Skip to contentSkip to conversation

kvm: don't lose the higher 32 bits of tlbs_dirty

PATCH In review Lai Jiangshan 4 participants 10 messages Dec 13, 2020 – Dec 21, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 kvm: don't lose the higher 32 bits of tlbs_dirty

From: Lai Jiangshan <[email protected]>

In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:

	need_tlb_flush |= kvm->tlbs_dirty;

with need_tlb_flush's type being int and tlbs_dirty's type being long.

It means that tlbs_dirty is always used as int and the higher 32 bits is useless. We can just change need_tlb_flush's type to long to make full use of tlbs_dirty.

---
 virt/kvm/kvm_main.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
virt/kvm/kvm_main.c +2 -1
index 2541a17ff1c4..4e519a517e9f 100644
@@ -470,7 +470,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 					const struct mmu_notifier_range *range)
 {
 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
-	int need_tlb_flush = 0, idx;
+	long need_tlb_flush = 0;
+	int idx;
 
 	idx = srcu_read_lock(&kvm->srcu);
 	spin_lock(&kvm->mmu_lock);
Signature · 1 line
2.19.1.6.gb485710b
#1 lore raw
Quoted 8 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:

	need_tlb_flush |= kvm->tlbs_dirty;

with need_tlb_flush's type being int and tlbs_dirty's type being long.

It means that tlbs_dirty is always used as int and the higher 32 bits is useless.

It's probably worth noting in the changelog that it's _extremely_ unlikely this bug can cause problems in practice. It would require encountering tlbs_dirty on a 4 billion count boundary, and KVM would need to be using shadow paging or be running a nested guest.

Quoted 2 lines

We can just change need_tlb_flush's type to long to make full use of tlbs_dirty.

Hrm, this does solve the problem, but I'm not a fan of continuing to use an integer variable as a boolean. Rather than propagate tlbs_dirty to need_tlb_flush, what if this bug fix patch checks tlbs_dirty directly, and then a follow up patch converts need_tlb_flush to a bool and removes the unnecessary initialization (see below).

E.g. the net result of both patches would be:

virt/kvm/kvm_main.c +5 -5
index 3abcb2ce5b7d..93b6986d3dfc 100644
@@ -473,7 +473,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
                                        const struct mmu_notifier_range *range)
 {
        struct kvm *kvm = mmu_notifier_to_kvm(mn);
-       int need_tlb_flush = 0, idx;
+       bool need_tlb_flush;
+       int idx;
 
        idx = srcu_read_lock(&kvm->srcu);
        spin_lock(&kvm->mmu_lock);
@@ -483,11 +484,10 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
         * count is also read inside the mmu_lock critical section.
         */
        kvm->mmu_notifier_count++;
-       need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
-                                            range->flags);
-       need_tlb_flush |= kvm->tlbs_dirty;
+       need_tlb_flush = !!kvm_unmap_hva_range(kvm, range->start, range->end,
+                                              range->flags);
        /* we've to flush the tlb before the pages can be freed */
-       if (need_tlb_flush)
+       if (need_tlb_flush || kvm->tlbs_dirty)
                kvm_flush_remote_tlbs(kvm);
 
        spin_unlock(&kvm->mmu_lock);
Fixes a4ee1ca4a36e ("KVM: MMU: delay flush all tlbs on sync_page path")
Quoted 10 lines
---
 virt/kvm/kvm_main.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
virt/kvm/kvm_main.c +0 -0
index 2541a17ff1c4..4e519a517e9f 100644

virt/kvm/kvm_main.c static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, · line 473

Quoted 5 lines
virt/kvm/kvm_main.c +1 -1
@@ -470,7 +470,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 					const struct mmu_notifier_range *range)
 {
 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
-	int need_tlb_flush = 0, idx;
+	long need_tlb_flush = 0;

need_tlb_flush doesn't need to be initialized here, it's explicitly set via the call to kvm_unmap_hva_range().

Quoted 7 lines
virt/kvm/kvm_main.c +1 -0
+	int idx;
 
 	idx = srcu_read_lock(&kvm->srcu);
 	spin_lock(&kvm->mmu_lock);
Signature · 1 line
2.19.1.6.gb485710b
#2 lore raw
Quoted 8 lines · Sean Christopherson
Quoted 6 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:

      need_tlb_flush |= kvm->tlbs_dirty;

with need_tlb_flush's type being int and tlbs_dirty's type being long.

