perf stat: don't report a null stalled cycles per insn metric
PATCH v2 1/3 perf stat: don't report a null stalled cycles per insn metric
For data collected on machines with front end stalled cycles supported, such as found on modern AMD CPU families, commit 146540fb545b ("perf
stat: Always separate stalled cycles per insn") introduces a new line in CSV output with a leading comma that upsets some automated scripts. Scripts have to use "-e ex_ret_instr" to work around this issue, after upgrading to a version of perf with that commit.
We could add "if (have_frontend_stalled && !config->csv_sep)" to the not (total && avg) else clause, to emphasize that CSV users are usually scripts, and are written to do only what is needed, i.e., they wouldn't typically invoke "perf stat" without specifying an explicit event list.
But - let alone CSV output - why should users now tolerate a constant 0-reporting extra line in regular terminal output?:
BEFORE:
$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1
Performance counter stats for 'system wide':
181,110,981 instructions # 0.58 insn per cycle
# 0.00 stalled cycles per insn
309,876,469 cycles
1.002202582 seconds time elapsedThe user would not like to see the now permanent "0.00 stalled cycles per insn" line fixture, as it gives no useful information.
So this patch removes the printing of the zeroed stalled cycles line altogether, almost reverting the very original commit fb4605ba47e7 ("perf stat: Check for frontend stalled for metrics"), which seems like it was written to normalize --metric-only column output of common Intel machines at the time: modern Intel machines have ceased to support the genericised frontend stalled metrics AFAICT.
AFTER:
$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1
Performance counter stats for 'system wide':
244,071,432 instructions # 0.69 insn per cycle
355,353,490 cycles
1.001862516 seconds time elapsedOutput behaviour when stalled cycles is indeed measured is not affected (BEFORE == AFTER):
$ sudo perf stat --all-cpus -einstructions,cycles,stalled-cycles-frontend -- sleep 1
Performance counter stats for 'system wide':
247,227,799 instructions # 0.63 insn per cycle
# 0.26 stalled cycles per insn
394,745,636 cycles
63,194,485 stalled-cycles-frontend # 16.01% frontend cycles idle
1.002079770 seconds time elapsed---
v2: Added Acked-bys from Jiri Olsa:
https://lore.kernel.org/lkml/20200120092928.GD608405@krava/
and Andi Kleen:
https://lore.kernel.org/lkml/[email protected]/
tools/perf/util/stat-shadow.c | 6 ------ 1 file changed, 6 deletions(-)
tools/perf/util/stat-shadow.c +0 -6
@@ -18,7 +18,6 @@* AGGR_NONE: Use matching CPU * AGGR_THREAD: Not supported? */ -static bool have_frontend_stalled;struct runtime_stat rt_stat;struct stats walltime_nsecs_stats;
@@ -144,7 +143,6 @@ void runtime_stat__exit(struct runtime_stat *st)void perf_stat__init_shadow_stats (void ) { - have_frontend_stalled =pmu_have_event ("cpu" ,"stalled-cycles-frontend" );runtime_stat__init (&rt_stat); }
@@ -853,10 +851,6 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,print_metric (config, ctxp,NULL ,"%7.2f " ,"stalled cycles per insn" , ratio); - }else if (have_frontend_stalled) { - out->new_line (config, ctxp); -print_metric (config, ctxp,NULL ,"%7.2f " , -"stalled cycles per insn" ,0 ); } }else if (perf_evsel__match (evsel,HARDWARE ,HW_BRANCH_MISSES )) {if (runtime_stat_n (st,STAT_BRANCHES , ctx, cpu) !=0 )
Signature · 1 line
2.25.0
PATCH v2 2/3 perf symbols: Update the list of kernel idle symbols
"acpi_idle_do_entry", "acpi_processor_ffh_cstate_enter", and "idle_cpu" appear in 'perf top' output, at least on AMD systems.
Add them to perf's idle_symbols list, so they don't dominate 'perf top' output.
