[RFC][PATCH 0/2] s390/kvm: add kvm support for guest page hinting v2

[RFC][PATCH 0/2] s390/kvm: add kvm support for guest page hinting v2

[Martin Schwidefsky]

v1->v2:
 - found a way to simplify the common code patch

Linux on s390 as a guest under z/VM has been using the guest page
hinting interface (alias collaborative memory management) for a long
time. The full version with volatile states has been deemed to be too
complicated (see the old discussion about guest page hinting e.g. on
http://marc.info/?l=linux-mm&m=123816662017742&w=2).
What is currently implemented for the guest is the unused and stable
states to mark unallocated pages as freely available to the host.
This works just fine with z/VM as the host.

The two patches in this series implement the guest page hinting
interface for the unused and stable states in the KVM host.
Most of the code specific to s390 but there is a common memory
management part as well, see patch #1.

The code is working stable now, from my point of view this is ready
for prime-time.

Konstantin Weitz (2):
  mm: add support for discard of unused ptes
  s390/kvm: support collaborative memory management

 arch/s390/include/asm/kvm_host.h |    5 ++-
 arch/s390/include/asm/pgtable.h  |   24 ++++++++++++
 arch/s390/kvm/kvm-s390.c         |   25 +++++++++++++
 arch/s390/kvm/kvm-s390.h         |    2 +
 arch/s390/kvm/priv.c             |   41 ++++++++++++++++++++
 arch/s390/mm/pgtable.c           |   77 ++++++++++++++++++++++++++++++++++++++
 include/asm-generic/pgtable.h    |   13 +++++++
 mm/rmap.c                        |   10 +++++
 8 files changed, 196 insertions(+), 1 deletion(-)

-- 
1.7.9.5

[PATCH 1/2] mm: add support for discard of unused ptes

[Martin Schwidefsky]

From: Konstantin Weitz <konstantin.weitz@gmail.com>

In a virtualized environment and given an appropriate interface the guest
can mark pages as unused while they are free (for the s390 implementation
see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
the unused state is a property of the pte.

This patch adds the primitive 'pte_unused' and code to the host swap out
handler so that pages marked as unused by all mappers are not swapped out
but discarded instead, thus saving one IO for swap out and potentially
another one for swap in.

[ Martin Schwidefsky: patch reordering and simplification ]

Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
---
 include/asm-generic/pgtable.h |   13 +++++++++++++
 mm/rmap.c                     |   10 ++++++++++
 2 files changed, 23 insertions(+)

diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
index 2f47ade..ec540c5 100644
--- a/include/asm-generic/pgtable.h
+++ b/include/asm-generic/pgtable.h
@@ -193,6 +193,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
 }
 #endif
 
+#ifndef __HAVE_ARCH_PTE_UNUSED
+/*
+ * Some architectures provide facilities to virtualization guests
+ * so that they can flag allocated pages as unused. This allows the
+ * host to transparently reclaim unused pages. This function returns
+ * whether the pte's page is unused.
+ */
+static inline int pte_unused(pte_t pte)
+{
+	return 0;
+}
+#endif
+
 #ifndef __HAVE_ARCH_PMD_SAME
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
diff --git a/mm/rmap.c b/mm/rmap.c
index cd356df..2291f25 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1234,6 +1234,16 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 		}
 		set_pte_at(mm, address, pte,
 			   swp_entry_to_pte(make_hwpoison_entry(page)));
+	} else if (pte_unused(pteval)) {
+		/*
+		 * The guest indicated that the page content is of no
+		 * interest anymore. Simply discard the pte, vmscan
+		 * will take care of the rest.
+		 */
+		if (PageAnon(page))
+			dec_mm_counter(mm, MM_ANONPAGES);
+		else
+			dec_mm_counter(mm, MM_FILEPAGES);
 	} else if (PageAnon(page)) {
 		swp_entry_t entry = { .val = page_private(page) };
 
-- 
1.7.9.5

[PATCH 1/2] mm: add support for discard of unused ptes

[Martin Schwidefsky]

From: Konstantin Weitz <konstantin.weitz@gmail.com>

In a virtualized environment and given an appropriate interface the guest
can mark pages as unused while they are free (for the s390 implementation
see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
the unused state is a property of the pte.

This patch adds the primitive 'pte_unused' and code to the host swap out
handler so that pages marked as unused by all mappers are not swapped out
but discarded instead, thus saving one IO for swap out and potentially
another one for swap in.

[ Martin Schwidefsky: patch reordering and simplification ]

Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
---
 include/asm-generic/pgtable.h |   13 +++++++++++++
 mm/rmap.c                     |   10 ++++++++++
 2 files changed, 23 insertions(+)

diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
index 2f47ade..ec540c5 100644
--- a/include/asm-generic/pgtable.h
+++ b/include/asm-generic/pgtable.h
@@ -193,6 +193,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
 }
 #endif
 
+#ifndef __HAVE_ARCH_PTE_UNUSED
+/*
+ * Some architectures provide facilities to virtualization guests
+ * so that they can flag allocated pages as unused. This allows the
+ * host to transparently reclaim unused pages. This function returns
+ * whether the pte's page is unused.
+ */
+static inline int pte_unused(pte_t pte)
+{
+	return 0;
+}
+#endif
+
 #ifndef __HAVE_ARCH_PMD_SAME
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
diff --git a/mm/rmap.c b/mm/rmap.c
index cd356df..2291f25 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1234,6 +1234,16 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 		}
 		set_pte_at(mm, address, pte,
 			   swp_entry_to_pte(make_hwpoison_entry(page)));
+	} else if (pte_unused(pteval)) {
+		/*
+		 * The guest indicated that the page content is of no
+		 * interest anymore. Simply discard the pte, vmscan
+		 * will take care of the rest.
+		 */
+		if (PageAnon(page))
+			dec_mm_counter(mm, MM_ANONPAGES);
+		else
+			dec_mm_counter(mm, MM_FILEPAGES);
 	} else if (PageAnon(page)) {
 		swp_entry_t entry = { .val = page_private(page) };
 
-- 
1.7.9.5

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 1/2] mm: add support for discard of unused ptes

[Christian Borntraeger]

On 25/07/13 10:54, Martin Schwidefsky wrote:
> From: Konstantin Weitz <konstantin.weitz@gmail.com>
> 
> In a virtualized environment and given an appropriate interface the guest
> can mark pages as unused while they are free (for the s390 implementation
> see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
> the unused state is a property of the pte.
> 
> This patch adds the primitive 'pte_unused' and code to the host swap out
> handler so that pages marked as unused by all mappers are not swapped out
> but discarded instead, thus saving one IO for swap out and potentially
> another one for swap in.
> 
> [ Martin Schwidefsky: patch reordering and simplification ]
> 
> Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>

