KVM MMU: fix for forgot mark parent->unsync_children bit

PATCH In review Xiao Guangrong 2 participants 7 messages May 5, 2010

Subsystems (3)

x86 KVM KVM x86

lore

Share & tools

lore

Xiao Guangrong

author

2010-05-05 12:19 UTC #0 lore raw

PATCH 1/2 KVM MMU: fix for forgot mark parent->unsync_children bit

When mapping a new parent to unsync shadow page, we should mark parent's unsync_children bit

Reported-by Marcelo Tosatti <[email protected]>

Signed-off-by Xiao Guangrong <[email protected]>

---
 arch/x86/kvm/mmu.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)

arch/x86/kvm/mmu.c +3 -1

index 97f2ea0..bf35a2f 100644
@@ -1374,7 +1374,9 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
 			if (sp->unsync_children) {
 				set_bit(KVM_REQ_MMU_SYNC, &vcpu->requests);
 				kvm_mmu_mark_parents_unsync(sp);
-			}
+			} else if (sp->unsync)
+				kvm_mmu_mark_parents_unsync(sp);
+
 			trace_kvm_mmu_get_page(sp, false);
 			return sp;
 		}

Signature · 1 line

1.6.1.2

Xiao Guangrong

author

2010-05-05 12:24 UTC #1 lore raw

PATCH 2/2 KVM MMU: fix race in invlpg code

It has race in invlpg code, like below sequences:

A: hold mmu_lock and get 'sp'
B: release mmu_lock and do other things
C: hold mmu_lock and continue use 'sp'

if other path freed 'sp' in stage B, then kernel will crash

This patch checks 'sp' whether lived before use 'sp' in stage C

Signed-off-by Xiao Guangrong <[email protected]>

---
 arch/x86/kvm/paging_tmpl.h |   22 ++++++++++++++++++++--
 1 files changed, 20 insertions(+), 2 deletions(-)

arch/x86/kvm/paging_tmpl.h +20 -2

index 624b38f..13ea675 100644
@@ -462,11 +462,16 @@ out_unlock:
 
 static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
 {
-	struct kvm_mmu_page *sp = NULL;
+	struct kvm_mmu_page *sp = NULL, *s;
 	struct kvm_shadow_walk_iterator iterator;
+	struct hlist_head *bucket;
+	struct hlist_node *node, *tmp;
 	gfn_t gfn = -1;
 	u64 *sptep = NULL, gentry;
 	int invlpg_counter, level, offset = 0, need_flush = 0;
+	unsigned index;
+	bool live = false;
+	union kvm_mmu_page_role role;
 
 	spin_lock(&vcpu->kvm->mmu_lock);

@@ -480,7 +485,7 @@ static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
 
 			if (!sp->unsync)
 				break;
-
+			role = sp->role;
 			WARN_ON(level != PT_PAGE_TABLE_LEVEL);
 			shift = PAGE_SHIFT -
 				  (PT_LEVEL_BITS - PT64_LEVEL_BITS) * level;

@@ -519,10 +524,23 @@ static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
 
 	mmu_guess_page_from_pte_write(vcpu, gfn_to_gpa(gfn) + offset, gentry);
 	spin_lock(&vcpu->kvm->mmu_lock);
+	index = kvm_page_table_hashfn(gfn);
+	bucket = &vcpu->kvm->arch.mmu_page_hash[index];
+	hlist_for_each_entry_safe(s, node, tmp, bucket, hash_link)
+		if (s == sp) {
+			if (s->gfn == gfn && s->role.word == role.word)
+				live = true;
+			break;
+		}
+
+	if (!live)
+		goto unlock_exit;
+
 	if (atomic_read(&vcpu->kvm->arch.invlpg_counter) == invlpg_counter) {
 		++vcpu->kvm->stat.mmu_pte_updated;
 		FNAME(update_pte)(vcpu, sp, sptep, &gentry);
 	}
+unlock_exit:
 	spin_unlock(&vcpu->kvm->mmu_lock);
 	mmu_release_page_from_pte_write(vcpu);
 }

