suppressing the warnings with -q is a good thing. Maybe we need to separate warning/debug messages from the output.
Quoted 6 lines
I don't see the issue with warnings being suppressed in quiet mode as long as errors are still printed. In other cases warnings have already been suppressed by quiet mode and this site is the odd one out.
What use case are you thinking of where someone explicitly adds -q but wants to see non fatal warnings?
I don't have any specific use case. If it's already suppressed in other cases, I'm fine with it.
Thanks,
Namhyung
Quoted 16 lines · Namhyung Kim
Quoted 15 lines · James Clark
Quoted 6 lines · Namhyung Kim
Quoted 2 lines · James Clark
Quoted 2 lines
The test already supplies -q to run in quiet mode, so extend quiet mode to
perf_stdio__warning()and alsoui__warning()for consistency.I'm not sure if suppressing the warnings with -q is a good thing. Maybe we need to separate warning/debug messages from the output.
I don't see the issue with warnings being suppressed in quiet mode as long as errors are still printed. In other cases warnings have already been suppressed by quiet mode and this site is the odd one out.
Quoted 6 lines
Quoted 2 lines
What use case are you thinking of where someone explicitly adds -q but wants to see non fatal warnings?
I don't have any specific use case. If it's already suppressed in other cases, I'm fine with it.
Actually I may have been mistaken. Seems like quiet is only used for "extra info" type messages rather than warnings. Although the commit message does say:
The -q/--quiet option is to suppress any message. Sometimes users just want to see the numbers and it can be used for that case.
With 'any' that I would take to include warnings as well. I could move warnings to stderr, but this has a much greater chance of breaking anyone's workflows that might be looking for warnings on stdout than removing warnings when -q is provided.
Also if warnings are moved to stderr and quiet isn't used, there would be no way to suppress warnings in the TUI which might actually be a useful feature.
So I'm still leaning towards the original change, if you are ok with that even though it's not done elsewhere?
Quoted 2 lines
Thanks,
Namhyung
Arnaldo Carvalho de Melo
maintainerEm Fri, Oct 14, 2022 at 10:47:34AM +0100, James Clark escreveu:
Quoted 33 lines · Namhyung Kim
Quoted 13 lines · James Clark
Quoted 12 lines · Namhyung Kim
Quoted 6 lines · James Clark
Quoted 2 lines
Quoted 2 lines
The test already supplies -q to run in quiet mode, so extend quiet mode to
perf_stdio__warning()and alsoui__warning()for consistency.I'm not sure if suppressing the warnings with -q is a good thing. Maybe we need to separate warning/debug messages from the output.
I don't see the issue with warnings being suppressed in quiet mode as long as errors are still printed. In other cases warnings have already been suppressed by quiet mode and this site is the odd one out.
Quoted 6 lines
Quoted 2 lines
What use case are you thinking of where someone explicitly adds -q but wants to see non fatal warnings?
I don't have any specific use case. If it's already suppressed in other cases, I'm fine with it.
Actually I may have been mistaken. Seems like quiet is only used for "extra info" type messages rather than warnings. Although the commit message does say:
The -q/--quiet option is to suppress any message. Sometimes users just want to see the numbers and it can be used for that case.
With 'any' that I would take to include warnings as well. I could move warnings to stderr, but this has a much greater chance of breaking anyone's workflows that might be looking for warnings on stdout than removing warnings when -q is provided.
Quoted 6 lines
Also if warnings are moved to stderr and quiet isn't used, there would be no way to suppress warnings in the TUI which might actually be a useful feature.
So I'm still leaning towards the original change, if you are ok with that even though it's not done elsewhere?
Namhyung? I tend to agree with James.
- Arnaldo
Namhyung Kim
maintainerQuoted 35 lines · Arnaldo Carvalho de Melo
Em Fri, Oct 14, 2022 at 10:47:34AM +0100, James Clark escreveu:
Quoted 33 lines · Namhyung Kim
Quoted 13 lines · James Clark
Quoted 12 lines · Namhyung Kim
Quoted 6 lines · James Clark
Quoted 2 lines
Quoted 2 lines
The test already supplies -q to run in quiet mode, so extend quiet mode to
perf_stdio__warning()and alsoui__warning()for consistency.I'm not sure if suppressing the warnings with -q is a good thing. Maybe we need to separate warning/debug messages from the output.
I don't see the issue with warnings being suppressed in quiet mode as long as errors are still printed. In other cases warnings have already been suppressed by quiet mode and this site is the odd one out.
Quoted 6 lines
Quoted 2 lines
What use case are you thinking of where someone explicitly adds -q but wants to see non fatal warnings?
I don't have any specific use case. If it's already suppressed in other cases, I'm fine with it.
Actually I may have been mistaken. Seems like quiet is only used for "extra info" type messages rather than warnings. Although the commit message does say:
The -q/--quiet option is to suppress any message. Sometimes users just want to see the numbers and it can be used for that case.
With 'any' that I would take to include warnings as well. I could move warnings to stderr, but this has a much greater chance of breaking anyone's workflows that might be looking for warnings on stdout than removing warnings when -q is provided.
Quoted 8 lines
Quoted 6 lines
Also if warnings are moved to stderr and quiet isn't used, there would be no way to suppress warnings in the TUI which might actually be a useful feature.
So I'm still leaning towards the original change, if you are ok with that even though it's not done elsewhere?
Namhyung? I tend to agree with James.
I think it's a matter of choice. With this change, users won't see warnings with -q anymore. This MIGHT affect some users as they can see low quality data silently due to missing data or something.
If you're ok with the behavior, I'm fine. But then we need to update the document to clarify the behavior.
Thanks,
Namhyung