Quoted 7 lines
Quoted 2 lines

It means that tlbs_dirty is always used as int and the higher 32 bits is useless.

It's probably worth noting in the changelog that it's _extremely_ unlikely this bug can cause problems in practice. It would require encountering tlbs_dirty on a 4 billion count boundary, and KVM would need to be using shadow paging or be running a nested guest.

You are right, I don't consider it would cause problems in practice, and I also tried to make tlbs_dirty as "int" and found that I have to change too many places.

And you are right it is worth noting about it, I'm sorry for neglecting.

Quoted 42 lines
Quoted 3 lines

We can just change need_tlb_flush's type to long to make full use of tlbs_dirty.

Hrm, this does solve the problem, but I'm not a fan of continuing to use an integer variable as a boolean. Rather than propagate tlbs_dirty to need_tlb_flush, what if this bug fix patch checks tlbs_dirty directly, and then a follow up patch converts need_tlb_flush to a bool and removes the unnecessary initialization (see below).

E.g. the net result of both patches would be:

virt/kvm/kvm_main.c +5 -5
index 3abcb2ce5b7d..93b6986d3dfc 100644
@@ -473,7 +473,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
                                        const struct mmu_notifier_range *range)
 {
        struct kvm *kvm = mmu_notifier_to_kvm(mn);
-       int need_tlb_flush = 0, idx;
+       bool need_tlb_flush;
+       int idx;
 
        idx = srcu_read_lock(&kvm->srcu);
        spin_lock(&kvm->mmu_lock);
@@ -483,11 +484,10 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
         * count is also read inside the mmu_lock critical section.
         */
        kvm->mmu_notifier_count++;
-       need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
-                                            range->flags);
-       need_tlb_flush |= kvm->tlbs_dirty;
+       need_tlb_flush = !!kvm_unmap_hva_range(kvm, range->start, range->end,
+                                              range->flags);
        /* we've to flush the tlb before the pages can be freed */
-       if (need_tlb_flush)
+       if (need_tlb_flush || kvm->tlbs_dirty)
                kvm_flush_remote_tlbs(kvm);
 
        spin_unlock(&kvm->mmu_lock);
Quoted 2 lines
Fixes a4ee1ca4a36e ("KVM: MMU: delay flush all tlbs on sync_page path")

I searched back, found it and considered adding this fixes tag, but I was afraid that this cleanup would be backported. Is it worth backporting such patch which is extremely hardly a problem in practice?

Quoted 27 lines
Quoted 11 lines
---
 virt/kvm/kvm_main.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
virt/kvm/kvm_main.c +0 -0
index 2541a17ff1c4..4e519a517e9f 100644

virt/kvm/kvm_main.c static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, · line 473

Quoted 5 lines
virt/kvm/kvm_main.c +1 -1
@@ -470,7 +470,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
                                      const struct mmu_notifier_range *range)
 {
      struct kvm *kvm = mmu_notifier_to_kvm(mn);
-     int need_tlb_flush = 0, idx;
+     long need_tlb_flush = 0;

need_tlb_flush doesn't need to be initialized here, it's explicitly set via the call to kvm_unmap_hva_range().

Quoted 7 lines
virt/kvm/kvm_main.c +1 -1
+     int idx;
 
      idx = srcu_read_lock(&kvm->srcu);
      spin_lock(&kvm->mmu_lock);
--

2.19.1.6.gb485710b

maintainer
#3 lore raw
In reply to Sean Christopherson
Quoted 15 lines · Sean Christopherson
Quoted 8 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:

	need_tlb_flush |= kvm->tlbs_dirty;

with need_tlb_flush's type being int and tlbs_dirty's type being long.

It means that tlbs_dirty is always used as int and the higher 32 bits is useless.

It's probably worth noting in the changelog that it's _extremely_ unlikely this bug can cause problems in practice. It would require encountering tlbs_dirty on a 4 billion count boundary, and KVM would need to be using shadow paging or be running a nested guest.

Quoted 8 lines
Quoted 2 lines

We can just change need_tlb_flush's type to long to make full use of tlbs_dirty.

Hrm, this does solve the problem, but I'm not a fan of continuing to use an integer variable as a boolean. Rather than propagate tlbs_dirty to need_tlb_flush, what if this bug fix patch checks tlbs_dirty directly, and then a follow up patch converts need_tlb_flush to a bool and removes the unnecessary initialization (see below).

