perf evlist: Allow setting arbitrary leader
Ian Rogers
reviewerPATCH v2 1/2 perf evlist: Allow setting arbitrary leader
The leader of a group is the first, but allow it to be an arbitrary list member so that for Intel topdown events slots may always be the group leader.
--- tools/lib/perf/evlist.c | 15 +++++++++------ tools/lib/perf/include/internal/evlist.h | 2 +- tools/perf/util/parse-events.c | 2 +- 3 files changed, 11 insertions(+), 8 deletions(-)
tools/lib/perf/evlist.c +9 -6
@@ -643,14 +643,14 @@ perf_evlist__next_mmap(struct perf_evlist *evlist, struct perf_mmap *map,return overwrite ? evlist->mmap_ovw_first : evlist->mmap_first; } -void __perf_evlist__set_leader (struct list_head *list) +void __perf_evlist__set_leader (struct list_head *list,struct perf_evsel *leader) { -struct perf_evsel *evsel, *leader; +struct perf_evsel *first, *last, *evsel; - leader = list_entry(list->next, struct perf_evsel, node); - evsel = list_entry(list->prev, struct perf_evsel, node); + first = list_entry(list->next, struct perf_evsel, node); + last = list_entry(list->prev, struct perf_evsel, node); - leader->nr_members = evsel->idx - leader->idx +1 ; + leader->nr_members = last->idx - first->idx +1 ;__perf_evlist__for_each_entry (list, evsel) evsel->leader = leader;
@@ -659,7 +659,10 @@ void __perf_evlist__set_leader(struct list_head *list)void perf_evlist__set_leader (struct perf_evlist *evlist) {if (evlist->nr_entries) { +struct perf_evsel *first =list_entry (evlist->entries.next, + struct perf_evsel, node); + evlist->nr_groups = evlist->nr_entries >1 ?1 :0 ; -__perf_evlist__set_leader (&evlist->entries); +__perf_evlist__set_leader (&evlist->entries, first); } }
tools/lib/perf/include/internal/evlist.h +1 -1
@@ -127,5 +127,5 @@ int perf_evlist__id_add_fd(struct perf_evlist *evlist,void perf_evlist__reset_id_hash (struct perf_evlist *evlist); -void __perf_evlist__set_leader (struct list_head *list); +void __perf_evlist__set_leader (struct list_head *list,struct perf_evsel *leader);#endif /* __LIBPERF_INTERNAL_EVLIST_H */
tools/perf/util/parse-events.c +1 -1
@@ -1834,8 +1834,8 @@ void parse_events__set_leader(char *name, struct list_head *list,if (parse_events__set_leader_for_uncore_aliase (name, list, parse_state))return ; -__perf_evlist__set_leader (list); leader = list_entry(list->next, struct evsel, core.node); +__perf_evlist__set_leader (list, &leader->core); leader->group_name = name ?strdup (name) :NULL ; }
Signature · 1 line
2.34.0.rc2.393.gf8c9666880-goog
Ian Rogers
reviewerPATCH v2 2/2 perf parse-events: Architecture specific leader override
Currently topdown events must appear after a slots event:
$ perf stat -e '{slots,topdown-fe-bound}' /bin/true Performance counter stats for '/bin/true':
3,183,090 slots
986,133 topdown-fe-boundReversing the events yields:
$ perf stat -e '{topdown-fe-bound,slots}' /bin/true
Error:
The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (topdown-fe-bound).For metrics the order of events is determined by iterating over a hashmap, and so slots isn't guaranteed to be first which can yield this error.
Change the set_leader in parse-events, called when a group is closed, so that rather than always making the first event the leader, if the slots event exists then it is made the leader. It is then moved to the head of the evlist otherwise it won't be opened in the correct order.
The result is:
$ perf stat -e '{topdown-fe-bound,slots}' /bin/true Performance counter stats for '/bin/true':
3,274,795 slots
1,001,702 topdown-fe-boundA problem with this approach is the slots event is identified by name, names can be overwritten like 'cpu/slots,name=foo/' and this causes the leader change to fail.
