perf tools: Add support for pinned modifier
PATCH v2 1/2 perf tools: Add support for pinned modifier
This commit adds support for a new modifier "D", which requests that the event, or group of events, be pinned to the PMU.
The "p" modifier is already taken for precise, and "P" may be used in future to mean "fully precise".
So we use "D", which stands for pinneD - and looks like a padlock, or if you're using the ":D" syntax perf smiles at you.
This is an oft-requested feature from our HW folks, who want to be able to run a large number of events, but also want 100% accurate results for instructions per cycle.
Comparison of results with and without pinning:
$ perf stat -e '{cycles,instructions}:D' -e cycles,instructions,... 79,590,480,683 cycles # 0.000 GHz
166,123,716,524 instructions # 2.09 insns per cycle
# 0.11 stalled cycles per insn 79,352,134,463 cycles # 0.000 GHz [11.11%]
165,178,301,818 instructions # 2.08 insns per cycle
# 0.11 stalled cycles per insn [11.13%]As you can see although perf does a very good job of scaling the values in the non-pinned case, there is some small discrepancy.
The patch is fairly straight forward, the one detail is that we need to make sure we only request pinning for the group leader when we have a group.
---
v2: Use "D" instead of "P" as discussed on the list.
tools/perf/Documentation/perf-list.txt | 1 + tools/perf/util/parse-events.c | 11 ++++++++++- tools/perf/util/parse-events.l | 3 ++- 3 files changed, 13 insertions(+), 2 deletions(-)
tools/perf/Documentation/perf-list.txt +1 -0
@@ -30,6 +30,7 @@ counted. The following modifiers exist: H - host counting (not in KVM guests) p - precise level S - read sample value (PERF_SAMPLE_READ) + D - pin the event to the PMU The 'p' modifier can be used for specifying how precise the instruction address should be. The 'p' modifier can be specified multiple times:
tools/perf/util/parse-events.c +10 -1
@@ -688,6 +688,7 @@ struct event_modifier {int precise;int exclude_GH;int sample_read; +int pinned; }; static int get_event_modifier(struct event_modifier *mod, char *str,
@@ -700,6 +701,7 @@ static int get_event_modifier(struct event_modifier *mod, char *str,int eG = evsel ? evsel->attr.exclude_guest :0 ;int precise = evsel ? evsel->attr.precise_ip :0 ;int sample_read =0 ; +int pinned = evsel ? evsel->attr.pinned :0 ;int exclude = eu | ek | eh;int exclude_GH = evsel ? evsel->exclude_GH :0 ;
@@ -734,6 +736,8 @@ static int get_event_modifier(struct event_modifier *mod, char *str, eG =1 ; }else if (*str =='S' ) { sample_read =1 ; + }else if (*str =='D' ) { + pinned =1 ; }else break ;
@@ -761,6 +765,8 @@ static int get_event_modifier(struct event_modifier *mod, char *str, mod->precise = precise; mod->exclude_GH = exclude_GH; mod->sample_read = sample_read; + mod->pinned = pinned; +return 0 ; }
@@ -773,7 +779,7 @@ static int check_modifier(char *str)char *p = str;/* The sizeof includes 0 byte as well. */ -if (strlen (str) > (sizeof ("ukhGHpppS" ) -1 )) +if (strlen (str) > (sizeof ("ukhGHpppSD" ) -1 ))return -1 ;while (*p) {
@@ -812,6 +818,9 @@ int parse_events__modifier_event(struct list_head *list, char *str, bool add) evsel->attr.exclude_guest = mod.eG; evsel->exclude_GH = mod.exclude_GH; evsel->sample_read = mod.sample_read; + +if (evsel->leader == evsel) + evsel->attr.pinned = mod.pinned; }return 0 ;
tools/perf/util/parse-events.l +2 -1
@@ -82,7 +82,8 @@ num_hex 0x[a-fA-F0-9]+ num_raw_hex [a-fA-F0-9]+ name [a-zA-Z_*?][a-zA-Z0-9_*?]* name_minus [a-zA-Z_*?][a-zA-Z0-9\-_*?]* -modifier_event [ukhpGHS]+ +/* If you add a modifier you need to update check_modifier() */ +modifier_event [ukhpGHSD]+ modifier_bp [rwx]{1,3} %%
Signature · 1 line
1.8.1.2
PATCH v2 2/2 perf tests: Add tests of new pinned modifier
Add a negative test to test__checkevent_pmu_events() to get lots of coverage of the negative case, ie. when the modifier is not specified.
Add a test of a single event, and of the group case.
