[PATCH 0/3] KVM SoftMMU fixes

[PATCH 0/3] KVM SoftMMU fixes

[Joerg Roedel]

Hi Avi, Marcelo,

this small patch series fixes two issues and include one cleanup I ran
into hacking in the KVM SoftMMU code. Please consider to apply.

Joerg

diffstat:

 arch/x86/kvm/mmu.c  |   10 +++-------
 virt/kvm/kvm_main.c |    6 ++++--
 2 files changed, 7 insertions(+), 9 deletions(-)

[PATCH 1/3] kvm mmu: handle compound pages in kvm_is_mmio_pfn

[Joerg Roedel]

The function kvm_is_mmio_pfn is called before put_page is called on a
page by KVM. This is a problem when when this function is called on some
struct page which is part of a compund page. It does not test the
reserved flag of the compound page but of the struct page within the
compount page. This is a problem when KVM works with hugepages allocated
at boot time. These pages have the reserved bit set in all tail pages.
Only the flag in the compount head is cleared. KVM would not put such a
page which results in a memory leak.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 virt/kvm/kvm_main.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 266bdaf..0ed662d 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -535,8 +535,10 @@ static inline int valid_vcpu(int n)
 
 inline int kvm_is_mmio_pfn(pfn_t pfn)
 {
-	if (pfn_valid(pfn))
-		return PageReserved(pfn_to_page(pfn));
+	if (pfn_valid(pfn)) {
+		struct page *page = compound_head(pfn_to_page(pfn));
+		return PageReserved(page);
+	}
 
 	return true;
 }
-- 
1.5.6.4

[PATCH 2/3] kvm mmu: remove redundant check in mmu_set_spte

[Joerg Roedel]

The following code flow is unnecessary:

	if (largepage)
		was_rmapped = is_large_pte(*shadow_pte);
	 else
	 	was_rmapped = 1;

The is_large_pte() function will always evaluate to one here because the
(largepage && !is_large_pte) case is already handled in the first
if-clause. So we can remove this check and set was_rmapped to one always
here.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kvm/mmu.c |    8 ++------
 1 files changed, 2 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index ef060ec..c90b4b2 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -1791,12 +1791,8 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
 			pgprintk("hfn old %lx new %lx\n",
 				 spte_to_pfn(*shadow_pte), pfn);
 			rmap_remove(vcpu->kvm, shadow_pte);
-		} else {
-			if (largepage)
-				was_rmapped = is_large_pte(*shadow_pte);
-			else
-				was_rmapped = 1;
-		}
+		} else
+			was_rmapped = 1;
 	}
 	if (set_spte(vcpu, shadow_pte, pte_access, user_fault, write_fault,
 		      dirty, largepage, global, gfn, pfn, speculative, true)) {
-- 
1.5.6.4

[PATCH 3/3] kvm mmu: alloc shadow pages with __GFP_ZERO

[Joerg Roedel]

Not using __GFP_ZERO when allocating shadow pages triggers the
assertion in the kvm_mmu_alloc_page() when MMU debugging is enabled.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kvm/mmu.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index c90b4b2..d93ecec 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -301,7 +301,7 @@ static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
 	if (cache->nobjs >= min)
 		return 0;
 	while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
-		page = alloc_page(GFP_KERNEL);
+		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 		if (!page)
 			return -ENOMEM;
 		set_page_private(page, 0);
-- 
1.5.6.4

Re: [PATCH 3/3] kvm mmu: alloc shadow pages with __GFP_ZERO

[Avi Kivity]

Joerg Roedel wrote:
> Not using __GFP_ZERO when allocating shadow pages triggers the
> assertion in the kvm_mmu_alloc_page() when MMU debugging is enabled.
>
> Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
> ---
>  arch/x86/kvm/mmu.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index c90b4b2..d93ecec 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -301,7 +301,7 @@ static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
>  	if (cache->nobjs >= min)
>  		return 0;
>  	while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
> -		page = alloc_page(GFP_KERNEL);
> +		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
>  		if (!page)
>  			return -ENOMEM;
>  		set_page_private(page, 0);
>   

What is the warning?

