KVM: x86: Retry APIC optimized map recalc if vCPU is added/enabled

Retry the optimized APIC map recalculation if an APIC-enabled vCPU shows
up between allocating the map and filling in the map data.  Conditionally
reschedule before retrying even though the number of vCPUs that can be
created is bounded by KVM.  Retrying a few thousand times isn't so slow
as to be hugely problematic, but it's not blazing fast either.

Reset xapic_id_mistach on each retry as a vCPU could change its xAPIC ID
between loops, but do NOT reset max_id.  The map size also factors in
whether or not a vCPU's local APIC is hardware-enabled, i.e. userspace
and/or the guest can theoretically keep KVM retrying indefinitely.  The
only downside is that KVM will allocate more memory than is strictly
necessary if the vCPU with the highest x2APIC ID disabled its APIC while
the recalculation was in-progress.

Refresh kvm->arch.apic_map_dirty to opportunistically change it from
DIRTY => UPDATE_IN_PROGRESS to avoid an unnecessary recalc from a
different task, i.e. if another task is waiting to attempt an update
(which is likely since a retry happens if and only if an update is
required).

Link: https://lore.kernel.org/r/20230602233250.1014316-3-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>

1 files changed, 25 insertions, 4 deletions

diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index 113ca9661ab2..673880bc0762 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -376,7 +376,8 @@ void kvm_recalculate_apic_map(struct kvm *kvm)
 	struct kvm_vcpu *vcpu;
 	unsigned long i;
 	u32 max_id = 255; /* enough space for any xAPIC ID */
-	bool xapic_id_mismatch = false;
+	bool xapic_id_mismatch;
+	int r;
 
 	/* Read kvm->arch.apic_map_dirty before kvm->arch.apic_map.  */
 	if (atomic_read_acquire(&kvm->arch.apic_map_dirty) == CLEAN)
@@ -386,9 +387,14 @@ void kvm_recalculate_apic_map(struct kvm *kvm)
 		  "Dirty APIC map without an in-kernel local APIC");
 
 	mutex_lock(&kvm->arch.apic_map_lock);
+
+retry:
 	/*
-	 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map
-	 * (if clean) or the APIC registers (if dirty).
+	 * Read kvm->arch.apic_map_dirty before kvm->arch.apic_map (if clean)
+	 * or the APIC registers (if dirty).  Note, on retry the map may have
+	 * not yet been marked dirty by whatever task changed a vCPU's x2APIC
+	 * ID, i.e. the map may still show up as in-progress.  In that case
+	 * this task still needs to retry and complete its calculation.
 	 */
 	if (atomic_cmpxchg_acquire(&kvm->arch.apic_map_dirty,
 				   DIRTY, UPDATE_IN_PROGRESS) == CLEAN) {
@@ -397,6 +403,15 @@ void kvm_recalculate_apic_map(struct kvm *kvm)
 		return;
 	}
 
+	/*
+	 * Reset the mismatch flag between attempts so that KVM does the right
+	 * thing if a vCPU changes its xAPIC ID, but do NOT reset max_id, i.e.
+	 * keep max_id strictly increasing.  Disallowing max_id from shrinking
+	 * ensures KVM won't get stuck in an infinite loop, e.g. if the vCPU
+	 * with the highest x2APIC ID is toggling its APIC on and off.
+	 */
+	xapic_id_mismatch = false;
+
 	kvm_for_each_vcpu(i, vcpu, kvm)
 		if (kvm_apic_present(vcpu))
 			max_id = max(max_id, kvm_x2apic_id(vcpu->arch.apic));
@@ -415,9 +430,15 @@ void kvm_recalculate_apic_map(struct kvm *kvm)
 		if (!kvm_apic_present(vcpu))
 			continue;
 
-		if (kvm_recalculate_phys_map(new, vcpu, &xapic_id_mismatch)) {
+		r = kvm_recalculate_phys_map(new, vcpu, &xapic_id_mismatch);
+		if (r) {
 			kvfree(new);
 			new = NULL;
+			if (r == -E2BIG) {
+				cond_resched();
+				goto retry;
+			}
+
 			goto out;
 		}