Re: [PATCH v3 09/22] KVM: arm64: Correct kvm_arm_pmu_get_max_counters()

[Marc Zyngier]

On Thu, 26 Jun 2025 21:04:45 +0100,
Colton Lewis <coltonlewis@xxxxxxxxxx> wrote:
> 
> Since cntr_mask is modified when the PMU is partitioned to remove some
> bits, make sure the missing counters are added back to get the right
> total.

Please fix the subject of the patch to be more descriptive. It is
worded like a bug fix, while it really is only a step in the patch
series.

Something like "Take partitioning into account for max number of
counters" would go a long way.

> 
> Signed-off-by: Colton Lewis <coltonlewis@xxxxxxxxxx>
> ---
>  arch/arm64/kvm/pmu.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/kvm/pmu.c b/arch/arm64/kvm/pmu.c
> index 79b7ea037153..67216451b8ce 100644
> --- a/arch/arm64/kvm/pmu.c
> +++ b/arch/arm64/kvm/pmu.c
> @@ -533,6 +533,8 @@ static bool pmu_irq_is_valid(struct kvm *kvm, int irq)
>  u8 kvm_arm_pmu_get_max_counters(struct kvm *kvm)
>  {
>  	struct arm_pmu *arm_pmu = kvm->arch.arm_pmu;
> +	u8 counters;
> +

nit: superfluous blank line.
>  
>  	/*
>  	 * PMUv3 requires that all event counters are capable of counting any
> @@ -545,7 +547,12 @@ u8 kvm_arm_pmu_get_max_counters(struct kvm *kvm)
>  	 * The arm_pmu->cntr_mask considers the fixed counter(s) as well.
>  	 * Ignore those and return only the general-purpose counters.
>  	 */
> -	return bitmap_weight(arm_pmu->cntr_mask, ARMV8_PMU_MAX_GENERAL_COUNTERS);
> +	counters = bitmap_weight(arm_pmu->cntr_mask, ARMV8_PMU_MAX_GENERAL_COUNTERS);
> +
> +	if (kvm_pmu_is_partitioned(arm_pmu))
> +		counters += arm_pmu->hpmn_max;

Why the check? Why can't we rely on hpmn_max to always give us the
correct value?

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.