> ---
>  include/asm-generic/pgtable.h |   13 +++++++++++++
>  mm/rmap.c                     |   10 ++++++++++
>  2 files changed, 23 insertions(+)
> 
> diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
> index 2f47ade..ec540c5 100644
> --- a/include/asm-generic/pgtable.h
> +++ b/include/asm-generic/pgtable.h
> @@ -193,6 +193,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
>  }
>  #endif
> 
> +#ifndef __HAVE_ARCH_PTE_UNUSED
> +/*
> + * Some architectures provide facilities to virtualization guests
> + * so that they can flag allocated pages as unused. This allows the
> + * host to transparently reclaim unused pages. This function returns
> + * whether the pte's page is unused.
> + */
> +static inline int pte_unused(pte_t pte)
> +{
> +	return 0;
> +}
> +#endif
> +
>  #ifndef __HAVE_ARCH_PMD_SAME
>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>  static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
> diff --git a/mm/rmap.c b/mm/rmap.c
> index cd356df..2291f25 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1234,6 +1234,16 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
>  		}
>  		set_pte_at(mm, address, pte,
>  			   swp_entry_to_pte(make_hwpoison_entry(page)));
> +	} else if (pte_unused(pteval)) {
> +		/*
> +		 * The guest indicated that the page content is of no
> +		 * interest anymore. Simply discard the pte, vmscan
> +		 * will take care of the rest.
> +		 */
> +		if (PageAnon(page))
> +			dec_mm_counter(mm, MM_ANONPAGES);
> +		else
> +			dec_mm_counter(mm, MM_FILEPAGES);
>  	} else if (PageAnon(page)) {
>  		swp_entry_t entry = { .val = page_private(page) };
> 

Re: [PATCH 1/2] mm: add support for discard of unused ptes

[Christian Borntraeger]

On 25/07/13 10:54, Martin Schwidefsky wrote:
> From: Konstantin Weitz <konstantin.weitz@gmail.com>
> 
> In a virtualized environment and given an appropriate interface the guest
> can mark pages as unused while they are free (for the s390 implementation
> see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
> the unused state is a property of the pte.
> 
> This patch adds the primitive 'pte_unused' and code to the host swap out
> handler so that pages marked as unused by all mappers are not swapped out
> but discarded instead, thus saving one IO for swap out and potentially
> another one for swap in.
> 
> [ Martin Schwidefsky: patch reordering and simplification ]
> 
> Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>

> ---
>  include/asm-generic/pgtable.h |   13 +++++++++++++
>  mm/rmap.c                     |   10 ++++++++++
>  2 files changed, 23 insertions(+)
> 
> diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
> index 2f47ade..ec540c5 100644
> --- a/include/asm-generic/pgtable.h
> +++ b/include/asm-generic/pgtable.h
> @@ -193,6 +193,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
>  }
>  #endif
> 
> +#ifndef __HAVE_ARCH_PTE_UNUSED
> +/*
> + * Some architectures provide facilities to virtualization guests
> + * so that they can flag allocated pages as unused. This allows the
> + * host to transparently reclaim unused pages. This function returns
> + * whether the pte's page is unused.
> + */
> +static inline int pte_unused(pte_t pte)
> +{
> +	return 0;
> +}
> +#endif
> +
>  #ifndef __HAVE_ARCH_PMD_SAME
>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>  static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
> diff --git a/mm/rmap.c b/mm/rmap.c
> index cd356df..2291f25 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1234,6 +1234,16 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
>  		}
>  		set_pte_at(mm, address, pte,
>  			   swp_entry_to_pte(make_hwpoison_entry(page)));
> +	} else if (pte_unused(pteval)) {
> +		/*
> +		 * The guest indicated that the page content is of no
> +		 * interest anymore. Simply discard the pte, vmscan
> +		 * will take care of the rest.
> +		 */
> +		if (PageAnon(page))
> +			dec_mm_counter(mm, MM_ANONPAGES);
> +		else
> +			dec_mm_counter(mm, MM_FILEPAGES);
>  	} else if (PageAnon(page)) {
>  		swp_entry_t entry = { .val = page_private(page) };
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 1/2] mm: add support for discard of unused ptes

[Andrew Morton]

On Thu, 25 Jul 2013 10:54:20 +0200 Martin Schwidefsky <schwidefsky@de.ibm.com> wrote:

> From: Konstantin Weitz <konstantin.weitz@gmail.com>
> 
> In a virtualized environment and given an appropriate interface the guest
> can mark pages as unused while they are free (for the s390 implementation
> see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
> the unused state is a property of the pte.
> 
> This patch adds the primitive 'pte_unused' and code to the host swap out
> handler so that pages marked as unused by all mappers are not swapped out
> but discarded instead, thus saving one IO for swap out and potentially
> another one for swap in.
> 
> ...
>
> --- a/include/asm-generic/pgtable.h
> +++ b/include/asm-generic/pgtable.h
> @@ -193,6 +193,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
>  }
>  #endif
>  
> +#ifndef __HAVE_ARCH_PTE_UNUSED
> +/*
> + * Some architectures provide facilities to virtualization guests
> + * so that they can flag allocated pages as unused. This allows the
> + * host to transparently reclaim unused pages. This function returns
> + * whether the pte's page is unused.
> + */
> +static inline int pte_unused(pte_t pte)
> +{
> +	return 0;
> +}
> +#endif
> +
>  #ifndef __HAVE_ARCH_PMD_SAME
>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>  static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
> diff --git a/mm/rmap.c b/mm/rmap.c
> index cd356df..2291f25 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1234,6 +1234,16 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
>  		}
>  		set_pte_at(mm, address, pte,
>  			   swp_entry_to_pte(make_hwpoison_entry(page)));
> +	} else if (pte_unused(pteval)) {
> +		/*
> +		 * The guest indicated that the page content is of no
> +		 * interest anymore. Simply discard the pte, vmscan
> +		 * will take care of the rest.
> +		 */
> +		if (PageAnon(page))
> +			dec_mm_counter(mm, MM_ANONPAGES);
> +		else
> +			dec_mm_counter(mm, MM_FILEPAGES);
>  	} else if (PageAnon(page)) {
>  		swp_entry_t entry = { .val = page_private(page) };

Obviously harmless.  Please include this in whatever tree carries
"[PATCH 2/2] s390/kvm: support collaborative memory management".

Re: [PATCH 1/2] mm: add support for discard of unused ptes

[Andrew Morton]

On Thu, 25 Jul 2013 10:54:20 +0200 Martin Schwidefsky <schwidefsky@de.ibm.com> wrote:

> From: Konstantin Weitz <konstantin.weitz@gmail.com>
> 
> In a virtualized environment and given an appropriate interface the guest
> can mark pages as unused while they are free (for the s390 implementation
> see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
> the unused state is a property of the pte.
> 
> This patch adds the primitive 'pte_unused' and code to the host swap out
> handler so that pages marked as unused by all mappers are not swapped out
> but discarded instead, thus saving one IO for swap out and potentially
> another one for swap in.
> 
> ...
>
> --- a/include/asm-generic/pgtable.h
> +++ b/include/asm-generic/pgtable.h
> @@ -193,6 +193,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
>  }
>  #endif
>  
> +#ifndef __HAVE_ARCH_PTE_UNUSED
> +/*
> + * Some architectures provide facilities to virtualization guests
> + * so that they can flag allocated pages as unused. This allows the
> + * host to transparently reclaim unused pages. This function returns
> + * whether the pte's page is unused.
> + */
> +static inline int pte_unused(pte_t pte)
> +{
> +	return 0;
> +}
> +#endif
> +
>  #ifndef __HAVE_ARCH_PMD_SAME
>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>  static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
> diff --git a/mm/rmap.c b/mm/rmap.c
> index cd356df..2291f25 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1234,6 +1234,16 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
>  		}
>  		set_pte_at(mm, address, pte,
>  			   swp_entry_to_pte(make_hwpoison_entry(page)));
> +	} else if (pte_unused(pteval)) {
> +		/*
> +		 * The guest indicated that the page content is of no
> +		 * interest anymore. Simply discard the pte, vmscan
> +		 * will take care of the rest.
> +		 */
> +		if (PageAnon(page))
> +			dec_mm_counter(mm, MM_ANONPAGES);
> +		else
> +			dec_mm_counter(mm, MM_FILEPAGES);
>  	} else if (PageAnon(page)) {
>  		swp_entry_t entry = { .val = page_private(page) };

Obviously harmless.  Please include this in whatever tree carries
"[PATCH 2/2] s390/kvm: support collaborative memory management".