Indeed, the compiler should be able to convert || to | if useful and valid (it may or may not do it depending on the sizes of types involved, but that's Someone Else's Problem and this is not really a path where every instruction matter).

Paolo

Quoted 62 lines

E.g. the net result of both patches would be:

virt/kvm/kvm_main.c +5 -5
index 3abcb2ce5b7d..93b6986d3dfc 100644
@@ -473,7 +473,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
                                         const struct mmu_notifier_range *range)
  {
         struct kvm *kvm = mmu_notifier_to_kvm(mn);
-       int need_tlb_flush = 0, idx;
+       bool need_tlb_flush;
+       int idx;
 
         idx = srcu_read_lock(&kvm->srcu);
         spin_lock(&kvm->mmu_lock);
@@ -483,11 +484,10 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
          * count is also read inside the mmu_lock critical section.
          */
         kvm->mmu_notifier_count++;
-       need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
-                                            range->flags);
-       need_tlb_flush |= kvm->tlbs_dirty;
+       need_tlb_flush = !!kvm_unmap_hva_range(kvm, range->start, range->end,
+                                              range->flags);
         /* we've to flush the tlb before the pages can be freed */
-       if (need_tlb_flush)
+       if (need_tlb_flush || kvm->tlbs_dirty)
                 kvm_flush_remote_tlbs(kvm);
 
         spin_unlock(&kvm->mmu_lock);
Fixes a4ee1ca4a36e ("KVM: MMU: delay flush all tlbs on sync_page path")
Quoted 10 lines
---
  virt/kvm/kvm_main.c | 3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)
virt/kvm/kvm_main.c +0 -0
index 2541a17ff1c4..4e519a517e9f 100644

virt/kvm/kvm_main.c static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, · line 473

Quoted 5 lines
virt/kvm/kvm_main.c +1 -1
@@ -470,7 +470,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
  					const struct mmu_notifier_range *range)
  {
  	struct kvm *kvm = mmu_notifier_to_kvm(mn);
-	int need_tlb_flush = 0, idx;
+	long need_tlb_flush = 0;

need_tlb_flush doesn't need to be initialized here, it's explicitly set via the call to kvm_unmap_hva_range().

Quoted 7 lines
virt/kvm/kvm_main.c +1 -0
+	int idx;
 
  	idx = srcu_read_lock(&kvm->srcu);
  	spin_lock(&kvm->mmu_lock);
Signature · 1 line
2.19.1.6.gb485710b
#4 lore raw

PATCH v2 1/1 kvm: check tlbs_dirty directly

From: Lai Jiangshan <[email protected]>

In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:

        need_tlb_flush |= kvm->tlbs_dirty;

with need_tlb_flush's type being int and tlbs_dirty's type being long.

It means that tlbs_dirty is always used as int and the higher 32 bits is useless. We need to check tlbs_dirty in a correct way and this change checks it directly without propagating it to need_tlb_flush.

And need_tlb_flush is changed to boolean because it is used as a boolean and its name starts with "need".

Note: it's _extremely_ unlikely this neglecting of higher 32 bits can cause problems in practice. It would require encountering tlbs_dirty on a 4 billion count boundary, and KVM would need to be using shadow paging or be running a nested guest.

Fixes a4ee1ca4a36e ("KVM: MMU: delay flush all tlbs on sync_page path")

---
Changed from V1:
Update the patch and the changelog as Sean Christopherson suggested.

 virt/kvm/kvm_main.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
virt/kvm/kvm_main.c +5 -5
index 2541a17ff1c4..1c17f3d073cb 100644
@@ -470,7 +470,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 					const struct mmu_notifier_range *range)
 {
 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
-	int need_tlb_flush = 0, idx;
+	int idx;
+	bool need_tlb_flush;
 
 	idx = srcu_read_lock(&kvm->srcu);
 	spin_lock(&kvm->mmu_lock);
@@ -480,11 +481,10 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 	 * count is also read inside the mmu_lock critical section.
 	 */
 	kvm->mmu_notifier_count++;
-	need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
-					     range->flags);
-	need_tlb_flush |= kvm->tlbs_dirty;
+	need_tlb_flush = !!kvm_unmap_hva_range(kvm, range->start, range->end,
+					       range->flags);
 	/* we've to flush the tlb before the pages can be freed */
-	if (need_tlb_flush)
+	if (need_tlb_flush || kvm->tlbs_dirty)
 		kvm_flush_remote_tlbs(kvm);
 
 	spin_unlock(&kvm->mmu_lock);
Signature · 1 line
2.19.1.6.gb485710b
#5 lore raw

Re: [PATCH V2] kvm: check tlbs_dirty directly

Note, you don't actually need to Cc [email protected] when sending the patch. If/when the patch is merged to Linus' tree, the stable tree maintainers, or more accurately their scripts, will automatically pick up the patch and apply it to the relevant stable trees.