--- tools/perf/tests/parse-events.c | 53 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 53 insertions(+)
tools/perf/tests/parse-events.c +53 -0
@@ -452,6 +452,7 @@ static int test__checkevent_pmu_events(struct perf_evlist *evlist) evsel->attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->attr.exclude_hv);TEST_ASSERT_VAL ("wrong precise_ip" , !evsel->attr.precise_ip); +TEST_ASSERT_VAL ("wrong pinned" , !evsel->attr.pinned);return 0 ; }
@@ -1070,6 +1071,50 @@ static int test__leader_sample2(struct perf_evlist *evlist __maybe_unused)return 0 ; } +static int test__checkevent_pinned_modifier (struct perf_evlist *evlist) +{ +struct perf_evsel *evsel =perf_evlist__first (evlist); + +TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->attr.exclude_user); +TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->attr.exclude_kernel); +TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->attr.exclude_hv); +TEST_ASSERT_VAL ("wrong precise_ip" , evsel->attr.precise_ip); +TEST_ASSERT_VAL ("wrong pinned" , evsel->attr.pinned); + +return test__checkevent_symbolic_name (evlist); +} + +static int test__pinned_group (struct perf_evlist *evlist) +{ +struct perf_evsel *evsel, *leader; + +TEST_ASSERT_VAL ("wrong number of entries" ,3 == evlist->nr_entries); + +/* cycles - group leader */ + evsel = leader =perf_evlist__first (evlist); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->attr.type); +TEST_ASSERT_VAL ("wrong config" , +PERF_COUNT_HW_CPU_CYCLES == evsel->attr.config); +TEST_ASSERT_VAL ("wrong group name" , !evsel->group_name); +TEST_ASSERT_VAL ("wrong leader" , evsel->leader == leader); +TEST_ASSERT_VAL ("wrong pinned" , evsel->attr.pinned); + +/* cache-misses - can not be pinned, but will go on with the leader */ + evsel =perf_evsel__next (evsel); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->attr.type); +TEST_ASSERT_VAL ("wrong config" , +PERF_COUNT_HW_CACHE_MISSES == evsel->attr.config); +TEST_ASSERT_VAL ("wrong pinned" , !evsel->attr.pinned); + +/* branch-misses - ditto */ + evsel =perf_evsel__next (evsel); +TEST_ASSERT_VAL ("wrong config" , +PERF_COUNT_HW_BRANCH_MISSES == evsel->attr.config); +TEST_ASSERT_VAL ("wrong pinned" , !evsel->attr.pinned); + +return 0 ; +} +static int count_tracepoints (void ) {char events_path[PATH_MAX ];
@@ -1294,6 +1339,14 @@ static struct evlist_test test__events[] = { .name = "{instructions,branch-misses}:Su", .check = test__leader_sample2, }, + [40] = { + .name = "instructions:uDp", + .check = test__checkevent_pinned_modifier, + }, + [41] = { + .name = "{cycles,cache-misses,branch-misses}:D", + .check = test__pinned_group, + }, }; static struct evlist_test test__events_pmu[] = {
Signature · 1 line
1.8.1.2
Re: [PATCH v2 2/2] perf tests: Add tests of new pinned modifier
Quoted 4 lines · Michael Ellerman
Add a negative test to
test__checkevent_pmu_events()to get lots of coverage of the negative case, ie. when the modifier is not specified.Add a test of a single event, and of the group case.
nice! :)
jirka
Re: [PATCH v2 2/2] perf tests: Add tests of new pinned modifier
In reply to Michael EllermanQuoted 4 lines · Michael Ellerman
Add a negative test to
test__checkevent_pmu_events()to get lots of coverage of the negative case, ie. when the modifier is not specified.Add a test of a single event, and of the group case.
Thanks,
Namhyung
tip-bot for Michael Ellerman
merge bot[tip:perf/core] perf tests: Add tests of new pinned modifier
In reply to Michael Ellermanperf tests: Add tests of new pinned modifier
Commit-ID: c9ee780f2736b7a149658b0cd8c8389da23e190a Gitweb: http://git.kernel.org/tip/c9ee780f2736b7a149658b0cd8c8389da23e190a Author: Michael Ellerman <[email protected]> AuthorDate: Tue, 6 Aug 2013 23:28:06 +1000 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Wed, 7 Aug 2013 17:35:40 -0300
perf tests: Add tests of new pinned modifier
Add a negative test to test__checkevent_pmu_events() to get lots of coverage of the negative case, ie. when the modifier is not specified.
Add a test of a single event, and of the group case.
