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 to suppress a
    WARN due to KVM expecting the BUSY bit to be cleared when sending IPIs.
 -----BEGIN PGP SIGNATURE-----
 
 iQJGBAABCgAwFiEEMHr+pfEFOIzK+KY1YJEiAU0MEvkFAmU8DrkSHHNlYW5qY0Bn
 b29nbGUuY29tAAoJEGCRIgFNDBL5Z54P/jQbjmwbvAVDa1g2HGr16ouFb3v+rCgV
 eXiN2/8tDlQl9JkvZsjOY1AtJauS3RZu/SncT3IxCCvHVrKKt0xgnhWT//rAMX3N
 zrrFxXJmilFdd2ijsuK8flHVFO5i06Kx1NRjSmAMb76et1et1EI0I62F1rom8ocn
 l0tryTkV0qhWKlgxzLBNzzhzxxA2Mvb6vFhut+nfa0IisnQ+roh5dV4JS7Cdhy+5
 qIQ4J86JdDR6kmraZLCv8r7vIQn5EQPVCAC2WWalds7iWUW3ixy01NLweVLaVtA3
 xWRHYK7azvu3SJ9HvmSMRZzzTElQBYnE2d37ycyYewHbDUJ6FCXxh7H+BehYaVBW
 CK6GLaMHY8vyXLcB2cRAzZ7GgBSSDevL11/CMc2ueGc51AOsDCxFh6v69TU+/PiQ
 dMOYiNTuwV9QC3raIXPCc2Q9rqGp20xBqXIv5XLEyIxqcFQESvtRggg6zJdBZ3zB
 T5/xAXDf++rKFiu1wskFAbwsk3M4T36u1WlK9iayC9dPvCDGRzsle3tt9npmoj2X
 3e0Uz7fkRLki1YZW1qLmvqoYJ9bX9B9BjuKkORIqSqXuwcgPYEzHt1GcRByI2FSk
 KSTnNwgs/Bs498LsDdU1Z740ZkYFGmu2nK+CUvwxP+uejPsnlhik5kGqLSIeA7zA
 9kDpcdV+tZKJ
 =atZ2
 -----END PGP SIGNATURE-----

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.
This commit is contained in:
Paolo Bonzini 2023-10-31 10:11:19 -04:00
commit f233646760
4 changed files with 21 additions and 15 deletions

View File

@ -108,6 +108,7 @@ KVM_X86_OP_OPTIONAL(vcpu_blocking)
KVM_X86_OP_OPTIONAL(vcpu_unblocking)
KVM_X86_OP_OPTIONAL(pi_update_irte)
KVM_X86_OP_OPTIONAL(pi_start_assignment)
KVM_X86_OP_OPTIONAL(apicv_pre_state_restore)
KVM_X86_OP_OPTIONAL(apicv_post_state_restore)
KVM_X86_OP_OPTIONAL_RET0(dy_apicv_has_pending_interrupt)
KVM_X86_OP_OPTIONAL(set_hv_timer)

View File

@ -1708,6 +1708,7 @@ struct kvm_x86_ops {
int (*pi_update_irte)(struct kvm *kvm, unsigned int host_irq,
uint32_t guest_irq, bool set);
void (*pi_start_assignment)(struct kvm *kvm);
void (*apicv_pre_state_restore)(struct kvm_vcpu *vcpu);
void (*apicv_post_state_restore)(struct kvm_vcpu *vcpu);
bool (*dy_apicv_has_pending_interrupt)(struct kvm_vcpu *vcpu);

View File

@ -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)));

View File

@ -6912,7 +6912,7 @@ static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
}
static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
static void vmx_apicv_pre_state_restore(struct kvm_vcpu *vcpu)
{
struct vcpu_vmx *vmx = to_vmx(vcpu);
@ -8286,7 +8286,7 @@ static struct kvm_x86_ops vmx_x86_ops __initdata = {
.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
.load_eoi_exitmap = vmx_load_eoi_exitmap,
.apicv_post_state_restore = vmx_apicv_post_state_restore,
.apicv_pre_state_restore = vmx_apicv_pre_state_restore,
.required_apicv_inhibits = VMX_REQUIRED_APICV_INHIBITS,
.hwapic_irr_update = vmx_hwapic_irr_update,
.hwapic_isr_update = vmx_hwapic_isr_update,