From: Kan Liang <kan.li...@linux.intel.com> When a task, which is sampled by a PEBS event with a fixed period, is sched_in, the fixed period will always be used as new period for counter. It's inaccurate, because the left period from last sched_out isn't taken into account.
The auto-reload feature is implicitly enabled for a PEBS event with a fixed period. The feature has specific intel_pmu_save_and_restart_reload(), which never records the left period. The period_left, which is used in perf_event_set_period() to calculate the new period, is always the user defined fixed period. With auto-reload, the counter always has a negative counter value. So the left period is -value. Update the period_left in intel_pmu_save_and_restart_reload(). Fixes: commit d31fc13fdcb2 ("perf/x86/intel: Fix event update for auto-reload") Signed-off-by: Kan Liang <kan.li...@linux.intel.com> --- arch/x86/events/intel/ds.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c index b7b01d7..a04b52b 100644 --- a/arch/x86/events/intel/ds.c +++ b/arch/x86/events/intel/ds.c @@ -1402,6 +1402,8 @@ intel_pmu_save_and_restart_reload(struct perf_event *event, int count) old = ((s64)(prev_raw_count << shift) >> shift); local64_add(new - old + count * period, &event->count); + local64_set(&hwc->period_left, -new); + perf_event_update_userpage(event); return 0; -- 2.7.4