---
v2: Added Jiri Olsa's acked-by:
https://lore.kernel.org/lkml/20200120092844.GC608405@krava/
tools/perf/util/symbol.c | 3 +++ 1 file changed, 3 insertions(+)
tools/perf/util/symbol.c +3 -0
@@ -635,9 +635,12 @@ int modules__parse(const char *filename, void *arg,static bool symbol__is_idle (const char *name) {const char *const idle_symbols[] = { +"acpi_idle_do_entry" , +"acpi_processor_ffh_cstate_enter" ,"arch_cpu_idle" ,"cpu_idle" ,"cpu_startup_entry" , +"idle_cpu" ,"intel_idle" ,"default_idle" ,"native_safe_halt" ,
Signature · 1 line
2.25.0
Re: [PATCH 2/3 v2] perf symbols: Update the list of kernel idle symbols
Quoted 20 lines · Kim Phillips
"
acpi_idle_do_entry", "acpi_processor_ffh_cstate_enter", and "idle_cpu" appear in 'perf top' output, at least on AMD systems.Add them to perf's
idle_symbolslist, so they don't dominate 'perf top' output.Cc Peter Zijlstra <[email protected]>Cc Ingo Molnar <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Mark Rutland <[email protected]>Cc Alexander Shishkin <[email protected]>Cc Jiri Olsa <[email protected]>Cc Namhyung Kim <[email protected]>Cc Cong Wang <[email protected]>Cc Andi Kleen <[email protected]>Cc Jin Yao <[email protected]>
Quoted 8 lines
Cc Kan Liang <[email protected]>Cc Kim Phillips <[email protected]>Cc Song Liu <[email protected]>Cc Davidlohr Bueso <[email protected]>Acked-by Jiri Olsa <[email protected]>Signed-off-by Kim Phillips <[email protected]>
[tip: perf/urgent] perf symbols: Update the list of kernel idle symbols
In reply to Kim PhillipsThe following commit has been merged into the perf/urgent branch of tip:
The following commit has been merged into the perf/urgent branch of tip:
Commit-ID: 0e71459afcbbf69e92a65085c45515d3f3f02c31 Gitweb: https://git.kernel.org/tip/0e71459afcbbf69e92a65085c45515d3f3f02c31 Author: Kim Phillips <[email protected]> AuthorDate: Fri, 07 Feb 2020 17:06:12 -06:00 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitterDate: Mon, 10 Feb 2020 16:30:13 -03:00
perf symbols: Update the list of kernel idle symbols
The "acpi_idle_do_entry", "acpi_processor_ffh_cstate_enter", and "idle_cpu" symbols appear in 'perf top' output, at least on AMD systems.
Add them to perf's idle_symbols list, so they don't dominate 'perf top' output.
--- tools/perf/util/symbol.c | 3 +++ 1 file changed, 3 insertions(+)
tools/perf/util/symbol.c +3 -0
@@ -635,9 +635,12 @@ out:static bool symbol__is_idle (const char *name) {const char *const idle_symbols[] = { +"acpi_idle_do_entry" , +"acpi_processor_ffh_cstate_enter" ,"arch_cpu_idle" ,"cpu_idle" ,"cpu_startup_entry" , +"idle_cpu" ,"intel_idle" ,"default_idle" ,"native_safe_halt" ,
PATCH v2 3/3 perf symbols: convert symbol__is_idle to use strlist
In reply to Kim PhillipsUse the more optimized strlist implementation to do the idle function lookup.
---
v2: new this series, based on Jiri's comment:
https://lore.kernel.org/lkml/20200120092844.GC608405@krava/
...and this time with the Cc list intact.
tools/perf/util/symbol.c | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-)
tools/perf/util/symbol.c +9 -5
@@ -654,13 +654,17 @@ static bool symbol__is_idle(const char *name)NULL };int i; +static struct strlist *idle_symbols_list; -for (i =0 ; idle_symbols[i]; i++) { -if (!strcmp (idle_symbols[i], name)) -return true; - } +if (idle_symbols_list) +return strlist__has_entry (idle_symbols_list, name); -return false; + idle_symbols_list =strlist__new (NULL ,NULL ); + +for (i =0 ; idle_symbols[i]; i++) +strlist__add (idle_symbols_list, idle_symbols[i]); + +return strlist__has_entry (idle_symbols_list, name); } static int map__process_kallsym_symbol(void *arg, const char *name,
Signature · 1 line
2.25.0
Re: [PATCH 3/3 v2] perf symbols: convert symbol__is_idle to use strlist
Quoted 40 lines · Kim Phillips
Use the more optimized strlist implementation to do the idle function lookup.
