Re: [PATCH v3 00/26] KVM: x86: Halt and APICv overhaul

From: Sean Christopherson
Date: Thu Dec 09 2021 - 10:45:32 EST


On Thu, Dec 09, 2021, Maxim Levitsky wrote:
> On Thu, 2021-12-09 at 15:29 +0100, Paolo Bonzini wrote:
> > On 12/9/21 01:02, Sean Christopherson wrote:
> > > RDX, a.k.a. ir_data is NULL. This check in svm_ir_list_add()
> > >
> > > if (pi->ir_data && (pi->prev_ga_tag != 0)) {
> > >
> > > implies pi->ir_data can be NULL, but neither avic_update_iommu_vcpu_affinity()
> > > nor amd_iommu_update_ga() check ir->data for NULL.
> > >
> > > amd_ir_set_vcpu_affinity() returns "success" without clearing pi.is_guest_mode
> > >
> > > /* Note:
> > > * This device has never been set up for guest mode.
> > > * we should not modify the IRTE
> > > */
> > > if (!dev_data || !dev_data->use_vapic)
> > > return 0;
> > >
> > > so it's plausible svm_ir_list_add() could add to the list with a NULL pi->ir_data.
> > >
> > > But none of the relevant code has seen any meaningful changes since 5.15, so odds
> > > are good I broke something :-/
>
> Doesn't reproduce here yet even with my iommu changes :-(
> Oh well.

Hmm, which suggests it could be an existing corner case.

Based on the above, this seems prudent and correct:

@@ -747,7 +754,7 @@ static int svm_ir_list_add(struct vcpu_svm *svm, struct amd_iommu_pi_data *pi)
* so we need to check here if it's already been * added
* to the ir_list.
*/
- if (pi->ir_data && (pi->prev_ga_tag != 0)) {
+ if (pi->prev_ga_tag != 0) {
struct kvm *kvm = svm->vcpu.kvm;
u32 vcpu_id = AVIC_GATAG_TO_VCPUID(pi->prev_ga_tag);
struct kvm_vcpu *prev_vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
@@ -877,7 +884,7 @@ int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
* we can reference to them directly when we update vcpu
* scheduling information in IOMMU irte.
*/
- if (!ret && pi.is_guest_mode)
+ if (!ret && pi.is_guest_mode && pi.ir_data)
svm_ir_list_add(svm, &pi);
} else {
/* Use legacy mode in IRTE */
@@ -898,7 +905,7 @@ int svm_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
* was cached. If so, we need to clean up the per-vcpu
* ir_list.
*/
- if (!ret && pi.prev_ga_tag) {
+ if (!ret && pi.prev_ga_tag && !WARN_ON(!pi.ir_data)) {
int id = AVIC_GATAG_TO_VCPUID(pi.prev_ga_tag);
struct kvm_vcpu *vcpu;