Signature · 1 line

1.6.1.2

Avi Kivity

2010-05-05 12:31 UTC #2 lore raw

Re: [PATCH 2/2] KVM MMU: fix race in invlpg code

Quoted 53 lines · Xiao Guangrong

It has race in invlpg code, like below sequences:

A: hold mmu_lock and get 'sp'
B: release mmu_lock and do other things
C: hold mmu_lock and continue use 'sp'

if other path freed 'sp' in stage B, then kernel will crash

This patch checks 'sp' whether lived before use 'sp' in stage C

Signed-off-by Xiao Guangrong<[email protected]>

---
  arch/x86/kvm/paging_tmpl.h |   22 ++++++++++++++++++++--
  1 files changed, 20 insertions(+), 2 deletions(-)

arch/x86/kvm/paging_tmpl.h +10 -2

index 624b38f..13ea675 100644
@@ -462,11 +462,16 @@ out_unlock:
 
  static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
  {
-	struct kvm_mmu_page *sp = NULL;
+	struct kvm_mmu_page *sp = NULL, *s;
  	struct kvm_shadow_walk_iterator iterator;
+	struct hlist_head *bucket;
+	struct hlist_node *node, *tmp;
  	gfn_t gfn = -1;
  	u64 *sptep = NULL, gentry;
  	int invlpg_counter, level, offset = 0, need_flush = 0;
+	unsigned index;
+	bool live = false;
+	union kvm_mmu_page_role role;
 
  	spin_lock(&vcpu->kvm->mmu_lock);

@@ -480,7 +485,7 @@ static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
 
  			if (!sp->unsync)
  				break;
-
+			role = sp->role;
  			WARN_ON(level != PT_PAGE_TABLE_LEVEL);
  			shift = PAGE_SHIFT -
  				  (PT_LEVEL_BITS - PT64_LEVEL_BITS) * level;

@@ -519,10 +524,23 @@ static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
 
  	mmu_guess_page_from_pte_write(vcpu, gfn_to_gpa(gfn) + offset, gentry);
  	spin_lock(&vcpu->kvm->mmu_lock);
+	index = kvm_page_table_hashfn(gfn);
+	bucket =&vcpu->kvm->arch.mmu_page_hash[index];
+	hlist_for_each_entry_safe(s, node, tmp, bucket, hash_link)

arch/x86/kvm/paging_tmpl.h static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) · line 539

Quoted 10 lines

arch/x86/kvm/paging_tmpl.h +9 -0

+		if (s == sp) {
+			if (s->gfn == gfn&&  s->role.word == role.word)
+				live = true;
+			break;
+		}
+
+	if (!live)
+		goto unlock_exit;
+

Did you try the root_count method? I think it's cleaner.

Signature · 1 line

error compiling committee.c: too many arguments to function

Xiao Guangrong

author

2010-05-05 12:48 UTC #3 lore raw

Re: [PATCH 2/2] KVM MMU: fix race in invlpg code

Quoted 8 lines · Avi Kivity

Quoted 8 lines

      spin_lock(&vcpu->kvm->mmu_lock);

arch/x86/kvm/paging_tmpl.h +7 -0

+    index = kvm_page_table_hashfn(gfn);
+    bucket =&vcpu->kvm->arch.mmu_page_hash[index];
+    hlist_for_each_entry_safe(s, node, tmp, bucket, hash_link)
+        if (s == sp) {
+            if (s->gfn == gfn&&  s->role.word == role.word)
+                live = true;
+            break;

Quoted 8 lines

arch/x86/kvm/paging_tmpl.h static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) · line 538
Quoted 6 lines
arch/x86/kvm/paging_tmpl.h +5 -0
+        }
+
+    if (!live)
+        goto unlock_exit;
+
Did you try the root_count method? I think it's cleaner.

Avi, Thanks for your idea.

