Merge tag 'kvm-x86-apic-6.7' of https://github.com/kvm-x86/linux into HEAD

KVM x86 APIC changes for 6.7:

 - Purge VMX's posted interrupt descriptor *before* loading APIC state when
   handling KVM_SET_LAPIC.  Purging the PID after loading APIC state results in
   lost APIC timer IRQs as the APIC timer can be armed as part of loading APIC
   state, i.e. can immediately pend an IRQ if the expiry is in the past.

 - Clear the ICR.BUSY bit when handling trap-like x2APIC writes.  This avoids a
   WARN, due to KVM expecting the BUSY bit to be cleared when sending IPIs.

1 files changed, 17 insertions, 13 deletions

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 3e977dbbf993..245b20973cae 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -2444,22 +2444,22 @@ EXPORT_SYMBOL_GPL(kvm_lapic_set_eoi);
 void kvm_apic_write_nodecode(struct kvm_vcpu *vcpu, u32 offset)
 {
 	struct kvm_lapic *apic = vcpu->arch.apic;
-	u64 val;
 
 	/*
-	 * ICR is a single 64-bit register when x2APIC is enabled.  For legacy
-	 * xAPIC, ICR writes need to go down the common (slightly slower) path
-	 * to get the upper half from ICR2.
+	 * ICR is a single 64-bit register when x2APIC is enabled, all others
+	 * registers hold 32-bit values.  For legacy xAPIC, ICR writes need to
+	 * go down the common path to get the upper half from ICR2.
+	 *
+	 * Note, using the write helpers may incur an unnecessary write to the
+	 * virtual APIC state, but KVM needs to conditionally modify the value
+	 * in certain cases, e.g. to clear the ICR busy bit.  The cost of extra
+	 * conditional branches is likely a wash relative to the cost of the
+	 * maybe-unecessary write, and both are in the noise anyways.
 	 */
-	if (apic_x2apic_mode(apic) && offset == APIC_ICR) {
-		val = kvm_lapic_get_reg64(apic, APIC_ICR);
-		kvm_apic_send_ipi(apic, (u32)val, (u32)(val >> 32));
-		trace_kvm_apic_write(APIC_ICR, val);
-	} else {
-		/* TODO: optimize to just emulate side effect w/o one more write */
-		val = kvm_lapic_get_reg(apic, offset);
-		kvm_lapic_reg_write(apic, offset, (u32)val);
-	}
+	if (apic_x2apic_mode(apic) && offset == APIC_ICR)
+		kvm_x2apic_icr_write(apic, kvm_lapic_get_reg64(apic, APIC_ICR));
+	else
+		kvm_lapic_reg_write(apic, offset, kvm_lapic_get_reg(apic, offset));
 }
 EXPORT_SYMBOL_GPL(kvm_apic_write_nodecode);
 
@@ -2670,6 +2670,8 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu, bool init_event)
 	u64 msr_val;
 	int i;
 
+	static_call_cond(kvm_x86_apicv_pre_state_restore)(vcpu);
+
 	if (!init_event) {
 		msr_val = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
 		if (kvm_vcpu_is_reset_bsp(vcpu))
@@ -2981,6 +2983,8 @@ int kvm_apic_set_state(struct kvm_vcpu *vcpu, struct kvm_lapic_state *s)
 	struct kvm_lapic *apic = vcpu->arch.apic;
 	int r;
 
+	static_call_cond(kvm_x86_apicv_pre_state_restore)(vcpu);
+
 	kvm_lapic_set_base(vcpu, vcpu->arch.apic_base);
 	/* set SPIV separately to get count of SW disabled APICs right */
 	apic_set_spiv(apic, *((u32 *)(s->regs + APIC_SPIV)));