Re: [PATCH] perf/x86/amd/uncore: Add group validation

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Group validation incorrectly counts disabled or errored siblings.
--

commit a3401bf46a6300cee39613401a4f3512876a47f1
Author: Sandipan Das <sandipan.das@amd.com>

perf/x86/amd/uncore: Add group validation

This patch introduces group validation in the pmu->event_init() path for
the AMD uncore driver. It counts the hardware events targeting the PMU
in an event group and ensures they do not exceed the available hardware
counters, preventing failures later in pmu->add().

> diff --git a/arch/x86/events/amd/uncore.c b/arch/x86/events/amd/uncore.c
> index 9293ce50574da..15330cbab6a69 100644
> --- a/arch/x86/events/amd/uncore.c
> +++ b/arch/x86/events/amd/uncore.c
[ ... ]
> +static bool amd_uncore_group_valid(struct perf_event *event)
> +{
> +	struct amd_uncore_pmu *pmu = event_to_amd_uncore_pmu(event);
> +	struct perf_event *leader = event->group_leader;
> +	struct perf_event *sibling;
> +	int counters = 0;
> +
> +	/* Ignore software events as they do not occupy hardware counters */
> +	if (leader->pmu == event->pmu && !is_software_event(leader))
> +		counters++;
> +
> +	for_each_sibling_event(sibling, leader) {
> +		if (sibling->pmu == event->pmu && !is_software_event(sibling))
> +			counters++;

[Severity: Medium]
Is it possible that disabled or errored siblings are incorrectly counted here?

Users might create oversized groups and manually multiplex them by toggling
events on and off. If this loop unconditionally increments the hardware
counter usage for all siblings, could it reject valid requests from userspace?

Unlike the core x86 and Intel uncore PMU drivers, it appears this fails to
ignore events with state <= PERF_EVENT_STATE_OFF.

> +	}
> +
> +	/*
> +	 * When pmu->event_init() is called, the event is yet to be linked to
> +	 * its leader's sibling list, so it is counted separately
> +	 */
> +	return (counters + 1) <= pmu->num_counters;
> +}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/0f2cff983d79521e153d211ce412fedbc4ed942c.1782211518.git.sandipan.das@amd.com?part=1