]> git.karo-electronics.de Git - linux-beck.git/commitdiff
KVM: MMU: use page track for non-leaf shadow pages
authorXiao Guangrong <guangrong.xiao@linux.intel.com>
Wed, 24 Feb 2016 09:51:14 +0000 (17:51 +0800)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 3 Mar 2016 13:36:23 +0000 (14:36 +0100)
non-leaf shadow pages are always write protected, it can be the user
of page track

Signed-off-by: Xiao Guangrong <guangrong.xiao@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/mmu.c

index 58c067da6efc8359e23cb0178e718a9ab24bd08c..fe03d2a1d4d3370293b3292d02a13fbdea06fe08 100644 (file)
@@ -806,11 +806,17 @@ static void account_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp)
        struct kvm_memory_slot *slot;
        gfn_t gfn;
 
+       kvm->arch.indirect_shadow_pages++;
        gfn = sp->gfn;
        slots = kvm_memslots_for_spte_role(kvm, sp->role);
        slot = __gfn_to_memslot(slots, gfn);
+
+       /* the non-leaf shadow pages are keeping readonly. */
+       if (sp->role.level > PT_PAGE_TABLE_LEVEL)
+               return kvm_slot_page_track_add_page(kvm, slot, gfn,
+                                                   KVM_PAGE_TRACK_WRITE);
+
        kvm_mmu_gfn_disallow_lpage(slot, gfn);
-       kvm->arch.indirect_shadow_pages++;
 }
 
 static void unaccount_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp)
@@ -819,11 +825,15 @@ static void unaccount_shadowed(struct kvm *kvm, struct kvm_mmu_page *sp)
        struct kvm_memory_slot *slot;
        gfn_t gfn;
 
+       kvm->arch.indirect_shadow_pages--;
        gfn = sp->gfn;
        slots = kvm_memslots_for_spte_role(kvm, sp->role);
        slot = __gfn_to_memslot(slots, gfn);
+       if (sp->role.level > PT_PAGE_TABLE_LEVEL)
+               return kvm_slot_page_track_remove_page(kvm, slot, gfn,
+                                                      KVM_PAGE_TRACK_WRITE);
+
        kvm_mmu_gfn_allow_lpage(slot, gfn);
-       kvm->arch.indirect_shadow_pages--;
 }
 
 static bool __mmu_gfn_lpage_is_disallowed(gfn_t gfn, int level,
@@ -2132,12 +2142,18 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
        hlist_add_head(&sp->hash_link,
                &vcpu->kvm->arch.mmu_page_hash[kvm_page_table_hashfn(gfn)]);
        if (!direct) {
-               if (rmap_write_protect(vcpu, gfn))
+               /*
+                * we should do write protection before syncing pages
+                * otherwise the content of the synced shadow page may
+                * be inconsistent with guest page table.
+                */
+               account_shadowed(vcpu->kvm, sp);
+               if (level == PT_PAGE_TABLE_LEVEL &&
+                     rmap_write_protect(vcpu, gfn))
                        kvm_flush_remote_tlbs(vcpu->kvm);
+
                if (level > PT_PAGE_TABLE_LEVEL && need_sync)
                        kvm_sync_pages(vcpu, gfn);
-
-               account_shadowed(vcpu->kvm, sp);
        }
        sp->mmu_valid_gen = vcpu->kvm->arch.mmu_valid_gen;
        clear_page(sp->spt);