You'll probably be getting the following letter from Greg KH any time now :-)

<formletter>

This is not the correct way to submit patches for inclusion in the stable kernel tree. Please read:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html
for how to do this properly.

</formletter>

Quoted 33 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:

        need_tlb_flush |= kvm->tlbs_dirty;

with need_tlb_flush's type being int and tlbs_dirty's type being long.

It means that tlbs_dirty is always used as int and the higher 32 bits is useless. We need to check tlbs_dirty in a correct way and this change checks it directly without propagating it to need_tlb_flush.

And need_tlb_flush is changed to boolean because it is used as a boolean and its name starts with "need".

Note: it's _extremely_ unlikely this neglecting of higher 32 bits can cause problems in practice. It would require encountering tlbs_dirty on a 4 billion count boundary, and KVM would need to be using shadow paging or be running a nested guest.

Fixes a4ee1ca4a36e ("KVM: MMU: delay flush all tlbs on sync_page path")

---
Changed from V1:
Update the patch and the changelog as Sean Christopherson suggested.

 virt/kvm/kvm_main.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
virt/kvm/kvm_main.c +0 -0
index 2541a17ff1c4..1c17f3d073cb 100644

virt/kvm/kvm_main.c static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, · line 474

Quoted 6 lines
virt/kvm/kvm_main.c +2 -1
@@ -470,7 +470,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 					const struct mmu_notifier_range *range)
 {
 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
-	int need_tlb_flush = 0, idx;
+	int idx;
+	bool need_tlb_flush;

It's a bit silly given how small this patch already is, but I do think this should be split into two patches so that the kvm->tlbs_dirty bug fix is fully isolated for backporting. I.e. patch 1/2 would simply be:

virt/kvm/kvm_main.c +1 -2
index 3abcb2ce5b7d..19dae28904f7 100644
@@ -485,9 +485,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
        kvm->mmu_notifier_count++;
        need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
                                             range->flags);
-       need_tlb_flush |= kvm->tlbs_dirty;
        /* we've to flush the tlb before the pages can be freed */
-       if (need_tlb_flush)
+       if (need_tlb_flush || kvm->tlbs_dirty)
                kvm_flush_remote_tlbs(kvm);
 
        spin_unlock(&kvm->mmu_lock);
Quoted 21 lines
virt/kvm/kvm_main.c +3 -4
 	idx = srcu_read_lock(&kvm->srcu);
 	spin_lock(&kvm->mmu_lock);
@@ -480,11 +481,10 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 	 * count is also read inside the mmu_lock critical section.
 	 */
 	kvm->mmu_notifier_count++;
-	need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
-					     range->flags);
-	need_tlb_flush |= kvm->tlbs_dirty;
+	need_tlb_flush = !!kvm_unmap_hva_range(kvm, range->start, range->end,
+					       range->flags);
 	/* we've to flush the tlb before the pages can be freed */
-	if (need_tlb_flush)
+	if (need_tlb_flush || kvm->tlbs_dirty)
 		kvm_flush_remote_tlbs(kvm);
 
 	spin_unlock(&kvm->mmu_lock);
Signature · 1 line
2.19.1.6.gb485710b
maintainer
#6 lore raw

Re: [PATCH V2] kvm: check tlbs_dirty directly

Quoted 6 lines · Sean Christopherson

Note, you don't actually need to Cc [email protected] when sending the patch. If/when the patch is merged to Linus' tree, the stable tree maintainers, or more accurately their scripts, will automatically pick up the patch and apply it to the relevant stable trees.

