KVM: MMU: move zap/flush to kvm_mmu_get_page
authorPaolo Bonzini <pbonzini@redhat.com>
Wed, 24 Feb 2016 10:26:10 +0000 (11:26 +0100)
committerPaolo Bonzini <pbonzini@redhat.com>
Tue, 8 Mar 2016 11:33:30 +0000 (12:33 +0100)
kvm_mmu_get_page is the only caller of kvm_sync_page_transient
and kvm_sync_pages.  Moving the handling of the invalid_list there
removes the need for the underdocumented kvm_sync_page_transient
function.

Reviewed-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/mmu.c

index 1c87102efb3d37b95bb935286bedb0f69431fe09..fecc9c51d9240e1f62caf711ed3239a4ff416bab 100644 (file)
@@ -1947,18 +1947,6 @@ static void kvm_mmu_flush_or_zap(struct kvm_vcpu *vcpu,
                kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
 }
 
-static bool kvm_sync_page_transient(struct kvm_vcpu *vcpu,
-                                   struct kvm_mmu_page *sp)
-{
-       LIST_HEAD(invalid_list);
-       int ret;
-
-       ret = __kvm_sync_page(vcpu, sp, &invalid_list);
-       kvm_mmu_flush_or_zap(vcpu, &invalid_list, false, ret);
-
-       return ret;
-}
-
 #ifdef CONFIG_KVM_MMU_AUDIT
 #include "mmu_audit.c"
 #else
@@ -1974,21 +1962,21 @@ static bool kvm_sync_page(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
 }
 
 /* @gfn should be write-protected at the call site */
-static void kvm_sync_pages(struct kvm_vcpu *vcpu,  gfn_t gfn)
+static bool kvm_sync_pages(struct kvm_vcpu *vcpu, gfn_t gfn,
+                          struct list_head *invalid_list)
 {
        struct kvm_mmu_page *s;
-       LIST_HEAD(invalid_list);
-       bool flush = false;
+       bool ret = false;
 
        for_each_gfn_indirect_valid_sp(vcpu->kvm, s, gfn) {
                if (!s->unsync)
                        continue;
 
                WARN_ON(s->role.level != PT_PAGE_TABLE_LEVEL);
-               flush |= kvm_sync_page(vcpu, s, &invalid_list);
+               ret |= kvm_sync_page(vcpu, s, invalid_list);
        }
 
-       kvm_mmu_flush_or_zap(vcpu, &invalid_list, false, flush);
+       return ret;
 }
 
 struct mmu_page_path {
@@ -2119,6 +2107,8 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
        unsigned quadrant;
        struct kvm_mmu_page *sp;
        bool need_sync = false;
+       bool flush = false;
+       LIST_HEAD(invalid_list);
 
        role = vcpu->arch.mmu.base_role;
        role.level = level;
@@ -2142,8 +2132,16 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
                if (sp->role.word != role.word)
                        continue;
 
-               if (sp->unsync && !kvm_sync_page_transient(vcpu, sp))
-                       break;
+               if (sp->unsync) {
+                       /* The page is good, but __kvm_sync_page might still end
+                        * up zapping it.  If so, break in order to rebuild it.
+                        */
+                       if (!__kvm_sync_page(vcpu, sp, &invalid_list))
+                               break;
+
+                       WARN_ON(!list_empty(&invalid_list));
+                       kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
+               }
 
                if (sp->unsync_children)
                        kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
@@ -2173,11 +2171,13 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
                        kvm_flush_remote_tlbs(vcpu->kvm);
 
                if (level > PT_PAGE_TABLE_LEVEL && need_sync)
-                       kvm_sync_pages(vcpu, gfn);
+                       flush |= kvm_sync_pages(vcpu, gfn, &invalid_list);
        }
        sp->mmu_valid_gen = vcpu->kvm->arch.mmu_valid_gen;
        clear_page(sp->spt);
        trace_kvm_mmu_get_page(sp, true);
+
+       kvm_mmu_flush_or_zap(vcpu, &invalid_list, false, flush);
        return sp;
 }