Adding __GFP_ZERO here will cause us to clear the page twice, which is 
wasteful.


-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Re: [PATCH 3/3] kvm mmu: alloc shadow pages with __GFP_ZERO

[Joerg Roedel]

On Wed, Feb 18, 2009 at 01:47:04PM +0000, Avi Kivity wrote:
> Joerg Roedel wrote:
> >Not using __GFP_ZERO when allocating shadow pages triggers the
> >assertion in the kvm_mmu_alloc_page() when MMU debugging is enabled.
> >
> >Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
> >---
> > arch/x86/kvm/mmu.c |    2 +-
> > 1 files changed, 1 insertions(+), 1 deletions(-)
> >
> >diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> >index c90b4b2..d93ecec 100644
> >--- a/arch/x86/kvm/mmu.c
> >+++ b/arch/x86/kvm/mmu.c
> >@@ -301,7 +301,7 @@ static int mmu_topup_memory_cache_page(struct kvm_mmu_memory_cache *cache,
> > 	if (cache->nobjs >= min)
> > 		return 0;
> > 	while (cache->nobjs < ARRAY_SIZE(cache->objects)) {
> >-		page = alloc_page(GFP_KERNEL);
> >+		page = alloc_page(GFP_KERNEL | __GFP_ZERO);
> > 		if (!page)
> > 			return -ENOMEM;
> > 		set_page_private(page, 0);
> >  
> 
> What is the warning?
> 
> Adding __GFP_ZERO here will cause us to clear the page twice, which is wasteful.

The assertion which the attached patch removes fails sometimes. Removing
this assertion is the alternative solution to this problem ;-)

>From ca45f3a2e45cd7e76ca624bb1098329db8ff83ab Mon Sep 17 00:00:00 2001
From: Joerg Roedel <joerg.roedel@amd.com>
Date: Wed, 18 Feb 2009 14:51:13 +0100
Subject: [PATCH] kvm mmu: remove assertion in kvm_mmu_alloc_page

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
---
 arch/x86/kvm/mmu.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index d93ecec..b226973 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -802,7 +802,6 @@ static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
 	set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
 	list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
 	INIT_LIST_HEAD(&sp->oos_link);
-	ASSERT(is_empty_shadow_page(sp->spt));
 	bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS);
 	sp->multimapped = 0;
 	sp->parent_pte = parent_pte;
-- 
1.5.6.4

-- 
           | Advanced Micro Devices GmbH
 Operating | Karl-Hammerschmidt-Str. 34, 85609 Dornach bei München
 System    | 
 Research  | Geschäftsführer: Jochen Polster, Thomas M. McCoy, Giuliano Meroni
 Center    | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
           | Registergericht München, HRB Nr. 43632

Re: [PATCH 3/3] kvm mmu: alloc shadow pages with __GFP_ZERO

[Avi Kivity]

Joerg Roedel wrote:
> The assertion which the attached patch removes fails sometimes. Removing
> this assertion is the alternative solution to this problem ;-)
>
> From ca45f3a2e45cd7e76ca624bb1098329db8ff83ab Mon Sep 17 00:00:00 2001
> From: Joerg Roedel <joerg.roedel@amd.com>
> Date: Wed, 18 Feb 2009 14:51:13 +0100
> Subject: [PATCH] kvm mmu: remove assertion in kvm_mmu_alloc_page
>
> Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
> ---
>  arch/x86/kvm/mmu.c |    1 -
>  1 files changed, 0 insertions(+), 1 deletions(-)
>
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index d93ecec..b226973 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -802,7 +802,6 @@ static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
>  	set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
>  	list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
>  	INIT_LIST_HEAD(&sp->oos_link);
> -	ASSERT(is_empty_shadow_page(sp->spt));
>  	bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS);
>  	sp->multimapped = 0;
>  	sp->parent_pte = parent_pte;
>   

sp->spt is allocated using mmu_memory_cache_alloc(), which zeros the 
page.  How can the assertion fail?