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 1/2] mm: add support for discard of unused ptes

[Martin Schwidefsky]

On Tue, 30 Jul 2013 13:44:22 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:

> On Thu, 25 Jul 2013 10:54:20 +0200 Martin Schwidefsky <schwidefsky@de.ibm.com> wrote:
> 
> > From: Konstantin Weitz <konstantin.weitz@gmail.com>
> > 
> > In a virtualized environment and given an appropriate interface the guest
> > can mark pages as unused while they are free (for the s390 implementation
> > see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
> > the unused state is a property of the pte.
> > 
> > This patch adds the primitive 'pte_unused' and code to the host swap out
> > handler so that pages marked as unused by all mappers are not swapped out
> > but discarded instead, thus saving one IO for swap out and potentially
> > another one for swap in.
> > 
> > ...
> >
> > --- a/include/asm-generic/pgtable.h
> > +++ b/include/asm-generic/pgtable.h
> > @@ -193,6 +193,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
> >  }
> >  #endif
> >  
> > +#ifndef __HAVE_ARCH_PTE_UNUSED
> > +/*
> > + * Some architectures provide facilities to virtualization guests
> > + * so that they can flag allocated pages as unused. This allows the
> > + * host to transparently reclaim unused pages. This function returns
> > + * whether the pte's page is unused.
> > + */
> > +static inline int pte_unused(pte_t pte)
> > +{
> > +	return 0;
> > +}
> > +#endif
> > +
> >  #ifndef __HAVE_ARCH_PMD_SAME
> >  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
> >  static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
> > diff --git a/mm/rmap.c b/mm/rmap.c
> > index cd356df..2291f25 100644
> > --- a/mm/rmap.c
> > +++ b/mm/rmap.c
> > @@ -1234,6 +1234,16 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> >  		}
> >  		set_pte_at(mm, address, pte,
> >  			   swp_entry_to_pte(make_hwpoison_entry(page)));
> > +	} else if (pte_unused(pteval)) {
> > +		/*
> > +		 * The guest indicated that the page content is of no
> > +		 * interest anymore. Simply discard the pte, vmscan
> > +		 * will take care of the rest.
> > +		 */
> > +		if (PageAnon(page))
> > +			dec_mm_counter(mm, MM_ANONPAGES);
> > +		else
> > +			dec_mm_counter(mm, MM_FILEPAGES);
> >  	} else if (PageAnon(page)) {
> >  		swp_entry_t entry = { .val = page_private(page) };
> 
> Obviously harmless.  Please include this in whatever tree carries
> "[PATCH 2/2] s390/kvm: support collaborative memory management".
 
Cool, thanks. This will go out via the KVM tree then.

-- 
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.

Re: [PATCH 1/2] mm: add support for discard of unused ptes

[Martin Schwidefsky]

On Tue, 30 Jul 2013 13:44:22 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:

> On Thu, 25 Jul 2013 10:54:20 +0200 Martin Schwidefsky <schwidefsky@de.ibm.com> wrote:
> 
> > From: Konstantin Weitz <konstantin.weitz@gmail.com>
> > 
> > In a virtualized environment and given an appropriate interface the guest
> > can mark pages as unused while they are free (for the s390 implementation
> > see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
> > the unused state is a property of the pte.
> > 
> > This patch adds the primitive 'pte_unused' and code to the host swap out
> > handler so that pages marked as unused by all mappers are not swapped out
> > but discarded instead, thus saving one IO for swap out and potentially
> > another one for swap in.
> > 
> > ...
> >
> > --- a/include/asm-generic/pgtable.h
> > +++ b/include/asm-generic/pgtable.h
> > @@ -193,6 +193,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
> >  }
> >  #endif
> >  
> > +#ifndef __HAVE_ARCH_PTE_UNUSED
> > +/*
> > + * Some architectures provide facilities to virtualization guests
> > + * so that they can flag allocated pages as unused. This allows the
> > + * host to transparently reclaim unused pages. This function returns
> > + * whether the pte's page is unused.
> > + */
> > +static inline int pte_unused(pte_t pte)
> > +{
> > +	return 0;
> > +}
> > +#endif
> > +
> >  #ifndef __HAVE_ARCH_PMD_SAME
> >  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
> >  static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
> > diff --git a/mm/rmap.c b/mm/rmap.c
> > index cd356df..2291f25 100644
> > --- a/mm/rmap.c
> > +++ b/mm/rmap.c
> > @@ -1234,6 +1234,16 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> >  		}
> >  		set_pte_at(mm, address, pte,
> >  			   swp_entry_to_pte(make_hwpoison_entry(page)));
> > +	} else if (pte_unused(pteval)) {
> > +		/*
> > +		 * The guest indicated that the page content is of no
> > +		 * interest anymore. Simply discard the pte, vmscan
> > +		 * will take care of the rest.
> > +		 */
> > +		if (PageAnon(page))
> > +			dec_mm_counter(mm, MM_ANONPAGES);
> > +		else
> > +			dec_mm_counter(mm, MM_FILEPAGES);
> >  	} else if (PageAnon(page)) {
> >  		swp_entry_t entry = { .val = page_private(page) };
> 
> Obviously harmless.  Please include this in whatever tree carries
> "[PATCH 2/2] s390/kvm: support collaborative memory management".
 
Cool, thanks. This will go out via the KVM tree then.

-- 
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[PATCH 2/2] s390/kvm: support collaborative memory management

[Martin Schwidefsky]

From: Konstantin Weitz <konstantin.weitz@gmail.com>

This patch enables Collaborative Memory Management (CMM) for kvm
on s390. CMM allows the guest to inform the host about page usage
(see arch/s390/mm/cmm.c). The host uses this information to avoid
swapping in unused pages in the page fault handler. Further, a CPU
provided list of unused invalid pages is processed to reclaim swap
space of not yet accessed unused pages.

