Re: [PATCH] KVM: arm64: Check ID_AA64DFR0_EL1.PMUVer in reset_pmevtyper()

[Oliver Upton <oupton@kernel.org>]

Hi Colton,

On Tue, Sep 29, 2026 at 10:04:11PM +0000, Colton Lewis wrote:
> When userspace initializes a vCPU with KVM_ARM_VCPU_PMU_V3 but sets
> ID_AA64DFR0_EL1.PMUVer to NI (0) or IMP_DEF (0xf, which is sanitized to
> NI), resetting the vCPU invokes reset_pmevtyper(). Because
> reset_pmevtyper() only checks kvm_vcpu_has_pmu(), it proceeds to call
> kvm_pmu_evtyper_mask(), which reads PMUVer == 0 from ID_AA64DFR0_EL1 and
> triggers a WARN_ONCE("Unknown PMU version 0") in __kvm_pmu_event_mask().
> 
> Check ID_AA64DFR0_EL1.PMUVer directly in reset_pmevtyper() via
> kvm_has_feat() instead of kvm_vcpu_has_pmu() so that PMEVTYPER<n>_EL0
> resets to 0 without querying the event mask when PMUv3 is not exposed to
> the guest.

So this goes a bit further than just this WARN_ONCE()... Ultimately if
FEAT_PMUv3 is hidden via the ID registers then we should expect KVM to
emulate these sysregs as UNDEF.

> Fixes: bc512d6a9b92 ("KVM: arm64: Make PMEVTYPER<n>_EL0.NSH RES0 if EL2 isn't advertised")
> Suggested-by: Oliver Upton <oupton@kernel.org>
> Cc: stable@vger.kernel.org

Avoiding a WARN_ONCE() is not stable material.

> Assisted-by: LLM
> Signed-off-by: Colton Lewis <coltonlewis@google.com>
> ---
>  arch/arm64/kvm/sys_regs.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> index a2f4e769a4282..0bb3c2f10a22c 100644
> --- a/arch/arm64/kvm/sys_regs.c
> +++ b/arch/arm64/kvm/sys_regs.c
> @@ -1060,7 +1060,7 @@ static u64 reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
>  static u64 reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
>  {
>  	/* This thing will UNDEF, who cares about the reset value? */
> -	if (!kvm_vcpu_has_pmu(vcpu))
> +	if (!kvm_has_feat(vcpu->kvm, ID_AA64DFR0_EL1, PMUVer, IMP))
>  		return 0;

I have this diff locally, separating the visibility change into a
separate patch. Do you want to give it a spin?

Thanks,
Oliver

diff --git a/arch/arm64/kvm/pmu-emul.c b/arch/arm64/kvm/pmu-emul.c
index 5b1af7e2176f..e2b28dbc0451 100644
--- a/arch/arm64/kvm/pmu-emul.c
+++ b/arch/arm64/kvm/pmu-emul.c
@@ -50,8 +50,11 @@ static u32 __kvm_pmu_event_mask(unsigned int pmuver)
 	case ID_AA64DFR0_EL1_PMUVer_V3P5:
 	case ID_AA64DFR0_EL1_PMUVer_V3P7:
 		return GENMASK(15, 0);
-	default:		/* Shouldn't be here, just for sanity */
+	default:
+		/* Shouldn't be here, just for sanity */
 		WARN_ONCE(1, "Unknown PMU version %d\n", pmuver);
+		fallthrough;
+	case ID_AA64DFR0_EL1_PMUVer_NI:
 		return 0;
 	}
 }
diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
index 44aae52c473d..8315a188676f 100644
--- a/arch/arm64/kvm/sys_regs.c
+++ b/arch/arm64/kvm/sys_regs.c
@@ -1028,7 +1028,7 @@ static unsigned int hidden_visibility(const struct kvm_vcpu *vcpu,
 static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
 				   const struct sys_reg_desc *r)
 {
-	if (kvm_vcpu_has_pmu(vcpu))
+	if (kvm_has_feat(vcpu->kvm, ID_AA64DFR0_EL1, PMUVer, IMP))
 		return 0;
 
 	return REG_HIDDEN;
@@ -1058,10 +1058,6 @@ static u64 reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 
 static u64 reset_pmevtyper(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
 {
-	/* This thing will UNDEF, who cares about the reset value? */
-	if (!kvm_vcpu_has_pmu(vcpu))
-		return 0;
-
 	reset_unknown(vcpu, r);
 	__vcpu_rmw_sys_reg(vcpu, r->reg, &=, kvm_pmu_evtyper_mask(vcpu->kvm));