-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Re: [PATCH 3/3] kvm mmu: alloc shadow pages with __GFP_ZERO

[Joerg Roedel]

On Wed, Feb 18, 2009 at 02:03:34PM +0000, Avi Kivity wrote:
> Joerg Roedel wrote:
> >The assertion which the attached patch removes fails sometimes. Removing
> >this assertion is the alternative solution to this problem ;-)
> >
> >From ca45f3a2e45cd7e76ca624bb1098329db8ff83ab Mon Sep 17 00:00:00 2001
> >From: Joerg Roedel <joerg.roedel@amd.com>
> >Date: Wed, 18 Feb 2009 14:51:13 +0100
> >Subject: [PATCH] kvm mmu: remove assertion in kvm_mmu_alloc_page
> >
> >Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
> >---
> > arch/x86/kvm/mmu.c |    1 -
> > 1 files changed, 0 insertions(+), 1 deletions(-)
> >
> >diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> >index d93ecec..b226973 100644
> >--- a/arch/x86/kvm/mmu.c
> >+++ b/arch/x86/kvm/mmu.c
> >@@ -802,7 +802,6 @@ static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
> > 	set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
> > 	list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
> > 	INIT_LIST_HEAD(&sp->oos_link);
> >-	ASSERT(is_empty_shadow_page(sp->spt));
> > 	bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS);
> > 	sp->multimapped = 0;
> > 	sp->parent_pte = parent_pte;
> >  
> 
> sp->spt is allocated using mmu_memory_cache_alloc(), which zeros the page.  How can the assertion fail?

In the code I see (current kvm-git) mmu_memory_cache_alloc() does zero
nothing. It takes the page from the preallocated pool and returns it.
The pool itself is filled with mmu_topup_memory_caches() which calls
mmu_topup_memory_cache_page() to fill the mmu_page_cache (from which the
sp->spt page is allocated later). And the mmu_topup_memory_cache_page()
function calls alloc_page() and does not zero the result. This let the
assertion trigger.

Joerg

-- 
           | Advanced Micro Devices GmbH
 Operating | Karl-Hammerschmidt-Str. 34, 85609 Dornach bei München
 System    | 
 Research  | Geschäftsführer: Jochen Polster, Thomas M. McCoy, Giuliano Meroni
 Center    | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
           | Registergericht München, HRB Nr. 43632

Re: [PATCH 3/3] kvm mmu: alloc shadow pages with __GFP_ZERO

[Avi Kivity]

Joerg Roedel wrote:
>> sp->spt is allocated using mmu_memory_cache_alloc(), which zeros the page.  How can the assertion fail?
>>     
>
> In the code I see (current kvm-git) mmu_memory_cache_alloc() does zero
> nothing. It takes the page from the preallocated pool and returns it.
> The pool itself is filled with mmu_topup_memory_caches() which calls
> mmu_topup_memory_cache_page() to fill the mmu_page_cache (from which the
> sp->spt page is allocated later). And the mmu_topup_memory_cache_page()
> function calls alloc_page() and does not zero the result. This let the
> assertion trigger.
>   

Right, I was looking at the 2.6.29 tree.  The patch is correct (and the 
others look good as well).  As usual, I'd like Marcelo to take a look as 
well.

-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.

Re: [PATCH 1/3] kvm mmu: handle compound pages in kvm_is_mmio_pfn

[Marcelo Tosatti]

BTW some page bits are erroneously transferred to the struct page's
within the compound page. We've got away with that so far because
these bits (such as dirty and accessed) are not used by the limited
hugetlb/hugetlbfs implementation ATM.

