3 files changed, 64 insertions, 6 deletions

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 388a2ef..65f2400b 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -4309,15 +4309,32 @@ void kvm_mmu_slot_remove_write_access(struct kvm *kvm, int slot)
 			if (*rmapp)
 				__rmap_write_protect(kvm, rmapp, false);
 
-			if (need_resched() || spin_needbreak(&kvm->mmu_lock)) {
-				kvm_flush_remote_tlbs(kvm);
+			if (need_resched() || spin_needbreak(&kvm->mmu_lock))
 				cond_resched_lock(&kvm->mmu_lock);
-			}
 		}
 	}
 
-	kvm_flush_remote_tlbs(kvm);
 	spin_unlock(&kvm->mmu_lock);
+
+	/*
+	 * kvm_mmu_slot_remove_write_access() and kvm_vm_ioctl_get_dirty_log()
+	 * which do tlb flush out of mmu-lock should be serialized by
+	 * kvm->slots_lock otherwise tlb flush would be missed.
+	 */
+	lockdep_assert_held(&kvm->slots_lock);
+
+	/*
+	 * We can flush all the TLBs out of the mmu lock without TLB
+	 * corruption since we just change the spte from writable to
+	 * readonly so that we only need to care the case of changing
+	 * spte from present to present (changing the spte from present
+	 * to nonpresent will flush all the TLBs immediately), in other
+	 * words, the only case we care is mmu_spte_update() where we
+	 * haved checked SPTE_HOST_WRITEABLE | SPTE_MMU_WRITEABLE
+	 * instead of PT_WRITABLE_MASK, that means it does not depend
+	 * on PT_WRITABLE_MASK anymore.
+	 */
+	kvm_flush_remote_tlbs(kvm);
 }
 
 #define BATCH_ZAP_PAGES	10
diff --git a/arch/x86/kvm/mmu.h b/arch/x86/kvm/mmu.h
index 3842e70..b982112 100644
--- a/arch/x86/kvm/mmu.h
+++ b/arch/x86/kvm/mmu.h
@@ -104,6 +104,39 @@ static inline int is_present_gpte(unsigned long pte)
 	return pte & PT_PRESENT_MASK;
 }
 
+/*
+ * Currently, we have two sorts of write-protection, a) the first one
+ * write-protects guest page to sync the guest modification, b) another one is
+ * used to sync dirty bitmap when we do KVM_GET_DIRTY_LOG. The differences
+ * between these two sorts are:
+ * 1) the first case clears SPTE_MMU_WRITEABLE bit.
+ * 2) the first case requires flushing tlb immediately avoiding corrupting
+ *    shadow page table between all vcpus so it should be in the protection of
+ *    mmu-lock. And the another case does not need to flush tlb until returning
+ *    the dirty bitmap to userspace since it only write-protects the page
+ *    logged in the bitmap, that means the page in the dirty bitmap is not
+ *    missed, so it can flush tlb out of mmu-lock.
+ *
+ * So, there is the problem: the first case can meet the corrupted tlb caused
+ * by another case which write-protects pages but without flush tlb
+ * immediately. In order to making the first case be aware this problem we let
+ * it flush tlb if we try to write-protect a spte whose SPTE_MMU_WRITEABLE bit
+ * is set, it works since another case never touches SPTE_MMU_WRITEABLE bit.
+ *
+ * Anyway, whenever a spte is updated (only permission and status bits are
+ * changed) we need to check whether the spte with SPTE_MMU_WRITEABLE becomes
+ * readonly, if that happens, we need to flush tlb. Fortunately,
+ * mmu_spte_update() has already handled it perfectly.
+ *
+ * The rules to use SPTE_MMU_WRITEABLE and PT_WRITABLE_MASK:
+ * - if we want to see if it has writable tlb entry or if the spte can be
+ *   writable on the mmu mapping, check SPTE_MMU_WRITEABLE, this is the most
+ *   case, otherwise
+ * - if we fix page fault on the spte or do write-protection by dirty logging,
+ *   check PT_WRITABLE_MASK.
+ *
+ * TODO: introduce APIs to split these two cases.
+ */
 static inline int is_writable_pte(unsigned long pte)
 {
 	return pte & PT_WRITABLE_MASK;
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 63a828d..c5582c3 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -3632,11 +3632,19 @@ int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 		offset = i * BITS_PER_LONG;
 		kvm_mmu_write_protect_pt_masked(kvm, memslot, offset, mask);
 	}
-	if (is_dirty)
-		kvm_flush_remote_tlbs(kvm);
 
 	spin_unlock(&kvm->mmu_lock);
 
+	/* See the comments in kvm_mmu_slot_remove_write_access(). */
+	lockdep_assert_held(&kvm->slots_lock);
+
+	/*
+	 * All the TLBs can be flushed out of mmu lock, see the comments in
+	 * kvm_mmu_slot_remove_write_access().
+	 */
+	if (is_dirty)
+		kvm_flush_remote_tlbs(kvm);
+
 	r = -EFAULT;
 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
 		goto out;