--- tools/perf/tests/parse-events.c | 53 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 53 insertions(+)
tools/perf/tests/parse-events.c +53 -0
@@ -452,6 +452,7 @@ static int test__checkevent_pmu_events(struct perf_evlist *evlist) evsel->attr.exclude_kernel);TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->attr.exclude_hv);TEST_ASSERT_VAL ("wrong precise_ip" , !evsel->attr.precise_ip); +TEST_ASSERT_VAL ("wrong pinned" , !evsel->attr.pinned);return 0 ; }
@@ -1070,6 +1071,50 @@ static int test__leader_sample2(struct perf_evlist *evlist __maybe_unused)return 0 ; } +static int test__checkevent_pinned_modifier (struct perf_evlist *evlist) +{ +struct perf_evsel *evsel =perf_evlist__first (evlist); + +TEST_ASSERT_VAL ("wrong exclude_user" , !evsel->attr.exclude_user); +TEST_ASSERT_VAL ("wrong exclude_kernel" , evsel->attr.exclude_kernel); +TEST_ASSERT_VAL ("wrong exclude_hv" , evsel->attr.exclude_hv); +TEST_ASSERT_VAL ("wrong precise_ip" , evsel->attr.precise_ip); +TEST_ASSERT_VAL ("wrong pinned" , evsel->attr.pinned); + +return test__checkevent_symbolic_name (evlist); +} + +static int test__pinned_group (struct perf_evlist *evlist) +{ +struct perf_evsel *evsel, *leader; + +TEST_ASSERT_VAL ("wrong number of entries" ,3 == evlist->nr_entries); + +/* cycles - group leader */ + evsel = leader =perf_evlist__first (evlist); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->attr.type); +TEST_ASSERT_VAL ("wrong config" , +PERF_COUNT_HW_CPU_CYCLES == evsel->attr.config); +TEST_ASSERT_VAL ("wrong group name" , !evsel->group_name); +TEST_ASSERT_VAL ("wrong leader" , evsel->leader == leader); +TEST_ASSERT_VAL ("wrong pinned" , evsel->attr.pinned); + +/* cache-misses - can not be pinned, but will go on with the leader */ + evsel =perf_evsel__next (evsel); +TEST_ASSERT_VAL ("wrong type" ,PERF_TYPE_HARDWARE == evsel->attr.type); +TEST_ASSERT_VAL ("wrong config" , +PERF_COUNT_HW_CACHE_MISSES == evsel->attr.config); +TEST_ASSERT_VAL ("wrong pinned" , !evsel->attr.pinned); + +/* branch-misses - ditto */ + evsel =perf_evsel__next (evsel); +TEST_ASSERT_VAL ("wrong config" , +PERF_COUNT_HW_BRANCH_MISSES == evsel->attr.config); +TEST_ASSERT_VAL ("wrong pinned" , !evsel->attr.pinned); + +return 0 ; +} +static int count_tracepoints (void ) {char events_path[PATH_MAX ];
@@ -1294,6 +1339,14 @@ static struct evlist_test test__events[] = { .name = "{instructions,branch-misses}:Su", .check = test__leader_sample2, }, + [40] = { + .name = "instructions:uDp", + .check = test__checkevent_pinned_modifier, + }, + [41] = { + .name = "{cycles,cache-misses,branch-misses}:D", + .check = test__pinned_group, + }, }; static struct evlist_test test__events_pmu[] = {
On Tue, Aug 06, 2013 at 11:28:05PM +1000, Michael Ellerman wrote:
SNIP
tools/perf/util/parse-events.c int parse_events__modifier_event(struct list_head *list, char *str, bool add) · line 823
Quoted 9 lines
tools/perf/util/parse-events.c +3
-0while (*p) {@@ -812,6 +818,9 @@ int parse_events__modifier_event(struct list_head *list, char *str, bool add) evsel->attr.exclude_guest = mod.eG; evsel->exclude_GH = mod.exclude_GH; evsel->sample_read = mod.sample_read; + +if (evsel->leader == evsel) + evsel->attr.pinned = mod.pinned;
nitpick, we have perf_evsel__is_group_leader for this check
jirka
Quoted 5 lines · Jiri Olsa
On Tue, Aug 06, 2013 at 11:28:05PM +1000, Michael Ellerman wrote:
SNIP
Quoted 1 line
Quoted 10 lines
tools/perf/util/parse-events.c int parse_events__modifier_event(struct list_head *list, char *str, bool add) · line 823
Quoted 8 lines
tools/perf/util/parse-events.c +3
-0while (*p) {@@ -812,6 +818,9 @@ int parse_events__modifier_event(struct list_head *list, char *str, bool add) evsel->attr.exclude_guest = mod.eG; evsel->exclude_GH = mod.exclude_GH; evsel->sample_read = mod.sample_read; + +if (evsel->leader == evsel) + evsel->attr.pinned = mod.pinned;nitpick, we have
perf_evsel__is_group_leaderfor this check
Ah darn, missed it.
