Re: [PATCH 2/3] perf: arm_spe: Suppress redundant ITRACE start records

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] ITRACE_START records are not suppressed for inherited events due to a child/parent event state mismatch.
--

commit 0683c3b2471a719bb066f5dd9a93d7a7f192399a
Author: Leo Yan <leo.yan@arm.com>

perf: arm_spe: Suppress redundant ITRACE start records

This commit calls perf_event_itrace_started() at the start of an event to
prevent the perf core from emitting redundant PERF_RECORD_ITRACE_START
records on subsequent schedule-ins.

> diff --git a/drivers/perf/arm_spe_pmu.c b/drivers/perf/arm_spe_pmu.c
> index b70f3fedf0282..40314ef8b731e 100644
> --- a/drivers/perf/arm_spe_pmu.c
> +++ b/drivers/perf/arm_spe_pmu.c
> @@ -890,6 +890,8 @@ static void arm_spe_pmu_start(struct perf_event *event, int flags)
>  		return;
>  	}
>  
> +	perf_event_itrace_started(event);
> +

[Severity: Medium]
Does this correctly suppress redundant records for inherited events?

The core event scheduling path in perf_log_itrace_start() internally resolves
to event->parent before checking for the PERF_ATTACH_ITRACE flag. 

Since perf_event_itrace_started() only mutates the local child event's
attach_state, the parent event's flag remains permanently unset. 

Won't this cause the perf core to continue emitting redundant ITRACE_START
records on every subsequent schedule-in for inherited events, negating the
intended fix?

>  	reg = arm_spe_event_to_pmsfcr(event);
>  	write_sysreg_s(reg, SYS_PMSFCR_EL1);
>

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260901-perf_suppress_itrace_start_records-v1-0-c4fbc6860859@arm.com?part=2