Skip to contentSkip to conversation

KVM: VMX: Update SVI during runtime APICv activation

PATCH Applied Jon Kohler 3 participants 4 messages Jun 12 – Jun 14

Subsystems (2)
lore
Share & tools
lore

Jon Kohler

author
#0 lore raw

PATCH 1/1 KVM: VMX: Update SVI during runtime APICv activation

From: Dongli Zhang <[email protected]>

commit b2849bec936be642b5420801f902337f2507648e upstream.

The APICv (apic->apicv_active) can be activated or deactivated at runtime, for instance, because of APICv inhibit reasons. Intel VMX employs different mechanisms to virtualize LAPIC based on whether APICv is active.

When APICv is activated at runtime, GUEST_INTR_STATUS is used to configure and report the current pending IRR and ISR states. Unless a specific vector is explicitly included in EOI_EXIT_BITMAP, its EOI will not be trapped to KVM. Intel VMX automatically clears the corresponding ISR bit based on the GUEST_INTR_STATUS.SVI field.

When APICv is deactivated at runtime, the VM_ENTRY_INTR_INFO_FIELD is used to specify the next interrupt vector to invoke upon VM-entry. The VMX IDT_VECTORING_INFO_FIELD is used to report un-invoked vectors on VM-exit. EOIs are always trapped to KVM, so the software can manually clear pending ISR bits.

There are scenarios where, with APICv activated at runtime, a guest-issued EOI may not be able to clear the pending ISR bit.

Taking vector 236 as an example, here is one scenario.

1. Suppose APICv is inactive. Vector 236 is pending in the IRR.
2. To handle KVM_REQ_EVENT, KVM moves vector 236 from the IRR to the ISR, and configures the VM_ENTRY_INTR_INFO_FIELD via vmx_inject_irq().
3. After VM-entry, vector 236 is invoked through the guest IDT. At this point, the data in VM_ENTRY_INTR_INFO_FIELD is no longer valid. The guest interrupt handler for vector 236 is invoked.
4. Suppose a VM exit occurs very early in the guest interrupt handler, before the EOI is issued.
5. Nothing is reported through the IDT_VECTORING_INFO_FIELD because vector 236 has already been invoked in the guest.
6. Now, suppose APICv is activated. Before the next VM-entry, KVM calls kvm_vcpu_update_apicv() to activate APICv.
7. Unfortunately, GUEST_INTR_STATUS.SVI is not configured, although vector 236 is still pending in the ISR.
8. After VM-entry, the guest finally issues the EOI for vector 236. However, because SVI is not configured, vector 236 is not cleared.
9. ISR is stalled forever on vector 236.

Here is another scenario.

1. Suppose APICv is inactive. Vector 236 is pending in the IRR.
2. To handle KVM_REQ_EVENT, KVM moves vector 236 from the IRR to the ISR, and configures the VM_ENTRY_INTR_INFO_FIELD via vmx_inject_irq().
3. VM-exit occurs immediately after the next VM-entry. The vector 236 is not invoked through the guest IDT. Instead, it is saved to the IDT_VECTORING_INFO_FIELD during the VM-exit.
4. KVM calls kvm_queue_interrupt() to re-queue the un-invoked vector 236 into vcpu->arch.interrupt. A KVM_REQ_EVENT is requested.
5. Now, suppose APICv is activated. Before the next VM-entry, KVM calls kvm_vcpu_update_apicv() to activate APICv.
6. Although APICv is now active, KVM still uses the legacy VM_ENTRY_INTR_INFO_FIELD to re-inject vector 236. GUEST_INTR_STATUS.SVI is not configured.
7. After the next VM-entry, vector 236 is invoked through the guest IDT. Finally, an EOI occurs. However, due to the lack of GUEST_INTR_STATUS.SVI configuration, vector 236 is not cleared from the ISR.
8. ISR is stalled forever on vector 236.

Using QEMU as an example, vector 236 is stuck in ISR forever.

