KVM: MMU: do not iterate over all VMs in mmu_shrink()
authorGleb Natapov <gleb@redhat.com>
Mon, 4 Jun 2012 11:53:23 +0000 (14:53 +0300)
committerAvi Kivity <avi@redhat.com>
Tue, 5 Jun 2012 14:46:43 +0000 (17:46 +0300)
mmu_shrink() needlessly iterates over all VMs even though it will not
attempt to free mmu pages from more than one on them. Fix that and also
check used mmu pages count outside of VM lock to skip inactive VMs faster.

Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
arch/x86/kvm/mmu.c

index d07e436..1ca7164 100644 (file)
@@ -3944,7 +3944,6 @@ static void kvm_mmu_remove_some_alloc_mmu_pages(struct kvm *kvm,
 static int mmu_shrink(struct shrinker *shrink, struct shrink_control *sc)
 {
        struct kvm *kvm;
-       struct kvm *kvm_freed = NULL;
        int nr_to_scan = sc->nr_to_scan;
 
        if (nr_to_scan == 0)
@@ -3956,22 +3955,30 @@ static int mmu_shrink(struct shrinker *shrink, struct shrink_control *sc)
                int idx;
                LIST_HEAD(invalid_list);
 
+               /*
+                * n_used_mmu_pages is accessed without holding kvm->mmu_lock
+                * here. We may skip a VM instance errorneosly, but we do not
+                * want to shrink a VM that only started to populate its MMU
+                * anyway.
+                */
+               if (kvm->arch.n_used_mmu_pages > 0) {
+                       if (!nr_to_scan--)
+                               break;
+                       continue;
+               }
+
                idx = srcu_read_lock(&kvm->srcu);
                spin_lock(&kvm->mmu_lock);
-               if (!kvm_freed && nr_to_scan > 0 &&
-                   kvm->arch.n_used_mmu_pages > 0) {
-                       kvm_mmu_remove_some_alloc_mmu_pages(kvm,
-                                                           &invalid_list);
-                       kvm_freed = kvm;
-               }
-               nr_to_scan--;
 
+               kvm_mmu_remove_some_alloc_mmu_pages(kvm, &invalid_list);
                kvm_mmu_commit_zap_page(kvm, &invalid_list);
+
                spin_unlock(&kvm->mmu_lock);
                srcu_read_unlock(&kvm->srcu, idx);
+
+               list_move_tail(&kvm->vm_list, &vm_list);
+               break;
        }
-       if (kvm_freed)
-               list_move_tail(&kvm_freed->vm_list, &vm_list);
 
        raw_spin_unlock(&kvm_lock);