You'll probably be getting the following letter from Greg KH any time now :-)

Nope, this was fine, I don't mind seeing the patch before it hits Linus's tree (or any tree) at all. It lets me know what to look out for, nothing was done incorrectly here at all from what I can tell.

thanks,

greg k-h

maintainer
#7 lore raw

Re: [PATCH V2] kvm: check tlbs_dirty directly

In reply to Sean Christopherson
Quoted 7 lines · Sean Christopherson

Note, you don't actually need to [email protected] when sending the patch. If/when the patch is merged to Linus' tree, the stable tree maintainers, or more accurately their scripts, will automatically pick up the patch and apply it to the relevant stable trees.

You'll probably be getting the following letter from Greg KH any time now:-)

Quoted 8 lines

<formletter>

This is not the correct way to submit patches for inclusion in the stable kernel tree. Please read:
https://www.kernel.org/doc/html/latest/process/stable-kernel-rules.html
for how to do this properly.

</formletter>

As far as I know, as long as he sees LKML or something else in Cc he assumes it's an indication _for the maintainer_. Don't underestimate the power of Greg's mailing list filters. :)

Paolo

#8 lore raw

PATCH v3 1/1 kvm: check tlbs_dirty directly

In reply to Sean Christopherson

From: Lai Jiangshan <[email protected]>

In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:

        need_tlb_flush |= kvm->tlbs_dirty;

with need_tlb_flush's type being int and tlbs_dirty's type being long.

It means that tlbs_dirty is always used as int and the higher 32 bits is useless. We need to check tlbs_dirty in a correct way and this change checks it directly without propagating it to need_tlb_flush.

Note: it's _extremely_ unlikely this neglecting of higher 32 bits can cause problems in practice. It would require encountering tlbs_dirty on a 4 billion count boundary, and KVM would need to be using shadow paging or be running a nested guest.

Fixes a4ee1ca4a36e ("KVM: MMU: delay flush all tlbs on sync_page path")

---
Changed from V1:
Update the patch and the changelog as Sean Christopherson suggested.

Changed from v2:
don't change the type of need_tlb_flush

 virt/kvm/kvm_main.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
virt/kvm/kvm_main.c +1 -2
index 2541a17ff1c4..3083fb53861d 100644
@@ -482,9 +482,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 	kvm->mmu_notifier_count++;
 	need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
 					     range->flags);
-	need_tlb_flush |= kvm->tlbs_dirty;
 	/* we've to flush the tlb before the pages can be freed */
-	if (need_tlb_flush)
+	if (need_tlb_flush || kvm->tlbs_dirty)
 		kvm_flush_remote_tlbs(kvm);
 
 	spin_unlock(&kvm->mmu_lock);
Signature · 1 line
2.19.1.6.gb485710b
maintainer Applied
#9 lore raw

Re: [PATCH V3] kvm: check tlbs_dirty directly

Quoted 34 lines · Lai Jiangshan

From: Lai Jiangshan <[email protected]>

In kvm_mmu_notifier_invalidate_range_start(), tlbs_dirty is used as:

         need_tlb_flush |= kvm->tlbs_dirty;

with need_tlb_flush's type being int and tlbs_dirty's type being long.

It means that tlbs_dirty is always used as int and the higher 32 bits is useless. We need to check tlbs_dirty in a correct way and this change checks it directly without propagating it to need_tlb_flush.

Note: it's _extremely_ unlikely this neglecting of higher 32 bits can cause problems in practice. It would require encountering tlbs_dirty on a 4 billion count boundary, and KVM would need to be using shadow paging or be running a nested guest.

Fixes a4ee1ca4a36e ("KVM: MMU: delay flush all tlbs on sync_page path")

---
Changed from V1:
Update the patch and the changelog as Sean Christopherson suggested.

Changed from v2:
don't change the type of need_tlb_flush

  virt/kvm/kvm_main.c | 3 +--
  1 file changed, 1 insertion(+), 2 deletions(-)
virt/kvm/kvm_main.c +0 -0
index 2541a17ff1c4..3083fb53861d 100644

virt/kvm/kvm_main.c static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, · line 489

Quoted 10 lines
virt/kvm/kvm_main.c +1 -2
@@ -482,9 +482,8 @@ static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
  	kvm->mmu_notifier_count++;
  	need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end,
  					     range->flags);
-	need_tlb_flush |= kvm->tlbs_dirty;
  	/* we've to flush the tlb before the pages can be freed */
-	if (need_tlb_flush)
+	if (need_tlb_flush || kvm->tlbs_dirty)
  		kvm_flush_remote_tlbs(kvm);
 
  	spin_unlock(&kvm->mmu_lock);

Queued, thanks.

Paolo