[ Martin Schwidefsky: patch reordering and cleanup ]

Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
---
 arch/s390/include/asm/kvm_host.h |    5 ++-
 arch/s390/include/asm/pgtable.h  |   24 ++++++++++++
 arch/s390/kvm/kvm-s390.c         |   25 +++++++++++++
 arch/s390/kvm/kvm-s390.h         |    2 +
 arch/s390/kvm/priv.c             |   41 ++++++++++++++++++++
 arch/s390/mm/pgtable.c           |   77 ++++++++++++++++++++++++++++++++++++++
 6 files changed, 173 insertions(+), 1 deletion(-)

diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
index 3238d40..de6450e 100644
--- a/arch/s390/include/asm/kvm_host.h
+++ b/arch/s390/include/asm/kvm_host.h
@@ -113,7 +113,9 @@ struct kvm_s390_sie_block {
 	__u64	gbea;			/* 0x0180 */
 	__u8	reserved188[24];	/* 0x0188 */
 	__u32	fac;			/* 0x01a0 */
-	__u8	reserved1a4[92];	/* 0x01a4 */
+	__u8	reserved1a4[20];	/* 0x01a4 */
+	__u64	cbrlo;			/* 0x01b8 */
+	__u8	reserved1c0[64];	/* 0x01c0 */
 } __attribute__((packed));
 
 struct kvm_vcpu_stat {
@@ -149,6 +151,7 @@ struct kvm_vcpu_stat {
 	u32 instruction_stsi;
 	u32 instruction_stfl;
 	u32 instruction_tprot;
+	u32 instruction_essa;
 	u32 instruction_sigp_sense;
 	u32 instruction_sigp_sense_running;
 	u32 instruction_sigp_external_call;
diff --git a/arch/s390/include/asm/pgtable.h b/arch/s390/include/asm/pgtable.h
index 75fb726..65d48b8 100644
--- a/arch/s390/include/asm/pgtable.h
+++ b/arch/s390/include/asm/pgtable.h
@@ -227,6 +227,7 @@ extern unsigned long MODULES_END;
 #define _PAGE_SWR	0x008		/* SW pte referenced bit */
 #define _PAGE_SWW	0x010		/* SW pte write bit */
 #define _PAGE_SPECIAL	0x020		/* SW associated with special page */
+#define _PAGE_UNUSED	0x040		/* SW bit for ptep_clear_flush() */
 #define __HAVE_ARCH_PTE_SPECIAL
 
 /* Set of bits not changed in pte_modify */
@@ -375,6 +376,12 @@ extern unsigned long MODULES_END;
 
 #endif /* CONFIG_64BIT */
 
+/* Guest Page State used for virtualization */
+#define _PGSTE_GPS_ZERO 	0x0000000080000000UL
+#define _PGSTE_GPS_USAGE_MASK	0x0000000003000000UL
+#define _PGSTE_GPS_USAGE_STABLE 0x0000000000000000UL
+#define _PGSTE_GPS_USAGE_UNUSED 0x0000000001000000UL
+
 /*
  * A user page table pointer has the space-switch-event bit, the
  * private-space-control bit and the storage-alteration-event-control
@@ -590,6 +597,12 @@ static inline int pte_file(pte_t pte)
 	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
 }
 
+static inline int pte_swap(pte_t pte)
+{
+	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
+	return (pte_val(pte) & mask) == _PAGE_TYPE_SWAP;
+}
+
 static inline int pte_special(pte_t pte)
 {
 	return (pte_val(pte) & _PAGE_SPECIAL);
@@ -794,6 +807,7 @@ unsigned long gmap_translate(unsigned long address, struct gmap *);
 unsigned long __gmap_fault(unsigned long address, struct gmap *);
 unsigned long gmap_fault(unsigned long address, struct gmap *);
 void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
+void __gmap_zap(unsigned long address, struct gmap *);
 
 void gmap_register_ipte_notifier(struct gmap_notifier *);
 void gmap_unregister_ipte_notifier(struct gmap_notifier *);
@@ -825,6 +839,7 @@ static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
 
 	if (mm_has_pgste(mm)) {
 		pgste = pgste_get_lock(ptep);
+		pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
 		pgste_set_key(ptep, pgste, entry);
 		pgste_set_pte(ptep, entry);
 		pgste_set_unlock(ptep, pgste);
@@ -858,6 +873,12 @@ static inline int pte_young(pte_t pte)
 	return 0;
 }
 
+#define __HAVE_ARCH_PTE_UNUSED
+static inline int pte_unused(pte_t pte)
+{
+	return pte_val(pte) & _PAGE_UNUSED;
+}
+
 /*
  * pgd/pmd/pte modification functions
  */
@@ -1142,6 +1163,9 @@ static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
 
 	if (mm_has_pgste(vma->vm_mm)) {
+		if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
+		    _PGSTE_GPS_USAGE_UNUSED)
+			pte_val(pte) |= _PAGE_UNUSED;
 		pgste = pgste_update_all(&pte, pgste);
 		pgste_set_unlock(ptep, pgste);
 	}
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index ba694d2..052a773 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -67,6 +67,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
 	{ "instruction_stsch", VCPU_STAT(instruction_stsch) },
 	{ "instruction_chsc", VCPU_STAT(instruction_chsc) },
+	{ "instruction_essa", VCPU_STAT(instruction_essa) },
 	{ "instruction_stsi", VCPU_STAT(instruction_stsi) },
 	{ "instruction_stfl", VCPU_STAT(instruction_stfl) },
 	{ "instruction_tprot", VCPU_STAT(instruction_tprot) },
@@ -276,7 +277,11 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 	if (kvm_is_ucontrol(vcpu->kvm))
 		gmap_free(vcpu->arch.gmap);
 
+	if (vcpu->arch.sie_block->cbrlo)
+		__free_page(__pfn_to_page(
+				vcpu->arch.sie_block->cbrlo >> PAGE_SHIFT));
 	free_page((unsigned long)(vcpu->arch.sie_block));
+
 	kvm_vcpu_uninit(vcpu);
 	kmem_cache_free(kvm_vcpu_cache, vcpu);
 }
@@ -380,6 +385,8 @@ int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 
 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 {
+	struct page *cbrl;
+
 	atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
 						    CPUSTAT_SM |
 						    CPUSTAT_STOPPED |
@@ -388,6 +395,14 @@ int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 	vcpu->arch.sie_block->ecb2  = 8;
 	vcpu->arch.sie_block->eca   = 0xC1002001U;
 	vcpu->arch.sie_block->fac   = (int) (long) facilities;
+	if (kvm_enabled_cmma()) {
+		cbrl = alloc_page(GFP_KERNEL | __GFP_ZERO);
+		if (cbrl) {
+			vcpu->arch.sie_block->ecb2 |= 0x80;
+			vcpu->arch.sie_block->ecb2 &= ~0x08;
+			vcpu->arch.sie_block->cbrlo = page_to_phys(cbrl);
+		}
+	}
 	hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 	tasklet_init(&vcpu->arch.tasklet, kvm_s390_tasklet,
 		     (unsigned long) vcpu);
@@ -682,6 +697,16 @@ static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+bool kvm_enabled_cmma(void)
+{
+	if (!MACHINE_IS_LPAR)
+		return false;
+	/* only enable for z10 and later */
+	if (!MACHINE_HAS_EDAT1)
+		return false;
+	return true;
+}
+
 static int __vcpu_run(struct kvm_vcpu *vcpu)
 {
 	int rc;
diff --git a/arch/s390/kvm/kvm-s390.h b/arch/s390/kvm/kvm-s390.h
index 028ca9f..ea3d5f7 100644
--- a/arch/s390/kvm/kvm-s390.h
+++ b/arch/s390/kvm/kvm-s390.h
@@ -145,6 +145,8 @@ void s390_vcpu_block(struct kvm_vcpu *vcpu);
 void s390_vcpu_unblock(struct kvm_vcpu *vcpu);
 void exit_sie(struct kvm_vcpu *vcpu);
 void exit_sie_sync(struct kvm_vcpu *vcpu);
+/* are we going to support cmma? */
+bool kvm_enabled_cmma(void);
 /* implemented in diag.c */
 int kvm_s390_handle_diag(struct kvm_vcpu *vcpu);
 