(qemu) info lapic 1
dumping local APIC state for CPU 1

LVT0	 0x00010700 active-hi edge  masked                      ExtINT (vec 0)
LVT1	 0x00010400 active-hi edge  masked                      NMI

LVTPC 0x00000400 active-hi edge NMI
LVTERR 0x000000fe active-hi edge Fixed (vec 254)
LVTTHMR 0x00010000 active-hi edge masked Fixed (vec 0)
LVTT 0x000400ec active-hi edge tsc-deadline Fixed (vec 236)
Timer DCR=0x0 (divide by 2) initial_count = 0 current_count = 0
SPIV 0x000001ff APIC enabled, focus=off, spurious vec 255
ICR 0x000000fd physical edge de-assert no-shorthand
ICR2 0x00000000 cpu 0 (X2APIC ID)
ESR 0x00000000
ISR 236
IRR 37(level) 236

The issue isn't applicable to AMD SVM as KVM simply writes vmcb01 directly irrespective of whether L1 (vmcs01) or L2 (vmcb02) is active (unlike VMX, there is no need/cost to switch between VMCBs). In addition, APICV_INHIBIT_REASON_IRQWIN ensures AMD SVM AVIC is not activated until the last interrupt is EOI'd.

Fix the bug by configuring Intel VMX GUEST_INTR_STATUS.SVI if APICv is activated at runtime.

[sean: call out that SVM writes vmcb01 directly, tweak comment]
(cherry picked from commit b2849bec936be642b5420801f902337f2507648e)
Cc [email protected] # 6.6.x and above
Cc Gulshan Gabel <[email protected]>

---

This issue is pervasive and has been observed in production with QEMU as the VMM. One scenario where this occurs is with Windows guests that use the AutoEOI feature, which inhibits APICv (APICV_INHIBIT_REASON_HYPERV).

The observed sequence is:

1. A VM is actively servicing vector 209 while live migrating, and before the guest issues EOI, the VM is paused and migrated. The LAPIC state (including ISR/IRR) is saved on the source. Until now, APICv has been inhibited by AutoEOI.
2. Upon arrival at the destination, the LAPIC state is restored via kvm_apic_set_state(). At this point, MSRs are not loaded, and since the inhibit is not yet in place, apicv_active is true, and vmx_hwapic_isr_update() writes SVI=209 into GUEST_INTR_STATUS.
3. When MSRs are subsequently loaded, the Hyper-V AutoEOI state is restored, causing KVM to set APICV_INHIBIT_REASON_HYPERV. On the first KVM_RUN, __kvm_vcpu_update_apicv() transitions apicv_active to false and disables VID, leaving SVI=209 stale in GUEST_INTR_STATUS.
4. When the VM is rebooted from inside the guest, kvm_lapic_reset() zeroes ISR/IRR in the virtual APIC page but does not update GUEST_INTR_STATUS because apicv_active is false — SVI=209 persists.
5. During the bootloader sequence, the guest clears the Hyper-V AutoEOI inhibit, and apicv_active transitions back to true. The stale SVI=209 is now live, causing hardware to block the delivery of all virtual interrupts with a lower priority.
6. In the observed case, the UEFI timer interrupt (vector 32) is blocked in IRR. The guest later reprograms this vector. When APICv is subsequently inhibited again, and the software interrupt path takes over, the stale IRR entry is injected to the wrong handler, and the guest panics.

With this fix, when APICv is reactivated in step 5, the SVI is recalculated from the current virtual ISR, which is the expected behavior.

Past that, we do also see this fail w/ selftest vmx_apicv_updates_test which fails with the following signature before this patch and afterwards it passes nicely.

./vmx_apicv_updates_test
Random seed: 0x6b8b4567
==== Test Assertion Failure ====
x86/vmx_apicv_updates_test.c:88: x2apic_read_reg(APIC_ISR + APIC_VECTOR_TO_REG_OFFSET(GOOD_IPI_VECTOR)) == 0 pid=154616 tid=154616 errno=4 - Interrupted system call

     1	0x0000000000411e41: assert_on_unhandled_exception at processor.c:778
     2	0x000000000040593d: _vcpu_run at kvm_util.c:1664
     3	0x000000000040595a: vcpu_run at kvm_util.c:1675
     4	0x0000000000400d1a: main at vmx_apicv_updates_test.c:131
     5	0x000000000042e9f8: __libc_start_main at ??:?
     6	0x0000000000400f3d: _start at ??:?

