The busy loop in hva_to_pfn_retry() is worse than a normal page fault
retry loop because it spins even while it's waiting for the invalidation
to complete. It isn't just that a page might get faulted out again before
it's actually accessed.
Introduce a wait queue to be woken when kvm->mn_active_invalidate_count
reaches zero, and wait on it if there is any pending invalidation which
affects the GPC being refreshed.
Signed-off-by: David Woodhouse <dwmw@xxxxxxxxxxxx>
[sean: massage comment as part of rebasing]
Signed-off-by: Sean Christopherson <seanjc@xxxxxxxxxx>
---
include/linux/kvm_host.h | 1 +
virt/kvm/kvm_main.c | 9 ++++++---
virt/kvm/pfncache.c | 30 ++++++++++++++++++++++++++----
3 files changed, 33 insertions(+), 7 deletions(-)
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index 2c0ed735f0f4..a9d7b2200b6f 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -787,6 +787,7 @@ struct kvm {
struct list_head gpc_list;
u64 mmu_gpc_invalidate_range_start;
u64 mmu_gpc_invalidate_range_end;
+ wait_queue_head_t gpc_invalidate_wq;
/*
* created_vcpus is protected by kvm->lock, and is incremented
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index b9223ecab2ca..3ba6d109a941 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -849,11 +849,13 @@ static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
spin_unlock(&kvm->mn_invalidate_lock);
/*
- * There can only be one waiter, since the wait happens under
- * slots_lock.
+ * There can only be one memslots waiter, since the wait happens under
+ * slots_lock, but there can be multiple gpc waiters.
*/
- if (wake)
+ if (wake) {
+ wake_up(&kvm->gpc_invalidate_wq);
rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
+ }
}
static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
@@ -1163,6 +1165,7 @@ static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
INIT_LIST_HEAD(&kvm->gpc_list);
spin_lock_init(&kvm->gpc_lock);
+ init_waitqueue_head(&kvm->gpc_invalidate_wq);
kvm->mmu_gpc_invalidate_range_start = KVM_HVA_ERR_BAD;
kvm->mmu_gpc_invalidate_range_end = KVM_HVA_ERR_BAD;
diff --git a/virt/kvm/pfncache.c b/virt/kvm/pfncache.c
index 2163bb6b899c..77cc5633636a 100644
--- a/virt/kvm/pfncache.c
+++ b/virt/kvm/pfncache.c
@@ -180,6 +180,30 @@ static bool gpc_invalidate_retry_hva(struct gfn_to_pfn_cache *gpc,
return gpc->kvm->mmu_invalidate_seq != mmu_seq;
}
+static void gpc_wait_for_invalidations(struct gfn_to_pfn_cache *gpc)
+{
+ struct kvm *kvm = gpc->kvm;
+
+ spin_lock(&kvm->mn_invalidate_lock);
+ if (gpc_invalidate_in_progress_hva(gpc)) {
+ DEFINE_WAIT(wait);
+
+ for (;;) {
+ prepare_to_wait(&kvm->gpc_invalidate_wq, &wait,
+ TASK_UNINTERRUPTIBLE);
+
+ if (!gpc_invalidate_in_progress_hva(gpc))
+ break;
+
+ spin_unlock(&kvm->mn_invalidate_lock);
+ schedule();
+ spin_lock(&kvm->mn_invalidate_lock);
+ }
+ finish_wait(&kvm->gpc_invalidate_wq, &wait);
+ }
+ spin_unlock(&kvm->mn_invalidate_lock);
+}
+
static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc)
{
/* Note, the new page offset may be different than the old! */
@@ -230,10 +254,8 @@ static kvm_pfn_t hva_to_pfn_retry(struct gfn_to_pfn_cache *gpc)
* trying to race ahead in the hope that a different task makes
* the cache valid.
*/
- while (gpc_invalidate_in_progress_hva(gpc)) {
- if (!cond_resched())
- cpu_relax();
- }
+ while (gpc_invalidate_in_progress_hva(gpc))
+ gpc_wait_for_invalidations(gpc);
mmu_seq = gpc->kvm->mmu_invalidate_seq;
smp_rmb();
--
2.47.0.rc1.288.g06298d1525-goog
--oDMvAmx7QP3BxXt9--