diff --git a/arch/s390/kvm/priv.c b/arch/s390/kvm/priv.c
index 0da3e6e..e0703db 100644
--- a/arch/s390/kvm/priv.c
+++ b/arch/s390/kvm/priv.c
@@ -581,9 +581,50 @@ static int handle_pfmf(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+static int handle_essa(struct kvm_vcpu *vcpu)
+{
+	/* entries expected to be 1FF */
+	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
+	unsigned long *cbrlo, cbrle;
+	struct gmap *gmap;
+	int i;
+
+	VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
+	gmap = vcpu->arch.gmap;
+	vcpu->stat.instruction_essa++;
+	if (!kvm_enabled_cmma() || !vcpu->arch.sie_block->cbrlo)
+		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
+
+	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
+		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
+
+	if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
+		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
+
+	/* Rewind PSW to repeat the ESSA instruction */
+	vcpu->arch.sie_block->gpsw.addr =
+		__rewind_psw(vcpu->arch.sie_block->gpsw, 4);
+	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
+	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
+	down_read(&gmap->mm->mmap_sem);
+	for (i = 0; i < entries; ++i) {
+		cbrle = cbrlo[i];
+		if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
+			/* invalid entry */
+			break;
+		/* try to free backing */
+		__gmap_zap(cbrle, gmap);
+	}
+	up_read(&gmap->mm->mmap_sem);
+	if (i < entries)
+		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
+	return 0;
+}
+
 static const intercept_handler_t b9_handlers[256] = {
 	[0x8d] = handle_epsw,
 	[0x9c] = handle_io_inst,
+	[0xab] = handle_essa,
 	[0xaf] = handle_pfmf,
 };
 
diff --git a/arch/s390/mm/pgtable.c b/arch/s390/mm/pgtable.c
index a8154a1..567f89e 100644
--- a/arch/s390/mm/pgtable.c
+++ b/arch/s390/mm/pgtable.c
@@ -17,6 +17,7 @@
 #include <linux/quicklist.h>
 #include <linux/rcupdate.h>
 #include <linux/slab.h>
+#include <linux/swapops.h>
 
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -573,6 +574,82 @@ unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
 }
 EXPORT_SYMBOL_GPL(gmap_fault);
 
+static void gmap_zap_swap_entry(swp_entry_t entry, struct mm_struct *mm)
+{
+	if (!non_swap_entry(entry))
+		dec_mm_counter(mm, MM_SWAPENTS);
+	else if (is_migration_entry(entry)) {
+		struct page *page = migration_entry_to_page(entry);
+
+		if (PageAnon(page))
+			dec_mm_counter(mm, MM_ANONPAGES);
+		else
+			dec_mm_counter(mm, MM_FILEPAGES);
+	}
+	free_swap_and_cache(entry);
+}
+
+/**
+ * The mm->mmap_sem lock must be held
+ */
+static void gmap_zap_unused(struct mm_struct *mm, unsigned long address)
+{
+	unsigned long ptev, pgstev;
+	spinlock_t *ptl;
+	pgste_t pgste;
+	pte_t *ptep, pte;
+
+	ptep = get_locked_pte(mm, address, &ptl);
+	if (unlikely(!ptep))
+		return;
+	pte = *ptep;
+	if (!pte_swap(pte))
+		goto out_pte;
+	/* Zap unused and logically-zero pages */
+	pgste = pgste_get_lock(ptep);
+	pgstev = pgste_val(pgste);
+	ptev = pte_val(pte);
+	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
+	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) {
+		gmap_zap_swap_entry(pte_to_swp_entry(pte), mm);
+		pte_clear(mm, address, ptep);
+	}
+	pgste_set_unlock(ptep, pgste);
+out_pte:
+	pte_unmap_unlock(*ptep, ptl);
+}
+
+/*
+ * this function is assumed to be called with mmap_sem held
+ */
+void __gmap_zap(unsigned long address, struct gmap *gmap)
+{
+	unsigned long *table, *segment_ptr;
+	unsigned long segment, pgstev, ptev;
+	struct gmap_pgtable *mp;
+	struct page *page;
+
+	segment_ptr = gmap_table_walk(address, gmap);
+	if (IS_ERR(segment_ptr))
+		return;
+	segment = *segment_ptr;
+	if (segment & _SEGMENT_ENTRY_INV)
+		return;
+	page = pfn_to_page(segment >> PAGE_SHIFT);
+	mp = (struct gmap_pgtable *) page->index;
+	address = mp->vmaddr | (address & ~PMD_MASK);
+	/* Page table is present */
+	table = (unsigned long *)(segment & _SEGMENT_ENTRY_ORIGIN);
+	table = table + ((address >> 12) & 0xff);
+	pgstev = table[PTRS_PER_PTE];
+	ptev = table[0];
+	/* quick check, checked again with locks held */
+	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
+	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID)))
+		gmap_zap_unused(gmap->mm, address);
+}
+EXPORT_SYMBOL_GPL(__gmap_zap);
+
 void gmap_discard(unsigned long from, unsigned long to, struct gmap *gmap)
 {
 
-- 
1.7.9.5

[PATCH 2/2] s390/kvm: support collaborative memory management

[Martin Schwidefsky]

From: Konstantin Weitz <konstantin.weitz@gmail.com>

This patch enables Collaborative Memory Management (CMM) for kvm
on s390. CMM allows the guest to inform the host about page usage
(see arch/s390/mm/cmm.c). The host uses this information to avoid
swapping in unused pages in the page fault handler. Further, a CPU
provided list of unused invalid pages is processed to reclaim swap
space of not yet accessed unused pages.

[ Martin Schwidefsky: patch reordering and cleanup ]

Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
---
 arch/s390/include/asm/kvm_host.h |    5 ++-
 arch/s390/include/asm/pgtable.h  |   24 ++++++++++++
 arch/s390/kvm/kvm-s390.c         |   25 +++++++++++++
 arch/s390/kvm/kvm-s390.h         |    2 +
 arch/s390/kvm/priv.c             |   41 ++++++++++++++++++++
 arch/s390/mm/pgtable.c           |   77 ++++++++++++++++++++++++++++++++++++++
 6 files changed, 173 insertions(+), 1 deletion(-)

diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
index 3238d40..de6450e 100644
--- a/arch/s390/include/asm/kvm_host.h
+++ b/arch/s390/include/asm/kvm_host.h
@@ -113,7 +113,9 @@ struct kvm_s390_sie_block {
 	__u64	gbea;			/* 0x0180 */
 	__u8	reserved188[24];	/* 0x0188 */
 	__u32	fac;			/* 0x01a0 */
-	__u8	reserved1a4[92];	/* 0x01a4 */
+	__u8	reserved1a4[20];	/* 0x01a4 */
+	__u64	cbrlo;			/* 0x01b8 */
+	__u8	reserved1c0[64];	/* 0x01c0 */
 } __attribute__((packed));
 
 struct kvm_vcpu_stat {
@@ -149,6 +151,7 @@ struct kvm_vcpu_stat {
 	u32 instruction_stsi;
 	u32 instruction_stfl;
 	u32 instruction_tprot;
+	u32 instruction_essa;
 	u32 instruction_sigp_sense;
 	u32 instruction_sigp_sense_running;
 	u32 instruction_sigp_external_call;
diff --git a/arch/s390/include/asm/pgtable.h b/arch/s390/include/asm/pgtable.h
index 75fb726..65d48b8 100644
--- a/arch/s390/include/asm/pgtable.h
+++ b/arch/s390/include/asm/pgtable.h
@@ -227,6 +227,7 @@ extern unsigned long MODULES_END;
 #define _PAGE_SWR	0x008		/* SW pte referenced bit */
 #define _PAGE_SWW	0x010		/* SW pte write bit */
 #define _PAGE_SPECIAL	0x020		/* SW associated with special page */
+#define _PAGE_UNUSED	0x040		/* SW bit for ptep_clear_flush() */
 #define __HAVE_ARCH_PTE_SPECIAL
 