Acked-by: Marcelo Tosatti <mtosatti@redhat.com>

On Wed, Feb 18, 2009 at 02:08:58PM +0100, Joerg Roedel wrote:
> The function kvm_is_mmio_pfn is called before put_page is called on a
> page by KVM. This is a problem when when this function is called on some
> struct page which is part of a compund page. It does not test the
> reserved flag of the compound page but of the struct page within the
> compount page. This is a problem when KVM works with hugepages allocated
> at boot time. These pages have the reserved bit set in all tail pages.
> Only the flag in the compount head is cleared. KVM would not put such a
> page which results in a memory leak.
> 
> Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
> ---
>  virt/kvm/kvm_main.c |    6 ++++--
>  1 files changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 266bdaf..0ed662d 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -535,8 +535,10 @@ static inline int valid_vcpu(int n)
>  
>  inline int kvm_is_mmio_pfn(pfn_t pfn)
>  {
> -	if (pfn_valid(pfn))
> -		return PageReserved(pfn_to_page(pfn));
> +	if (pfn_valid(pfn)) {
> +		struct page *page = compound_head(pfn_to_page(pfn));
> +		return PageReserved(page);
> +	}
>  
>  	return true;
>  }

Re: [PATCH 2/3] kvm mmu: remove redundant check in mmu_set_spte

[Marcelo Tosatti]

The following code flow is unnecessary:

	if (largepage)
		was_rmapped = is_large_pte(*shadow_pte);
	 else
	 	was_rmapped = 1;

The is_large_pte() function will always evaluate to one here because the
(largepage && !is_large_pte) case is already handled in the first
if-clause. So we can remove this check and set was_rmapped to one always
here.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Acked-by: Marcelo Tosatti <mtosatti@redhat.com>

---
 arch/x86/kvm/mmu.c |    8 ++------
 1 files changed, 2 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index ef060ec..c90b4b2 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -1791,12 +1791,8 @@ static void mmu_set_spte(struct kvm_vcpu *vcpu, u64 *shadow_pte,
 			pgprintk("hfn old %lx new %lx\n",
 				 spte_to_pfn(*shadow_pte), pfn);
 			rmap_remove(vcpu->kvm, shadow_pte);
-		} else {
-			if (largepage)
-				was_rmapped = is_large_pte(*shadow_pte);
-			else
-				was_rmapped = 1;
-		}
+		} else
+			was_rmapped = 1;
 	}
 	if (set_spte(vcpu, shadow_pte, pte_access, user_fault, write_fault,
 		      dirty, largepage, global, gfn, pfn, speculative, true)) {

Re: [PATCH 3/3] kvm mmu: alloc shadow pages with __GFP_ZERO

[Marcelo Tosatti]

On Wed, Feb 18, 2009 at 02:54:37PM +0100, Joerg Roedel wrote:
> > Adding __GFP_ZERO here will cause us to clear the page twice, which is wasteful.
> 
> The assertion which the attached patch removes fails sometimes. Removing
> this assertion is the alternative solution to this problem ;-)
> 

From: Joerg Roedel <joerg.roedel@amd.com>
Date: Wed, 18 Feb 2009 14:51:13 +0100
Subject: [PATCH] kvm mmu: remove assertion in kvm_mmu_alloc_page

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Acked-by: Marcelo Tosatti <mtosatti@redhat.com>

---
 arch/x86/kvm/mmu.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)
 
 diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
 index d93ecec..b226973 100644
 --- a/arch/x86/kvm/mmu.c
 +++ b/arch/x86/kvm/mmu.c
 @@ -802,7 +802,6 @@ static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
  	set_page_private(virt_to_page(sp->spt), (unsigned long)sp);
  	list_add(&sp->link, &vcpu->kvm->arch.active_mmu_pages);
  	INIT_LIST_HEAD(&sp->oos_link);
 -	ASSERT(is_empty_shadow_page(sp->spt));
  	bitmap_zero(sp->slot_bitmap, KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS);
  	sp->multimapped = 0;
  	sp->parent_pte = parent_pte;