0x1 != 0x0 (x2apic_read_reg(APIC_ISR + APIC_VECTOR_TO_REG_OFFSET(GOOD_IPI_VECTOR)) != 0)

 arch/x86/kvm/vmx/vmx.c | 9 ---------
 arch/x86/kvm/x86.c     | 7 +++++++
 2 files changed, 7 insertions(+), 9 deletions(-)
arch/x86/kvm/vmx/vmx.c +0 -9
index c084f48e2b0b..b7798ced7b50 100644
@@ -6886,15 +6886,6 @@ void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
 	 * VM-Exit, otherwise L1 with run with a stale SVI.
 	 */
 	if (is_guest_mode(vcpu)) {
-		/*
-		 * KVM is supposed to forward intercepted L2 EOIs to L1 if VID
-		 * is enabled in vmcs12; as above, the EOIs affect L2's vAPIC.
-		 * Note, userspace can stuff state while L2 is active; assert
-		 * that VID is disabled if and only if the vCPU is in KVM_RUN
-		 * to avoid false positives if userspace is setting APIC state.
-		 */
-		WARN_ON_ONCE(vcpu->wants_to_run &&
-			     nested_cpu_has_vid(get_vmcs12(vcpu)));
 		to_vmx(vcpu)->nested.update_vmcs01_hwapic_isr = true;
 		return;
 	}
arch/x86/kvm/x86.c +7 -0
index ad2b7158b9c8..a21ebe04aa23 100644
@@ -10950,9 +10950,16 @@ void __kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
 	 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
 	 * still active when the interrupt got accepted. Make sure
 	 * kvm_check_and_inject_events() is called to check for that.
+	 *
+	 * Update SVI when APICv gets enabled, otherwise SVI won't reflect the
+	 * highest bit in vISR and the next accelerated EOI in the guest won't
+	 * be virtualized correctly (the CPU uses SVI to determine which vISR
+	 * vector to clear).
 	 */
 	if (!apic->apicv_active)
 		kvm_make_request(KVM_REQ_EVENT, vcpu);
+	else
+		kvm_apic_update_hwapic_isr(vcpu);
 
 out:
 	preempt_enable();
Signature · 1 line
2.43.0
reviewer Acked-by on v1 1/1
#1 lore raw
Quoted 3 lines · Jon Kohler

From: Dongli Zhang <[email protected]>

commit b2849bec936be642b5420801f902337f2507648e upstream.

...

Quoted 10 lines
[sean: call out that SVM writes vmcb01 directly, tweak comment]
(cherry picked from commit b2849bec936be642b5420801f902337f2507648e)
Cc [email protected] # 6.6.x and above
Cc Gulshan Gabel <[email protected]>
Quoted 6 lines

---

This issue is pervasive and has been observed in production with QEMU as the VMM. One scenario where this occurs is with Windows guests that use the AutoEOI feature, which inhibits APICv (APICV_INHIBIT_REASON_HYPERV).

Gah, sorry, my bad. I don't know why I didn't tag this for stable@.

author
#3 lore raw
Quoted 7 lines · Sasha Levin
Quoted 2 lines · Jon Kohler

From: Dongli Zhang <[email protected]>

Quoted 7 lines
Quoted 1 line

commit b2849bec936be642b5420801f902337f2507648e upstream.

Queued for 6.18.y, thanks. (And thanks Sean for the ack.)

--
Thanks,
Sasha

Bringing a side thread back on list, Sasha is going to queue this for 6.12.y as well because it applies cleanly there; however, it does not apply cleanly to 6.6.y.

Gulshan (CC’d) did all the great investigation work on this (and the commit msg narrative on my original email was all him!), so he's going to pick up the 6.6.y backport here as we’ve already backported/shipped that to our internal 6.6.y production branch, so that should be straight forward.

Jon