Cc Peter Zijlstra <[email protected]>Cc Ingo Molnar <[email protected]>Cc Andi Kleen <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Mark Rutland <[email protected]>Cc Alexander Shishkin <[email protected]>Cc Jiri Olsa <[email protected]>Cc Jiri Olsa <[email protected]>Cc Namhyung Kim <[email protected]>Cc Cong Wang <[email protected]>Cc Jin Yao <[email protected]>Cc Kan Liang <[email protected]>Cc Kim Phillips <[email protected]>Cc Song Liu <[email protected]>Cc Davidlohr Bueso <[email protected]>Signed-off-by Kim Phillips <[email protected]>---
v2: new this series, based on Jiri's comment:https://lore.kernel.org/lkml/20200120092844.GC608405@krava/
...and this time with the Cc list intact.
tools/perf/util/symbol.c| 14 +++++++++-----1 file changed, 9 insertions(+), 5 deletions(-)tools/perf/util/symbol.c +0
-0
tools/perf/util/symbol.c static bool symbol__is_idle(const char *name) · line 657
Quoted 4 lines
tools/perf/util/symbol.c +1
-0@@ -654,13 +654,17 @@ static bool symbol__is_idle(const char *name) NULL };int i; +static struct strlist *idle_symbols_list;
nit, probably just personal preference:
Maybe move idle_symbols_list out of the function and add the logic to symbol__init()?
Other than this:
Quoted 21 lines
tools/perf/util/symbol.c +8
-5-for (i =0 ; idle_symbols[i]; i++) { -if (!strcmp (idle_symbols[i], name)) -return true; - } +if (idle_symbols_list) +return strlist__has_entry (idle_symbols_list, name); -return false; + idle_symbols_list =strlist__new (NULL ,NULL ); + +for (i =0 ; idle_symbols[i]; i++) +strlist__add (idle_symbols_list, idle_symbols[i]); + +return strlist__has_entry (idle_symbols_list, name); } static int map__process_kallsym_symbol(void *arg, const char *name,Signature · 1 line
2.25.0
Re: [PATCH 3/3 v2] perf symbols: convert symbol__is_idle to use strlist
Em Mon, Feb 10, 2020 at 06:51:53PM +0000, Song Liu escreveu:
Quoted 47 lines
Quoted 40 lines · Kim Phillips
Use the more optimized strlist implementation to do the idle function lookup.
Cc Peter Zijlstra <[email protected]>Cc Ingo Molnar <[email protected]>Cc Andi Kleen <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Mark Rutland <[email protected]>Cc Alexander Shishkin <[email protected]>Cc Jiri Olsa <[email protected]>Cc Jiri Olsa <[email protected]>Cc Namhyung Kim <[email protected]>Cc Cong Wang <[email protected]>Cc Jin Yao <[email protected]>Cc Kan Liang <[email protected]>Cc Kim Phillips <[email protected]>Cc Song Liu <[email protected]>Cc Davidlohr Bueso <[email protected]>Signed-off-by Kim Phillips <[email protected]>---
v2: new this series, based on Jiri's comment:https://lore.kernel.org/lkml/20200120092844.GC608405@krava/
...and this time with the Cc list intact.
tools/perf/util/symbol.c| 14 +++++++++-----1 file changed, 9 insertions(+), 5 deletions(-)tools/perf/util/symbol.c +0
-0tools/perf/util/symbol.c static bool symbol__is_idle(const char *name) · line 657
Quoted 4 lines
tools/perf/util/symbol.c +1
-0@@ -654,13 +654,17 @@ static bool symbol__is_idle(const char *name) NULL };int i; +static struct strlist *idle_symbols_list;nit, probably just personal preference:
Quoted 6 lines
Maybe move
idle_symbols_listout of the function and add the logic tosymbol__init()?Other than this:
Acked-by Song Liu <[email protected]>
I applied it as is, improvements can be made on top of it.