 /* Set of bits not changed in pte_modify */
@@ -375,6 +376,12 @@ extern unsigned long MODULES_END;
 
 #endif /* CONFIG_64BIT */
 
+/* Guest Page State used for virtualization */
+#define _PGSTE_GPS_ZERO 	0x0000000080000000UL
+#define _PGSTE_GPS_USAGE_MASK	0x0000000003000000UL
+#define _PGSTE_GPS_USAGE_STABLE 0x0000000000000000UL
+#define _PGSTE_GPS_USAGE_UNUSED 0x0000000001000000UL
+
 /*
  * A user page table pointer has the space-switch-event bit, the
  * private-space-control bit and the storage-alteration-event-control
@@ -590,6 +597,12 @@ static inline int pte_file(pte_t pte)
 	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
 }
 
+static inline int pte_swap(pte_t pte)
+{
+	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
+	return (pte_val(pte) & mask) == _PAGE_TYPE_SWAP;
+}
+
 static inline int pte_special(pte_t pte)
 {
 	return (pte_val(pte) & _PAGE_SPECIAL);
@@ -794,6 +807,7 @@ unsigned long gmap_translate(unsigned long address, struct gmap *);
 unsigned long __gmap_fault(unsigned long address, struct gmap *);
 unsigned long gmap_fault(unsigned long address, struct gmap *);
 void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
+void __gmap_zap(unsigned long address, struct gmap *);
 
 void gmap_register_ipte_notifier(struct gmap_notifier *);
 void gmap_unregister_ipte_notifier(struct gmap_notifier *);
@@ -825,6 +839,7 @@ static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
 
 	if (mm_has_pgste(mm)) {
 		pgste = pgste_get_lock(ptep);
+		pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
 		pgste_set_key(ptep, pgste, entry);
 		pgste_set_pte(ptep, entry);
 		pgste_set_unlock(ptep, pgste);
@@ -858,6 +873,12 @@ static inline int pte_young(pte_t pte)
 	return 0;
 }
 
+#define __HAVE_ARCH_PTE_UNUSED
+static inline int pte_unused(pte_t pte)
+{
+	return pte_val(pte) & _PAGE_UNUSED;
+}
+
 /*
  * pgd/pmd/pte modification functions
  */
@@ -1142,6 +1163,9 @@ static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
 	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
 
 	if (mm_has_pgste(vma->vm_mm)) {
+		if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
+		    _PGSTE_GPS_USAGE_UNUSED)
+			pte_val(pte) |= _PAGE_UNUSED;
 		pgste = pgste_update_all(&pte, pgste);
 		pgste_set_unlock(ptep, pgste);
 	}
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index ba694d2..052a773 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -67,6 +67,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
 	{ "instruction_stsch", VCPU_STAT(instruction_stsch) },
 	{ "instruction_chsc", VCPU_STAT(instruction_chsc) },
+	{ "instruction_essa", VCPU_STAT(instruction_essa) },
 	{ "instruction_stsi", VCPU_STAT(instruction_stsi) },
 	{ "instruction_stfl", VCPU_STAT(instruction_stfl) },
 	{ "instruction_tprot", VCPU_STAT(instruction_tprot) },
@@ -276,7 +277,11 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 	if (kvm_is_ucontrol(vcpu->kvm))
 		gmap_free(vcpu->arch.gmap);
 
+	if (vcpu->arch.sie_block->cbrlo)
+		__free_page(__pfn_to_page(
+				vcpu->arch.sie_block->cbrlo >> PAGE_SHIFT));
 	free_page((unsigned long)(vcpu->arch.sie_block));
+
 	kvm_vcpu_uninit(vcpu);
 	kmem_cache_free(kvm_vcpu_cache, vcpu);
 }
@@ -380,6 +385,8 @@ int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 
 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 {
+	struct page *cbrl;
+
 	atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
 						    CPUSTAT_SM |
 						    CPUSTAT_STOPPED |
@@ -388,6 +395,14 @@ int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 	vcpu->arch.sie_block->ecb2  = 8;
 	vcpu->arch.sie_block->eca   = 0xC1002001U;
 	vcpu->arch.sie_block->fac   = (int) (long) facilities;
+	if (kvm_enabled_cmma()) {
+		cbrl = alloc_page(GFP_KERNEL | __GFP_ZERO);
+		if (cbrl) {
+			vcpu->arch.sie_block->ecb2 |= 0x80;
+			vcpu->arch.sie_block->ecb2 &= ~0x08;
+			vcpu->arch.sie_block->cbrlo = page_to_phys(cbrl);
+		}
+	}
 	hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
 	tasklet_init(&vcpu->arch.tasklet, kvm_s390_tasklet,
 		     (unsigned long) vcpu);
@@ -682,6 +697,16 @@ static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+bool kvm_enabled_cmma(void)
+{
+	if (!MACHINE_IS_LPAR)
+		return false;
+	/* only enable for z10 and later */
+	if (!MACHINE_HAS_EDAT1)
+		return false;
+	return true;
+}
+
 static int __vcpu_run(struct kvm_vcpu *vcpu)
 {
 	int rc;
diff --git a/arch/s390/kvm/kvm-s390.h b/arch/s390/kvm/kvm-s390.h
index 028ca9f..ea3d5f7 100644
--- a/arch/s390/kvm/kvm-s390.h
+++ b/arch/s390/kvm/kvm-s390.h
@@ -145,6 +145,8 @@ void s390_vcpu_block(struct kvm_vcpu *vcpu);
 void s390_vcpu_unblock(struct kvm_vcpu *vcpu);
 void exit_sie(struct kvm_vcpu *vcpu);
 void exit_sie_sync(struct kvm_vcpu *vcpu);
+/* are we going to support cmma? */
+bool kvm_enabled_cmma(void);
 /* implemented in diag.c */
 int kvm_s390_handle_diag(struct kvm_vcpu *vcpu);
 
diff --git a/arch/s390/kvm/priv.c b/arch/s390/kvm/priv.c
index 0da3e6e..e0703db 100644
--- a/arch/s390/kvm/priv.c
+++ b/arch/s390/kvm/priv.c
@@ -581,9 +581,50 @@ static int handle_pfmf(struct kvm_vcpu *vcpu)
 	return 0;
 }
 