The change also modifies and fixes mixed groups like, with the change:
$ perf stat -e '{instructions,slots,topdown-fe-bound}' -a -- sleep 2Performance counter stats for 'system wide':
5574985410 slots
971981616 instructions
1348461887 topdown-fe-bound2.001263120 seconds time elapsed
Without the change:
$ perf stat -e '{instructions,slots,topdown-fe-bound}' -a -- sleep 2Performance counter stats for 'system wide':
<not counted> instructions
<not counted> slots
<not supported> topdown-fe-bound2.006247990 seconds time elapsed
Something that may be undesirable here is that the events are reordered in the output.
v2. Move the list_move operation into parse_events__set_leader as
suggested by Jiri Olsa <[email protected]>.
--- tools/perf/arch/x86/util/evlist.c | 17 +++++++++++++++++ tools/perf/util/evlist.h | 1 + tools/perf/util/parse-events.c | 8 +++++++- 3 files changed, 25 insertions(+), 1 deletion(-)
tools/perf/arch/x86/util/evlist.c +17 -0
@@ -17,3 +17,20 @@ int arch_evlist__add_default_attrs(struct evlist *evlist)else return parse_events (evlist,TOPDOWN_L1_EVENTS ,NULL ); } + +struct evsel *arch_evlist__leader (struct list_head *list) +{ +struct evsel *evsel, *first; + + first = list_entry(list->next, struct evsel, core.node); + +if (!pmu_have_event ("cpu" ,"slots" )) +return first; + +__evlist__for_each_entry (list, evsel) { +if (evsel->pmu_name && !strcmp (evsel->pmu_name,"cpu" ) && + evsel->name &&strstr (evsel->name,"slots" )) +return evsel; + } +return first; +}
tools/perf/util/evlist.h +1 -0
@@ -110,6 +110,7 @@ int __evlist__add_default_attrs(struct evlist *evlist, __evlist__add_default_attrs(evlist, array, ARRAY_SIZE(array))int arch_evlist__add_default_attrs (struct evlist *evlist); +struct evsel *arch_evlist__leader (struct list_head *list);int evlist__add_dummy (struct evlist *evlist);
tools/perf/util/parse-events.c +7 -1
@@ -1821,6 +1821,11 @@ parse_events__set_leader_for_uncore_aliase(char *name, struct list_head *list,return ret; } +__weak struct evsel *arch_evlist__leader (struct list_head *list) +{ + return list_entry(list->next, struct evsel, core.node); +} +void parse_events__set_leader (char *name,struct list_head *list,struct parse_events_state *parse_state) {
@@ -1834,9 +1839,10 @@ void parse_events__set_leader(char *name, struct list_head *list,if (parse_events__set_leader_for_uncore_aliase (name, list, parse_state))return ; - leader = list_entry(list->next, struct evsel, core.node); + leader =arch_evlist__leader (list);__perf_evlist__set_leader (list, &leader->core); leader->group_name = name ?strdup (name) :NULL ; +list_move (&leader->core.node, list); }/* list_event is assumed to point to malloc'ed memory */
Signature · 1 line
2.34.0.rc2.393.gf8c9666880-goog
Re: [PATCH v2 2/2] perf parse-events: Architecture specific leader override
Quoted 19 lines · Ian Rogers
Currently topdown events must appear after a slots event:
$ perf stat -e '{slots,topdown-fe-bound}' /bin/truePerformance counter stats for '
/bin/true':3,183,090 slots 986,133 topdown-fe-boundReversing the events yields:
$ perf stat -e '{topdown-fe-bound,slots}' /bin/true Error: The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (topdown-fe-bound).For metrics the order of events is determined by iterating over a hashmap, and so slots isn't guaranteed to be first which can yield this error.
Quoted 8 lines
Change the
set_leaderin parse-events, called when a group is closed, so that rather than always making the first event the leader, if the slots event exists then it is made the leader. It is then moved to the head of the evlist otherwise it won't be opened in the correct order.The result is:
$ perf stat -e '{topdown-fe-bound,slots}' /bin/true
Just curious - does this just affect topdown events? I think x86 is the only arch which has them.
Thanks,
John
Ian Rogers
reviewerRe: [PATCH v2 2/2] perf parse-events: Architecture specific leader override
Quoted 26 lines · John Garry
Quoted 24 lines · Ian Rogers
Currently topdown events must appear after a slots event:
$ perf stat -e '{slots,topdown-fe-bound}' /bin/truePerformance counter stats for '
/bin/true':3,183,090 slots 986,133 topdown-fe-boundReversing the events yields:
$ perf stat -e '{topdown-fe-bound,slots}' /bin/true Error: The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (topdown-fe-bound).For metrics the order of events is determined by iterating over a hashmap, and so slots isn't guaranteed to be first which can yield this error.