- Arnaldo
Quoted 22 lines
Quoted 21 lines
tools/perf/util/symbol.c +8
-5-for (i =0 ; idle_symbols[i]; i++) { -if (!strcmp (idle_symbols[i], name)) -return true; - } +if (idle_symbols_list) +return strlist__has_entry (idle_symbols_list, name); -return false; + idle_symbols_list =strlist__new (NULL ,NULL ); + +for (i =0 ; idle_symbols[i]; i++) +strlist__add (idle_symbols_list, idle_symbols[i]); + +return strlist__has_entry (idle_symbols_list, name); } static int map__process_kallsym_symbol(void *arg, const char *name,Signature · 1 line
2.25.0
Signature · 1 line
- Arnaldo
[tip: perf/urgent] perf symbols: Convert symbol__is_idle() to use strlist
In reply to Kim PhillipsThe following commit has been merged into the perf/urgent branch of tip:
The following commit has been merged into the perf/urgent branch of tip:
Commit-ID: bc5f15be2c814ca1ff6bb4e62d5b275a8c88cbb1 Gitweb: https://git.kernel.org/tip/bc5f15be2c814ca1ff6bb4e62d5b275a8c88cbb1 Author: Kim Phillips <[email protected]> AuthorDate: Mon, 10 Feb 2020 10:31:47 -06:00 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitterDate: Mon, 10 Feb 2020 16:30:51 -03:00
perf symbols: Convert symbol__is_idle() to use strlist
Use the more optimized strlist implementation to do the idle function lookup.
--- tools/perf/util/symbol.c | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-)
tools/perf/util/symbol.c +9 -5
@@ -654,13 +654,17 @@ static bool symbol__is_idle(const char *name)NULL };int i; +static struct strlist *idle_symbols_list; -for (i =0 ; idle_symbols[i]; i++) { -if (!strcmp (idle_symbols[i], name)) -return true; - } +if (idle_symbols_list) +return strlist__has_entry (idle_symbols_list, name); -return false; + idle_symbols_list =strlist__new (NULL ,NULL ); + +for (i =0 ; idle_symbols[i]; i++) +strlist__add (idle_symbols_list, idle_symbols[i]); + +return strlist__has_entry (idle_symbols_list, name); } static int map__process_kallsym_symbol(void *arg, const char *name,
Quoted 79 lines · Kim Phillips
For data collected on machines with front end stalled cycles supported, such as found on modern AMD CPU families, commit
146540fb545b("perf
stat: Always separate stalled cycles per insn") introduces a new line in CSV output with a leading comma that upsets some automated scripts. Scripts have to use "-eex_ret_instr" to work around this issue, after upgrading to a version of perf with that commit.We could add "if (
have_frontend_stalled&& !config->csv_sep)" to the not (total && avg) else clause, to emphasize that CSV users are usually scripts, and are written to do only what is needed, i.e., they wouldn't typically invoke "perf stat" without specifying an explicit event list.But - let alone CSV output - why should users now tolerate a constant 0-reporting extra line in regular terminal output?:
BEFORE:
$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1Performance counter stats for 'system wide':
181,110,981 instructions # 0.58 insn per cycle # 0.00 stalled cycles per insn 309,876,469 cycles 1.002202582 seconds time elapsedThe user would not like to see the now permanent "0.00 stalled cycles per insn" line fixture, as it gives no useful information.
So this patch removes the printing of the zeroed stalled cycles line altogether, almost reverting the very original commit
fb4605ba47e7("perf stat: Check for frontend stalled for metrics"), which seems like it was written to normalize --metric-only column output of common Intel machines at the time: modern Intel machines have ceased to support the genericised frontend stalled metrics AFAICT.AFTER:
$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1Performance counter stats for 'system wide':
244,071,432 instructions # 0.69 insn per cycle 355,353,490 cycles 1.001862516 seconds time elapsedOutput behaviour when stalled cycles is indeed measured is not affected (BEFORE == AFTER):
$ sudo perf stat --all-cpus -einstructions,cycles,stalled-cycles-frontend -- sleep 1Performance counter stats for 'system wide':
247,227,799 instructions # 0.63 insn per cycle # 0.26 stalled cycles per insn 394,745,636 cycles 63,194,485 stalled-cycles-frontend # 16.01% frontend cycles idle 1.002079770 seconds time elapsedCc Peter Zijlstra <[email protected]>Cc Ingo Molnar <[email protected]>Cc Arnaldo Carvalho de Melo <[email protected]>Cc Mark Rutland <[email protected]>Cc Alexander Shishkin <[email protected]>Cc Jiri Olsa <[email protected]>Cc Namhyung Kim <[email protected]>Cc Cong Wang <[email protected]>Cc Jin Yao <[email protected]>Cc Kan Liang <[email protected]>Cc Kim Phillips <[email protected]>
Quoted 8 lines
Cc Song Liu <[email protected]>Cc Davidlohr Bueso <[email protected]>Acked-by Jiri Olsa <[email protected]>Acked-by Andi Kleen <[email protected]>Fixes 146540fb545b ("perf stat: Always separate stalled cycles per insn")Signed-off-by Kim Phillips <[email protected]>
Thanks for improving perf on AMD systems!