Should I do a respin?
cheers
Em Wed, Aug 07, 2013 at 03:19:57PM +1000, Michael Ellerman escreveu:
Quoted 10 lines · Jiri Olsa
Quoted 5 lines · Michael Ellerman
tools/perf/util/parse-events.c int parse_events__modifier_event(struct list_head *list, char *str, bool add) · line 823
Quoted 2 lines
tools/perf/util/parse-events.c +2
-0+if (evsel->leader == evsel) + evsel->attr.pinned = mod.pinned;nitpick, we have
perf_evsel__is_group_leaderfor this checkAh darn, missed it.
Should I do a respin?
No need, I did it.
- Arnaldo
Quoted 8 lines · Arnaldo Carvalho de Melo
Em Wed, Aug 07, 2013 at 03:19:57PM +1000, Michael Ellerman escreveu:
Quoted 6 lines · Jiri Olsa
Quoted 5 lines · Michael Ellerman
tools/perf/util/parse-events.c int parse_events__modifier_event(struct list_head *list, char *str, bool add) · line 823
Quoted 2 lines
tools/perf/util/parse-events.c +2
-0+if (evsel->leader == evsel) + evsel->attr.pinned = mod.pinned;nitpick, we have
perf_evsel__is_group_leaderfor this check
Quoted 5 lines
Quoted 3 lines
Ah darn, missed it.
Should I do a respin?
No need, I did it.
Thanks.
cheers
Signature · 1 line
Sent from my Android phone with K-9 Mail. Please excuse my brevity.
Quoted 25 lines · Michael Ellerman
This commit adds support for a new modifier "D", which requests that the event, or group of events, be pinned to the PMU.
The "p" modifier is already taken for precise, and "P" may be used in future to mean "fully precise".
So we use "D", which stands for pinneD - and looks like a padlock, or if you're using the ":D" syntax perf smiles at you.
This is an oft-requested feature from our HW folks, who want to be able to run a large number of events, but also want 100% accurate results for instructions per cycle.
Comparison of results with and without pinning:
$ perf stat -e '{cycles,instructions}:D' -e cycles,instructions,...79,590,480,683 cycles # 0.000 GHz 166,123,716,524 instructions # 2.09 insns per cycle # 0.11 stalled cycles per insn79,352,134,463 cycles # 0.000 GHz [11.11%] 165,178,301,818 instructions # 2.08 insns per cycle # 0.11 stalled cycles per insn [11.13%]
Quoted 8 lines
As you can see although perf does a very good job of scaling the values in the non-pinned case, there is some small discrepancy.
The patch is fairly straight forward, the one detail is that we need to make sure we only request pinning for the group leader when we have a group.
Signed-off-by Michael Ellerman <[email protected]>
Tested-by/Acked-by: Jiri Olsa <[email protected]>
thanks,
jirka
Hi Michael,
Quoted 25 lines · Michael Ellerman
This commit adds support for a new modifier "D", which requests that the event, or group of events, be pinned to the PMU.
The "p" modifier is already taken for precise, and "P" may be used in future to mean "fully precise".
So we use "D", which stands for pinneD - and looks like a padlock, or if you're using the ":D" syntax perf smiles at you.
This is an oft-requested feature from our HW folks, who want to be able to run a large number of events, but also want 100% accurate results for instructions per cycle.
Comparison of results with and without pinning:
$ perf stat -e '{cycles,instructions}:D' -e cycles,instructions,...79,590,480,683 cycles # 0.000 GHz 166,123,716,524 instructions # 2.09 insns per cycle # 0.11 stalled cycles per insn79,352,134,463 cycles # 0.000 GHz [11.11%] 165,178,301,818 instructions # 2.08 insns per cycle # 0.11 stalled cycles per insn [11.13%]
Quoted 6 lines
As you can see although perf does a very good job of scaling the values in the non-pinned case, there is some small discrepancy.
The patch is fairly straight forward, the one detail is that we need to make sure we only request pinning for the group leader when we have a group.
