Re: [PATCH 6.12.y] RISC-V: KVM: Serialize virtual interrupt pending state updates

[Sasha Levin <sashal@kernel.org>]

On Mon, Jul 27, 2026 at 10:33:48AM +0800, Xie Bo wrote:
>@@ -174,6 +187,9 @@ int kvm_riscv_vcpu_aia_get_csr(struct kvm_vcpu *vcpu,
> 			       unsigned long *out_val)
> {
> 	struct kvm_vcpu_aia_csr *csr = &vcpu->arch.aia_context.guest_csr;
>+#ifdef CONFIG_32BIT
>+	unsigned long flags;
>+#endif

This declaration landed in kvm_riscv_vcpu_aia_get_csr(), but the use site
is in kvm_riscv_vcpu_aia_set_csr():

>@@ -198,8 +214,12 @@ int kvm_riscv_vcpu_aia_set_csr(struct kvm_vcpu *vcpu,
>+		if (reg_num == KVM_REG_RISCV_CSR_AIA_REG(siph)) {
>+			raw_spin_lock_irqsave(&vcpu->arch.irqs_pending_lock, flags);

So on CONFIG_32BIT builds 'flags' is undeclared here and rv32 KVM fails to
build (KVM on riscv only depends on RISCV_SBI && MMU, so this is a real
configuration), plus the declaration in the getter is unused. The two
function prologues are identical, which is presumably how the hunk
anchored in the wrong one.

Could you send a v2 with the declaration moved into
kvm_riscv_vcpu_aia_set_csr()? Everything else in the patch matches
upstream d024a0a7879e ("RISC-V: KVM: Serialize virtual interrupt pending
state updates").

I've queued your 6.18.y patch. I'm holding the 6.6.y one until the
corrected 6.12.y lands, so that 6.6 doesn't ship a fix that 6.12 is
missing - both will go in together.

-- 
Thanks,
Sasha