Change the
set_leaderin parse-events, called when a group is closed, so that rather than always making the first event the leader, if the slots event exists then it is made the leader. It is then moved to the head of the evlist otherwise it won't be opened in the correct order.
Quoted 6 lines
Quoted 3 lines
The result is:
$ perf stat -e '{topdown-fe-bound,slots}' /bin/trueJust curious - does this just affect topdown events? I think x86 is the only arch which has them.
The change is specific to x86 as the weak symbol override only happens on x86. For x86 it only applies if the cpu/slots/ event exists. In the future it may be used for more than just this - Kajol Jain mentioned a similar problem for powerpc:
https://lore.kernel.org/lkml/[email protected]/
Thanks,
Ian
Quoted 2 lines
Thanks,
John
Arnaldo Carvalho de Melo
maintainerRe: [PATCH v2 2/2] perf parse-events: Architecture specific leader override
In reply to Ian RogersEm Thu, Nov 18, 2021 at 02:06:47PM -0800, Ian Rogers escreveu:
Quoted 80 lines
Currently topdown events must appear after a slots event:
$ perf stat -e '{slots,topdown-fe-bound}' /bin/truePerformance counter stats for '
/bin/true':3,183,090 slots 986,133 topdown-fe-boundReversing the events yields:
$ perf stat -e '{topdown-fe-bound,slots}' /bin/true Error: The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (topdown-fe-bound).For metrics the order of events is determined by iterating over a hashmap, and so slots isn't guaranteed to be first which can yield this error.
Change the
set_leaderin parse-events, called when a group is closed, so that rather than always making the first event the leader, if the slots event exists then it is made the leader. It is then moved to the head of the evlist otherwise it won't be opened in the correct order.The result is:
$ perf stat -e '{topdown-fe-bound,slots}' /bin/truePerformance counter stats for '
/bin/true':3,274,795 slots 1,001,702 topdown-fe-boundA problem with this approach is the slots event is identified by name, names can be overwritten like 'cpu/slots,name=foo/' and this causes the leader change to fail.
The change also modifies and fixes mixed groups like, with the change:
$ perf stat -e '{instructions,slots,topdown-fe-bound}' -a -- sleep 2Performance counter stats for 'system wide':
5574985410 slots 971981616 instructions 1348461887 topdown-fe-bound2.001263120 seconds time elapsedWithout the change:
$ perf stat -e '{instructions,slots,topdown-fe-bound}' -a -- sleep 2Performance counter stats for 'system wide':
<not counted> instructions <not counted> slots <not supported> topdown-fe-bound2.006247990 seconds time elapsedSomething that may be undesirable here is that the events are reordered in the output.
v2. Move the
list_moveoperation intoparse_events__set_leaderas
suggested by Jiri Olsa <[email protected]>.Reviewed-by Kajol Jain<[email protected]>Signed-off-by Ian Rogers <[email protected]>--- tools/perf/arch/x86/util/evlist.c | 17 +++++++++++++++++ tools/perf/util/evlist.h | 1 + tools/perf/util/parse-events.c | 8 +++++++- 3 files changed, 25 insertions(+), 1 deletion(-)tools/perf/arch/x86/util/evlist.c +0
-0
tools/perf/arch/x86/util/evlist.c int arch_evlist__add_default_attrs(struct evlist *evlist) · line 25
Quoted 9 lines
tools/perf/arch/x86/util/evlist.c +6
-0@@ -17,3 +17,20 @@ int arch_evlist__add_default_attrs(struct evlist *evlist)else return parse_events (evlist,TOPDOWN_L1_EVENTS ,NULL ); } + +struct evsel *arch_evlist__leader (struct list_head *list) +{ +struct evsel *evsel, *first; + + first = list_entry(list->next, struct evsel, core.node);
Can we make this:
struct *evsel, *first = list_first_entry(list, struct evsel, core.node);
?