+static int handle_essa(struct kvm_vcpu *vcpu)
+{
+	/* entries expected to be 1FF */
+	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
+	unsigned long *cbrlo, cbrle;
+	struct gmap *gmap;
+	int i;
+
+	VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
+	gmap = vcpu->arch.gmap;
+	vcpu->stat.instruction_essa++;
+	if (!kvm_enabled_cmma() || !vcpu->arch.sie_block->cbrlo)
+		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
+
+	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
+		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
+
+	if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
+		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
+
+	/* Rewind PSW to repeat the ESSA instruction */
+	vcpu->arch.sie_block->gpsw.addr =
+		__rewind_psw(vcpu->arch.sie_block->gpsw, 4);
+	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
+	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
+	down_read(&gmap->mm->mmap_sem);
+	for (i = 0; i < entries; ++i) {
+		cbrle = cbrlo[i];
+		if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
+			/* invalid entry */
+			break;
+		/* try to free backing */
+		__gmap_zap(cbrle, gmap);
+	}
+	up_read(&gmap->mm->mmap_sem);
+	if (i < entries)
+		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
+	return 0;
+}
+
 static const intercept_handler_t b9_handlers[256] = {
 	[0x8d] = handle_epsw,
 	[0x9c] = handle_io_inst,
+	[0xab] = handle_essa,
 	[0xaf] = handle_pfmf,
 };
 
diff --git a/arch/s390/mm/pgtable.c b/arch/s390/mm/pgtable.c
index a8154a1..567f89e 100644
--- a/arch/s390/mm/pgtable.c
+++ b/arch/s390/mm/pgtable.c
@@ -17,6 +17,7 @@
 #include <linux/quicklist.h>
 #include <linux/rcupdate.h>
 #include <linux/slab.h>
+#include <linux/swapops.h>
 
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -573,6 +574,82 @@ unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
 }
 EXPORT_SYMBOL_GPL(gmap_fault);
 
+static void gmap_zap_swap_entry(swp_entry_t entry, struct mm_struct *mm)
+{
+	if (!non_swap_entry(entry))
+		dec_mm_counter(mm, MM_SWAPENTS);
+	else if (is_migration_entry(entry)) {
+		struct page *page = migration_entry_to_page(entry);
+
+		if (PageAnon(page))
+			dec_mm_counter(mm, MM_ANONPAGES);
+		else
+			dec_mm_counter(mm, MM_FILEPAGES);
+	}
+	free_swap_and_cache(entry);
+}
+
+/**
+ * The mm->mmap_sem lock must be held
+ */
+static void gmap_zap_unused(struct mm_struct *mm, unsigned long address)
+{
+	unsigned long ptev, pgstev;
+	spinlock_t *ptl;
+	pgste_t pgste;
+	pte_t *ptep, pte;
+
+	ptep = get_locked_pte(mm, address, &ptl);
+	if (unlikely(!ptep))
+		return;
+	pte = *ptep;
+	if (!pte_swap(pte))
+		goto out_pte;
+	/* Zap unused and logically-zero pages */
+	pgste = pgste_get_lock(ptep);
+	pgstev = pgste_val(pgste);
+	ptev = pte_val(pte);
+	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
+	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) {
+		gmap_zap_swap_entry(pte_to_swp_entry(pte), mm);
+		pte_clear(mm, address, ptep);
+	}
+	pgste_set_unlock(ptep, pgste);
+out_pte:
+	pte_unmap_unlock(*ptep, ptl);
+}
+
+/*
+ * this function is assumed to be called with mmap_sem held
+ */
+void __gmap_zap(unsigned long address, struct gmap *gmap)
+{
+	unsigned long *table, *segment_ptr;
+	unsigned long segment, pgstev, ptev;
+	struct gmap_pgtable *mp;
+	struct page *page;
+
+	segment_ptr = gmap_table_walk(address, gmap);
+	if (IS_ERR(segment_ptr))
+		return;
+	segment = *segment_ptr;
+	if (segment & _SEGMENT_ENTRY_INV)
+		return;
+	page = pfn_to_page(segment >> PAGE_SHIFT);
+	mp = (struct gmap_pgtable *) page->index;
+	address = mp->vmaddr | (address & ~PMD_MASK);
+	/* Page table is present */
+	table = (unsigned long *)(segment & _SEGMENT_ENTRY_ORIGIN);
+	table = table + ((address >> 12) & 0xff);
+	pgstev = table[PTRS_PER_PTE];
+	ptev = table[0];
+	/* quick check, checked again with locks held */
+	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
+	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID)))
+		gmap_zap_unused(gmap->mm, address);
+}
+EXPORT_SYMBOL_GPL(__gmap_zap);
+
 void gmap_discard(unsigned long from, unsigned long to, struct gmap *gmap)
 {
 
-- 
1.7.9.5

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH 2/2] s390/kvm: support collaborative memory management

[Christian Borntraeger]

On 25/07/13 10:54, Martin Schwidefsky wrote:
> From: Konstantin Weitz <konstantin.weitz@gmail.com>
> 
> This patch enables Collaborative Memory Management (CMM) for kvm
> on s390. CMM allows the guest to inform the host about page usage
> (see arch/s390/mm/cmm.c). The host uses this information to avoid
> swapping in unused pages in the page fault handler. Further, a CPU
> provided list of unused invalid pages is processed to reclaim swap
> space of not yet accessed unused pages.
> 
> [ Martin Schwidefsky: patch reordering and cleanup ]
> 
> Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Two things to consider: life migration and reset

When we implement life migration, we need to add some additional magic for
userspace to query/set unused state. But this can be a followup patch, 
whenever this becomes necessary.

As of today it should be enough to add some code to the diag308 handler to
make reset save. For other kinds of reset (e.g. those for kdump) we need
to make this accessible to userspace. Again, this can be added later on
when we implement the other missing pieces for kdump and friends.

