Currently, perf-stat with armv8_pmu events with a workload is broken.
This patch set just fixes that.
Before the patch set:
[root@localhost hulk]# tools/perf/perf stat -e armv8_pmuv3_0/ll_cache_rd/,armv8_pmuv3_0/ll_cache_miss_rd/ ls > /dev/null
Segmentation fault
After the patch set:
[root@localhost hulk]# tools/perf/perf stat -e armv8_pmuv3_0/ll_cache_rd/,armv8_pmuv3_0/ll_cache_miss_rd/ ls > /dev/null
Performance counter stats for 'ls':
39,882 armv8_pmuv3_0/ll_cache_rd/
9,639 armv8_pmuv3_0/ll_cache_miss_rd/
0.001416690 seconds time elapsed
0.001469000 seconds user
0.000000000 seconds sys
Wei Li (2):
perf stat: Fix segfault when counting armv8 PMU events
perf stat: Unbreak perf stat with armv8 PMU events
tools/lib/perf/include/internal/evlist.h | 1 +
tools/perf/builtin-stat.c | 37 ++++++++++++++++--------
tools/perf/util/evlist.c | 23 ++++++++++++++-
3 files changed, 48 insertions(+), 13 deletions(-)
--
2.17.1
After the segfault is fixed, perf-stat with armv8_pmu events with a
workload is still broken:
[root@localhost hulk]# tools/perf/perf stat -e armv8_pmuv3_0/ll_cache_rd/,armv8_pmuv3_0/ll_cache_miss_rd/ ls > /dev/null
Performance counter stats for 'ls':
<not counted> armv8_pmuv3_0/ll_cache_rd/ (0.00%)
<not counted> armv8_pmuv3_0/ll_cache_miss_rd/ (0.00%)
0.002052670 seconds time elapsed
0.000000000 seconds user
0.002086000 seconds sys
In fact, while the event will be opened per-thread,
create_perf_stat_counter() is called as many times as the count of cpu
in the evlist's cpumap, and lost all the file descriptors except the
last one. If this counter is not scheduled during the period of time,
it will be "not counted".
Add the process to don't open the needless events in such situation.
Fixes: 4804e0111662 ("perf stat: Use affinity for opening events")
Signed-off-by: Wei Li <[email protected]>
---
tools/perf/builtin-stat.c | 36 +++++++++++++++++++++++-------------
1 file changed, 23 insertions(+), 13 deletions(-)
diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c
index 6e6ceacce634..9a43b3de26d1 100644
--- a/tools/perf/builtin-stat.c
+++ b/tools/perf/builtin-stat.c
@@ -712,6 +712,7 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
struct affinity affinity;
int i, cpu;
bool second_pass = false;
+ bool open_per_thread = false;
if (forks) {
if (perf_evlist__prepare_workload(evsel_list, &target, argv, is_pipe,
@@ -726,16 +727,17 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
perf_evlist__set_leader(evsel_list);
if (!(target__has_cpu(&target) && !target__has_per_thread(&target)))
- evsel_list->core.open_per_thread = true;
+ evsel_list->core.open_per_thread = open_per_thread = true;
if (affinity__setup(&affinity) < 0)
return -1;
evlist__for_each_cpu (evsel_list, i, cpu) {
- affinity__set(&affinity, cpu);
+ if (!open_per_thread)
+ affinity__set(&affinity, cpu);
evlist__for_each_entry(evsel_list, counter) {
- if (evsel__cpu_iter_skip(counter, cpu))
+ if (!open_per_thread && evsel__cpu_iter_skip(counter, cpu))
continue;
if (counter->reset_group || counter->errored)
continue;
@@ -753,7 +755,8 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
if ((errno == EINVAL || errno == EBADF) &&
counter->leader != counter &&
counter->weak_group) {
- perf_evlist__reset_weak_group(evsel_list, counter, false);
+ perf_evlist__reset_weak_group(evsel_list, counter,
+ open_per_thread);
assert(counter->reset_group);
second_pass = true;
continue;
@@ -773,6 +776,9 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
}
counter->supported = true;
}
+
+ if (open_per_thread)
+ break;
}
if (second_pass) {
@@ -782,20 +788,22 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
*/
evlist__for_each_cpu(evsel_list, i, cpu) {
- affinity__set(&affinity, cpu);
- /* First close errored or weak retry */
- evlist__for_each_entry(evsel_list, counter) {
- if (!counter->reset_group && !counter->errored)
- continue;
- if (evsel__cpu_iter_skip_no_inc(counter, cpu))
- continue;
- perf_evsel__close_cpu(&counter->core, counter->cpu_iter);
+ if (!open_per_thread) {
+ affinity__set(&affinity, cpu);
+ /* First close errored or weak retry */
+ evlist__for_each_entry(evsel_list, counter) {
+ if (!counter->reset_group && !counter->errored)
+ continue;
+ if (evsel__cpu_iter_skip_no_inc(counter, cpu))
+ continue;
+ perf_evsel__close_cpu(&counter->core, counter->cpu_iter);
+ }
}
/* Now reopen weak */
evlist__for_each_entry(evsel_list, counter) {
if (!counter->reset_group && !counter->errored)
continue;
- if (evsel__cpu_iter_skip(counter, cpu))
+ if (!open_per_thread && evsel__cpu_iter_skip(counter, cpu))
continue;
if (!counter->reset_group)
continue;
@@ -817,6 +825,8 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx)
}
counter->supported = true;
}
+ if (open_per_thread)
+ break;
}
}
affinity__cleanup(&affinity);
--
2.17.1