Quoted 53 lines
tools/perf/arch/x86/util/evlist.c +11
-0+ +if (!pmu_have_event ("cpu" ,"slots" )) +return first; + +__evlist__for_each_entry (list, evsel) { +if (evsel->pmu_name && !strcmp (evsel->pmu_name,"cpu" ) && + evsel->name &&strstr (evsel->name,"slots" )) +return evsel; + } +return first; +}tools/perf/util/evlist.h +1
-0@@ -110,6 +110,7 @@ int __evlist__add_default_attrs(struct evlist *evlist, __evlist__add_default_attrs(evlist, array, ARRAY_SIZE(array))int arch_evlist__add_default_attrs (struct evlist *evlist); +struct evsel *arch_evlist__leader (struct list_head *list);int evlist__add_dummy (struct evlist *evlist);tools/perf/util/parse-events.c +7
-1@@ -1821,6 +1821,11 @@ parse_events__set_leader_for_uncore_aliase(char *name, struct list_head *list,return ret; } +__weak struct evsel *arch_evlist__leader (struct list_head *list) +{ + return list_entry(list->next, struct evsel, core.node); +} +void parse_events__set_leader (char *name,struct list_head *list,struct parse_events_state *parse_state) {@@ -1834,9 +1839,10 @@ void parse_events__set_leader(char *name, struct list_head *list,if (parse_events__set_leader_for_uncore_aliase (name, list, parse_state))return ; - leader = list_entry(list->next, struct evsel, core.node); + leader =arch_evlist__leader (list);__perf_evlist__set_leader (list, &leader->core); leader->group_name = name ?strdup (name) :NULL ; +list_move (&leader->core.node, list); }/* list_event is assumed to point to malloc'ed memory */ Signature · 1 line
2.34.0.rc2.393.gf8c9666880-goog
Signature · 1 line
- Arnaldo
Quoted 6 lines · Ian Rogers
The leader of a group is the first, but allow it to be an arbitrary list member so that for Intel topdown events slots may always be the group leader.
Reviewed-by Kajol Jain<[email protected]>Signed-off-by Ian Rogers <[email protected]>
hi,
for both patches:
thanks,
jirka
Quoted 70 lines
--- tools/lib/perf/evlist.c | 15 +++++++++------ tools/lib/perf/include/internal/evlist.h | 2 +- tools/perf/util/parse-events.c | 2 +- 3 files changed, 11 insertions(+), 8 deletions(-)tools/lib/perf/evlist.c +9
-6@@ -643,14 +643,14 @@ perf_evlist__next_mmap(struct perf_evlist *evlist, struct perf_mmap *map,return overwrite ? evlist->mmap_ovw_first : evlist->mmap_first; } -void __perf_evlist__set_leader (struct list_head *list) +void __perf_evlist__set_leader (struct list_head *list,struct perf_evsel *leader) { -struct perf_evsel *evsel, *leader; +struct perf_evsel *first, *last, *evsel; - leader = list_entry(list->next, struct perf_evsel, node); - evsel = list_entry(list->prev, struct perf_evsel, node); + first = list_entry(list->next, struct perf_evsel, node); + last = list_entry(list->prev, struct perf_evsel, node); - leader->nr_members = evsel->idx - leader->idx +1 ; + leader->nr_members = last->idx - first->idx +1 ;__perf_evlist__for_each_entry (list, evsel) evsel->leader = leader;@@ -659,7 +659,10 @@ void __perf_evlist__set_leader(struct list_head *list)void perf_evlist__set_leader (struct perf_evlist *evlist) {if (evlist->nr_entries) { +struct perf_evsel *first =list_entry (evlist->entries.next, + struct perf_evsel, node); + evlist->nr_groups = evlist->nr_entries >1 ?1 :0 ; -__perf_evlist__set_leader (&evlist->entries); +__perf_evlist__set_leader (&evlist->entries, first); } }tools/lib/perf/include/internal/evlist.h +1
-1@@ -127,5 +127,5 @@ int perf_evlist__id_add_fd(struct perf_evlist *evlist,void perf_evlist__reset_id_hash (struct perf_evlist *evlist); -void __perf_evlist__set_leader (struct list_head *list); +void __perf_evlist__set_leader (struct list_head *list,struct perf_evsel *leader);#endif /* __LIBPERF_INTERNAL_EVLIST_H */ tools/perf/util/parse-events.c +1
-1@@ -1834,8 +1834,8 @@ void parse_events__set_leader(char *name, struct list_head *list,if (parse_events__set_leader_for_uncore_aliase (name, list, parse_state))return ; -__perf_evlist__set_leader (list); leader = list_entry(list->next, struct evsel, core.node); +__perf_evlist__set_leader (list, &leader->core); leader->group_name = name ?strdup (name) :NULL ; }Signature · 1 line
2.34.0.rc2.393.gf8c9666880-goog
Arnaldo Carvalho de Melo
maintainerEm Thu, Nov 18, 2021 at 02:06:46PM -0800, Ian Rogers escreveu:
Quoted 20 lines
The leader of a group is the first, but allow it to be an arbitrary list member so that for Intel topdown events slots may always be the group leader.