[tip: perf/urgent] perf stat: Don't report a null stalled cycles per insn metric
In reply to Kim PhillipsThe following commit has been merged into the perf/urgent branch of tip:
The following commit has been merged into the perf/urgent branch of tip:
Commit-ID: 80cc7bb6c104d733bff60ddda09f19139c61507c Gitweb: https://git.kernel.org/tip/80cc7bb6c104d733bff60ddda09f19139c61507c Author: Kim Phillips <[email protected]> AuthorDate: Fri, 07 Feb 2020 17:06:11 -06:00 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitterDate: Mon, 10 Feb 2020 16:30:09 -03:00
perf stat: Don't report a null stalled cycles per insn metric
For data collected on machines with front end stalled cycles supported, such as found on modern AMD CPU families, commit 146540fb545b ("perf
stat: Always separate stalled cycles per insn") introduces a new line in CSV output with a leading comma that upsets some automated scripts. Scripts have to use "-e ex_ret_instr" to work around this issue, after upgrading to a version of perf with that commit.
We could add "if (have_frontend_stalled && !config->csv_sep)" to the not (total && avg) else clause, to emphasize that CSV users are usually scripts, and are written to do only what is needed, i.e., they wouldn't typically invoke "perf stat" without specifying an explicit event list.
But - let alone CSV output - why should users now tolerate a constant 0-reporting extra line in regular terminal output?:
BEFORE:
$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1
Performance counter stats for 'system wide':
181,110,981 instructions # 0.58 insn per cycle
# 0.00 stalled cycles per insn
309,876,469 cycles
1.002202582 seconds time elapsedThe user would not like to see the now permanent:
"0.00 stalled cycles per insn"
line fixture, as it gives no useful information.
So this patch removes the printing of the zeroed stalled cycles line altogether, almost reverting the very original commit fb4605ba47e7 ("perf stat: Check for frontend stalled for metrics"), which seems like it was written to normalize --metric-only column output of common Intel machines at the time: modern Intel machines have ceased to support the genericised frontend stalled metrics AFAICT.
AFTER:
$ sudo perf stat --all-cpus -einstructions,cycles -- sleep 1
Performance counter stats for 'system wide':
244,071,432 instructions # 0.69 insn per cycle
355,353,490 cycles
1.001862516 seconds time elapsedOutput behaviour when stalled cycles is indeed measured is not affected (BEFORE == AFTER):
$ sudo perf stat --all-cpus -einstructions,cycles,stalled-cycles-frontend -- sleep 1
Performance counter stats for 'system wide':
247,227,799 instructions # 0.63 insn per cycle
# 0.26 stalled cycles per insn
394,745,636 cycles
63,194,485 stalled-cycles-frontend # 16.01% frontend cycles idle
1.002079770 seconds time elapsed--- tools/perf/util/stat-shadow.c | 6 ------ 1 file changed, 6 deletions(-)
tools/perf/util/stat-shadow.c +0 -6
@@ -18,7 +18,6 @@* AGGR_NONE: Use matching CPU * AGGR_THREAD: Not supported? */ -static bool have_frontend_stalled;struct runtime_stat rt_stat;struct stats walltime_nsecs_stats;
@@ -144,7 +143,6 @@ void runtime_stat__exit(struct runtime_stat *st)void perf_stat__init_shadow_stats (void ) { - have_frontend_stalled =pmu_have_event ("cpu" ,"stalled-cycles-frontend" );runtime_stat__init (&rt_stat); }
@@ -853,10 +851,6 @@ void perf_stat__print_shadow_stats(struct perf_stat_config *config,print_metric (config, ctxp,NULL ,"%7.2f " ,"stalled cycles per insn" , ratio); - }else if (have_frontend_stalled) { - out->new_line (config, ctxp); -print_metric (config, ctxp,NULL ,"%7.2f " , -"stalled cycles per insn" ,0 ); } }else if (perf_evsel__match (evsel,HARDWARE ,HW_BRANCH_MISSES )) {if (runtime_stat_n (st,STAT_BRANCHES , ctx, cpu) !=0 )