Thanks,
Namhyung
tip-bot for Michael Ellerman
merge botperf tools: Add support for pinned modifier
Commit-ID: e9a7c414477d20c3cc56f90f29c35b06f0f15e25 Gitweb: http://git.kernel.org/tip/e9a7c414477d20c3cc56f90f29c35b06f0f15e25 Author: Michael Ellerman <[email protected]> AuthorDate: Tue, 6 Aug 2013 23:28:05 +1000 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Wed, 7 Aug 2013 17:35:40 -0300
perf tools: Add support for pinned modifier
This commit adds support for a new modifier "D", which requests that the event, or group of events, be pinned to the PMU.
The "p" modifier is already taken for precise, and "P" may be used in future to mean "fully precise".
So we use "D", which stands for pinneD - and looks like a padlock, or if you're using the ":D" syntax perf smiles at you.
This is an oft-requested feature from our HW folks, who want to be able to run a large number of events, but also want 100% accurate results for instructions per cycle.
Comparison of results with and without pinning:
$ perf stat -e '{cycles,instructions}:D' -e cycles,instructions,... 79,590,480,683 cycles # 0.000 GHz
166,123,716,524 instructions # 2.09 insns per cycle
# 0.11 stalled cycles per insn 79,352,134,463 cycles # 0.000 GHz [11.11%]
165,178,301,818 instructions # 2.08 insns per cycle
# 0.11 stalled cycles per insn [11.13%]As you can see although perf does a very good job of scaling the values in the non-pinned case, there is some small discrepancy.
The patch is fairly straight forward, the one detail is that we need to make sure we only request pinning for the group leader when we have a group.
[ Use perf_evsel__is_group_leader instead of open coded equivalent, as
suggested by Jiri Olsa ]
--- tools/perf/Documentation/perf-list.txt | 1 + tools/perf/util/parse-events.c | 11 ++++++++++- tools/perf/util/parse-events.l | 3 ++- 3 files changed, 13 insertions(+), 2 deletions(-)
tools/perf/Documentation/perf-list.txt +1 -0
@@ -30,6 +30,7 @@ counted. The following modifiers exist: H - host counting (not in KVM guests) p - precise level S - read sample value (PERF_SAMPLE_READ) + D - pin the event to the PMU The 'p' modifier can be used for specifying how precise the instruction address should be. The 'p' modifier can be specified multiple times:
tools/perf/util/parse-events.c +10 -1
@@ -688,6 +688,7 @@ struct event_modifier {int precise;int exclude_GH;int sample_read; +int pinned; }; static int get_event_modifier(struct event_modifier *mod, char *str,
@@ -700,6 +701,7 @@ static int get_event_modifier(struct event_modifier *mod, char *str,int eG = evsel ? evsel->attr.exclude_guest :0 ;int precise = evsel ? evsel->attr.precise_ip :0 ;int sample_read =0 ; +int pinned = evsel ? evsel->attr.pinned :0 ;int exclude = eu | ek | eh;int exclude_GH = evsel ? evsel->exclude_GH :0 ;
@@ -734,6 +736,8 @@ static int get_event_modifier(struct event_modifier *mod, char *str, eG =1 ; }else if (*str =='S' ) { sample_read =1 ; + }else if (*str =='D' ) { + pinned =1 ; }else break ;
@@ -761,6 +765,8 @@ static int get_event_modifier(struct event_modifier *mod, char *str, mod->precise = precise; mod->exclude_GH = exclude_GH; mod->sample_read = sample_read; + mod->pinned = pinned; +return 0 ; }
@@ -773,7 +779,7 @@ static int check_modifier(char *str)char *p = str;/* The sizeof includes 0 byte as well. */ -if (strlen (str) > (sizeof ("ukhGHpppS" ) -1 )) +if (strlen (str) > (sizeof ("ukhGHpppSD" ) -1 ))return -1 ;while (*p) {
@@ -812,6 +818,9 @@ int parse_events__modifier_event(struct list_head *list, char *str, bool add) evsel->attr.exclude_guest = mod.eG; evsel->exclude_GH = mod.exclude_GH; evsel->sample_read = mod.sample_read; + +if (perf_evsel__is_group_leader (evsel)) + evsel->attr.pinned = mod.pinned; }return 0 ;
tools/perf/util/parse-events.l +2 -1
@@ -82,7 +82,8 @@ num_hex 0x[a-fA-F0-9]+ num_raw_hex [a-fA-F0-9]+ name [a-zA-Z_*?][a-zA-Z0-9_*?]* name_minus [a-zA-Z_*?][a-zA-Z0-9\-_*?]* -modifier_event [ukhpGHS]+ +/* If you add a modifier you need to update check_modifier() */ +modifier_event [ukhpGHSD]+ modifier_bp [rwx]{1,3} %%