Reviewed-by Kajol Jain<[email protected]>Signed-off-by Ian Rogers <[email protected]>--- tools/lib/perf/evlist.c | 15 +++++++++------ tools/lib/perf/include/internal/evlist.h | 2 +- tools/perf/util/parse-events.c | 2 +- 3 files changed, 11 insertions(+), 8 deletions(-)tools/lib/perf/evlist.c +0
-0@@ -643,14 +643,14 @@ perf_evlist__next_mmap(struct perf_evlist *evlist, struct perf_mmap *map,return overwrite ? evlist->mmap_ovw_first : evlist->mmap_first; }
tools/lib/perf/evlist.c perf_evlist__next_mmap(struct perf_evlist *evlist, struct perf_mmap *map, · line 651
Quoted 10 lines
tools/lib/perf/evlist.c +4
-4-void __perf_evlist__set_leader (struct list_head *list) +void __perf_evlist__set_leader (struct list_head *list,struct perf_evsel *leader) { -struct perf_evsel *evsel, *leader; +struct perf_evsel *first, *last, *evsel; - leader = list_entry(list->next, struct perf_evsel, node); - evsel = list_entry(list->prev, struct perf_evsel, node); + first = list_entry(list->next, struct perf_evsel, node); + last = list_entry(list->prev, struct perf_evsel, node);
We have list_first_entry() and list_last_entry():
For instance:
tools/perf/util/parse-events.c: leader = list_first_entry(list, struct evsel, core.node);
Can we use it here?
Quoted 33 lines
tools/lib/perf/evlist.c +5
-2- leader->nr_members = evsel->idx - leader->idx +1 ; + leader->nr_members = last->idx - first->idx +1 ;__perf_evlist__for_each_entry (list, evsel) evsel->leader = leader;@@ -659,7 +659,10 @@ void __perf_evlist__set_leader(struct list_head *list)void perf_evlist__set_leader (struct perf_evlist *evlist) {if (evlist->nr_entries) { +struct perf_evsel *first =list_entry (evlist->entries.next, + struct perf_evsel, node); + evlist->nr_groups = evlist->nr_entries >1 ?1 :0 ; -__perf_evlist__set_leader (&evlist->entries); +__perf_evlist__set_leader (&evlist->entries, first); } }tools/lib/perf/include/internal/evlist.h +1
-1@@ -127,5 +127,5 @@ int perf_evlist__id_add_fd(struct perf_evlist *evlist,void perf_evlist__reset_id_hash (struct perf_evlist *evlist); -void __perf_evlist__set_leader (struct list_head *list); +void __perf_evlist__set_leader (struct list_head *list,struct perf_evsel *leader);#endif /* __LIBPERF_INTERNAL_EVLIST_H */ tools/perf/util/parse-events.c +0
-0
tools/perf/util/parse-events.c void parse_events__set_leader(char *name, struct list_head *list, · line 1839
Quoted 7 lines
tools/perf/util/parse-events.c +1
-1@@ -1834,8 +1834,8 @@ void parse_events__set_leader(char *name, struct list_head *list,if (parse_events__set_leader_for_uncore_aliase (name, list, parse_state))return ; -__perf_evlist__set_leader (list); leader = list_entry(list->next, struct evsel, core.node); +__perf_evlist__set_leader (list, &leader->core); leader->group_name = name ?strdup (name) :NULL ;
Ditto
Quoted 4 lines
tools/perf/util/parse-events.c +0
-0}Signature · 1 line
2.34.0.rc2.393.gf8c9666880-goog
Signature · 1 line
- Arnaldo