I have considered this method, but i'm not sure when it's the good time to real free this page, and i think we also need a way to synchronize the real free path and this path. Do you have any comment for it :-(

Xiao

Avi Kivity

2010-05-05 12:52 UTC #4 lore raw

Re: [PATCH 2/2] KVM MMU: fix race in invlpg code

Quoted 21 lines · Xiao Guangrong

Quoted 8 lines · Avi Kivity
Quoted 8 lines
       spin_lock(&vcpu->kvm->mmu_lock);
arch/x86/kvm/paging_tmpl.h +7 -0
+    index = kvm_page_table_hashfn(gfn);
+    bucket =&vcpu->kvm->arch.mmu_page_hash[index];
+    hlist_for_each_entry_safe(s, node, tmp, bucket, hash_link)
+        if (s == sp) {
+            if (s->gfn == gfn&&   s->role.word == role.word)
+                live = true;
+            break;
Quoted 9 lines
arch/x86/kvm/paging_tmpl.h static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva) · line 538
Quoted 7 lines
arch/x86/kvm/paging_tmpl.h +5 -0
+        }
+
+    if (!live)
+        goto unlock_exit;
+
Did you try the root_count method? I think it's cleaner.

Quoted 6 lines

Avi, Thanks for your idea.
I have considered this method, but i'm not sure when it's the good time to real free this page, and i think we also need a way to synchronize the real free path and this path. Do you have any comment for it :-(

Same as mmu_free_roots():

        --sp->root_count;
        if (!sp->root_count && sp->role.invalid) {
                kvm_mmu_zap_page(vcpu->kvm, sp);
                goto unlock_exit;
        }

Signature · 1 line

error compiling committee.c: too many arguments to function

Avi Kivity

2010-05-05 12:30 UTC #5 lore raw

In reply to Xiao Guangrong

Quoted 15 lines · Xiao Guangrong

When mapping a new parent to unsync shadow page, we should mark parent's unsync_children bit
Reported-by Marcelo Tosatti<[email protected]>
Signed-off-by Xiao Guangrong<[email protected]>
---
  arch/x86/kvm/mmu.c |    4 +++-
  1 files changed, 3 insertions(+), 1 deletions(-)
arch/x86/kvm/mmu.c +0 -0
index 97f2ea0..bf35a2f 100644

arch/x86/kvm/mmu.c static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu, · line 1382

Quoted 10 lines

arch/x86/kvm/mmu.c +3 -1

@@ -1374,7 +1374,9 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
  			if (sp->unsync_children) {
  				set_bit(KVM_REQ_MMU_SYNC,&vcpu->requests);
  				kvm_mmu_mark_parents_unsync(sp);
-			}
+			} else if (sp->unsync)
+				kvm_mmu_mark_parents_unsync(sp);
+
  			trace_kvm_mmu_get_page(sp, false);
  			return sp;
  		}

Which patch does this fix? If it wasn't merge yet, please repost with the fix included.

Signature · 1 line

error compiling committee.c: too many arguments to function

Xiao Guangrong

author

2010-05-05 12:38 UTC #6 lore raw

Quoted 23 lines · Avi Kivity

Quoted 22 lines · Xiao Guangrong

When mapping a new parent to unsync shadow page, we should mark parent's unsync_children bit

Reported-by Marcelo Tosatti<[email protected]>

Signed-off-by Xiao Guangrong<[email protected]>

---
  arch/x86/kvm/mmu.c |    4 +++-
  1 files changed, 3 insertions(+), 1 deletions(-)

arch/x86/kvm/mmu.c +3 -1

index 97f2ea0..bf35a2f 100644
@@ -1374,7 +1374,9 @@ static struct kvm_mmu_page
 *kvm_mmu_get_page(struct kvm_vcpu *vcpu,
              if (sp->unsync_children) {
                  set_bit(KVM_REQ_MMU_SYNC,&vcpu->requests);
                  kvm_mmu_mark_parents_unsync(sp);
-            }
+            } else if (sp->unsync)
+                kvm_mmu_mark_parents_unsync(sp);
+

Quoted 8 lines

Quoted 4 lines
arch/x86/kvm/mmu.c +0 -0
              trace_kvm_mmu_get_page(sp, false);
              return sp;
          }
Which patch does this fix? If it wasn't merge yet, please repost with the fix included.

Oh, OK, i'll sent the previous pathset that are reverted by Marcelo.

Thanks,
Xiao