KVM: x86: Try to enable irr_pending state with disabled APICv

Commit Message

Yong He Oct. 23, 2024, 12:45 p.m. UTC

From: Yong He <[e-mail]>

Try to enable irr_pending when set APIC state, if there is
pending interrupt in IRR with disabled APICv.

In save/restore VM scenery with disabled APICv. Qemu/CloudHypervisor
always send signals to stop running vcpu threads, then save
entire VM state, including APIC state. There may be a pending
timer interrupt in the saved APIC IRR that is injected before
vcpu_run return. But when restoring the VM, since APICv is
disabled, irr_pending is disabled by default, so this may cause
the timer interrupt in the IRR to be suspended for a long time,
until the next interrupt comes.

Signed-off-by: Yong He <[e-mail]>
---
 arch/x86/kvm/lapic.c | 4 ++++
 1 file changed, 4 insertions(+)

Patch

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 2098dc689088..7373f649958b 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -3099,6 +3099,10 @@  int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
 						apic_find_highest_irr(apic));
 		kvm_x86_call(hwapic_isr_update)(apic_find_highest_isr(apic));
 	}
+
+	/* Search the IRR and enable irr_pending state with disabled APICv*/
+	if (!enable_apicv && apic_search_irr(apic) != -1)
+		apic->irr_pending = true;
 	kvm_make_request(KVM_REQ_EVENT, vcpu);
 	if (ioapic_in_kernel(vcpu->kvm))
 		kvm_rtc_eoi_tracking_restore_one(vcpu);