So

Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>



> ---
>  arch/s390/include/asm/kvm_host.h |    5 ++-
>  arch/s390/include/asm/pgtable.h  |   24 ++++++++++++
>  arch/s390/kvm/kvm-s390.c         |   25 +++++++++++++
>  arch/s390/kvm/kvm-s390.h         |    2 +
>  arch/s390/kvm/priv.c             |   41 ++++++++++++++++++++
>  arch/s390/mm/pgtable.c           |   77 ++++++++++++++++++++++++++++++++++++++
>  6 files changed, 173 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
> index 3238d40..de6450e 100644
> --- a/arch/s390/include/asm/kvm_host.h
> +++ b/arch/s390/include/asm/kvm_host.h
> @@ -113,7 +113,9 @@ struct kvm_s390_sie_block {
>  	__u64	gbea;			/* 0x0180 */
>  	__u8	reserved188[24];	/* 0x0188 */
>  	__u32	fac;			/* 0x01a0 */
> -	__u8	reserved1a4[92];	/* 0x01a4 */
> +	__u8	reserved1a4[20];	/* 0x01a4 */
> +	__u64	cbrlo;			/* 0x01b8 */
> +	__u8	reserved1c0[64];	/* 0x01c0 */
>  } __attribute__((packed));
> 
>  struct kvm_vcpu_stat {
> @@ -149,6 +151,7 @@ struct kvm_vcpu_stat {
>  	u32 instruction_stsi;
>  	u32 instruction_stfl;
>  	u32 instruction_tprot;
> +	u32 instruction_essa;
>  	u32 instruction_sigp_sense;
>  	u32 instruction_sigp_sense_running;
>  	u32 instruction_sigp_external_call;
> diff --git a/arch/s390/include/asm/pgtable.h b/arch/s390/include/asm/pgtable.h
> index 75fb726..65d48b8 100644
> --- a/arch/s390/include/asm/pgtable.h
> +++ b/arch/s390/include/asm/pgtable.h
> @@ -227,6 +227,7 @@ extern unsigned long MODULES_END;
>  #define _PAGE_SWR	0x008		/* SW pte referenced bit */
>  #define _PAGE_SWW	0x010		/* SW pte write bit */
>  #define _PAGE_SPECIAL	0x020		/* SW associated with special page */
> +#define _PAGE_UNUSED	0x040		/* SW bit for ptep_clear_flush() */
>  #define __HAVE_ARCH_PTE_SPECIAL
> 
>  /* Set of bits not changed in pte_modify */
> @@ -375,6 +376,12 @@ extern unsigned long MODULES_END;
> 
>  #endif /* CONFIG_64BIT */
> 
> +/* Guest Page State used for virtualization */
> +#define _PGSTE_GPS_ZERO 	0x0000000080000000UL
> +#define _PGSTE_GPS_USAGE_MASK	0x0000000003000000UL
> +#define _PGSTE_GPS_USAGE_STABLE 0x0000000000000000UL
> +#define _PGSTE_GPS_USAGE_UNUSED 0x0000000001000000UL
> +
>  /*
>   * A user page table pointer has the space-switch-event bit, the
>   * private-space-control bit and the storage-alteration-event-control
> @@ -590,6 +597,12 @@ static inline int pte_file(pte_t pte)
>  	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
>  }
> 
> +static inline int pte_swap(pte_t pte)
> +{
> +	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
> +	return (pte_val(pte) & mask) == _PAGE_TYPE_SWAP;
> +}
> +
>  static inline int pte_special(pte_t pte)
>  {
>  	return (pte_val(pte) & _PAGE_SPECIAL);
> @@ -794,6 +807,7 @@ unsigned long gmap_translate(unsigned long address, struct gmap *);
>  unsigned long __gmap_fault(unsigned long address, struct gmap *);
>  unsigned long gmap_fault(unsigned long address, struct gmap *);
>  void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
> +void __gmap_zap(unsigned long address, struct gmap *);
> 
>  void gmap_register_ipte_notifier(struct gmap_notifier *);
>  void gmap_unregister_ipte_notifier(struct gmap_notifier *);
> @@ -825,6 +839,7 @@ static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
> 
>  	if (mm_has_pgste(mm)) {
>  		pgste = pgste_get_lock(ptep);
> +		pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
>  		pgste_set_key(ptep, pgste, entry);
>  		pgste_set_pte(ptep, entry);
>  		pgste_set_unlock(ptep, pgste);
> @@ -858,6 +873,12 @@ static inline int pte_young(pte_t pte)
>  	return 0;
>  }
> 
> +#define __HAVE_ARCH_PTE_UNUSED
> +static inline int pte_unused(pte_t pte)
> +{
> +	return pte_val(pte) & _PAGE_UNUSED;
> +}
> +
>  /*
>   * pgd/pmd/pte modification functions
>   */
> @@ -1142,6 +1163,9 @@ static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
>  	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
> 
>  	if (mm_has_pgste(vma->vm_mm)) {
> +		if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
> +		    _PGSTE_GPS_USAGE_UNUSED)
> +			pte_val(pte) |= _PAGE_UNUSED;
>  		pgste = pgste_update_all(&pte, pgste);
>  		pgste_set_unlock(ptep, pgste);
>  	}
> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
> index ba694d2..052a773 100644
> --- a/arch/s390/kvm/kvm-s390.c
> +++ b/arch/s390/kvm/kvm-s390.c
> @@ -67,6 +67,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
>  	{ "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
>  	{ "instruction_stsch", VCPU_STAT(instruction_stsch) },
>  	{ "instruction_chsc", VCPU_STAT(instruction_chsc) },
> +	{ "instruction_essa", VCPU_STAT(instruction_essa) },
>  	{ "instruction_stsi", VCPU_STAT(instruction_stsi) },
>  	{ "instruction_stfl", VCPU_STAT(instruction_stfl) },
>  	{ "instruction_tprot", VCPU_STAT(instruction_tprot) },
> @@ -276,7 +277,11 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
>  	if (kvm_is_ucontrol(vcpu->kvm))
>  		gmap_free(vcpu->arch.gmap);
> 
> +	if (vcpu->arch.sie_block->cbrlo)
> +		__free_page(__pfn_to_page(
> +				vcpu->arch.sie_block->cbrlo >> PAGE_SHIFT));
>  	free_page((unsigned long)(vcpu->arch.sie_block));
> +
>  	kvm_vcpu_uninit(vcpu);
>  	kmem_cache_free(kvm_vcpu_cache, vcpu);
>  }
> @@ -380,6 +385,8 @@ int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
> 
>  int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
>  {
> +	struct page *cbrl;
> +
>  	atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
>  						    CPUSTAT_SM |
>  						    CPUSTAT_STOPPED |
> @@ -388,6 +395,14 @@ int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
>  	vcpu->arch.sie_block->ecb2  = 8;
>  	vcpu->arch.sie_block->eca   = 0xC1002001U;
>  	vcpu->arch.sie_block->fac   = (int) (long) facilities;
> +	if (kvm_enabled_cmma()) {
> +		cbrl = alloc_page(GFP_KERNEL | __GFP_ZERO);
> +		if (cbrl) {
> +			vcpu->arch.sie_block->ecb2 |= 0x80;
> +			vcpu->arch.sie_block->ecb2 &= ~0x08;
> +			vcpu->arch.sie_block->cbrlo = page_to_phys(cbrl);
> +		}
> +	}
>  	hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
>  	tasklet_init(&vcpu->arch.tasklet, kvm_s390_tasklet,
>  		     (unsigned long) vcpu);
> @@ -682,6 +697,16 @@ static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
>  	return 0;
>  }
> 
> +bool kvm_enabled_cmma(void)
> +{
> +	if (!MACHINE_IS_LPAR)
> +		return false;
> +	/* only enable for z10 and later */
> +	if (!MACHINE_HAS_EDAT1)
> +		return false;
> +	return true;
> +}
> +
>  static int __vcpu_run(struct kvm_vcpu *vcpu)
>  {
>  	int rc;
> diff --git a/arch/s390/kvm/kvm-s390.h b/arch/s390/kvm/kvm-s390.h
> index 028ca9f..ea3d5f7 100644
> --- a/arch/s390/kvm/kvm-s390.h
> +++ b/arch/s390/kvm/kvm-s390.h
> @@ -145,6 +145,8 @@ void s390_vcpu_block(struct kvm_vcpu *vcpu);
>  void s390_vcpu_unblock(struct kvm_vcpu *vcpu);
>  void exit_sie(struct kvm_vcpu *vcpu);
>  void exit_sie_sync(struct kvm_vcpu *vcpu);
> +/* are we going to support cmma? */
> +bool kvm_enabled_cmma(void);
>  /* implemented in diag.c */
>  int kvm_s390_handle_diag(struct kvm_vcpu *vcpu);
> 
> diff --git a/arch/s390/kvm/priv.c b/arch/s390/kvm/priv.c
> index 0da3e6e..e0703db 100644
> --- a/arch/s390/kvm/priv.c
> +++ b/arch/s390/kvm/priv.c
> @@ -581,9 +581,50 @@ static int handle_pfmf(struct kvm_vcpu *vcpu)
>  	return 0;
>  }
> 
> +static int handle_essa(struct kvm_vcpu *vcpu)
> +{
> +	/* entries expected to be 1FF */
> +	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
> +	unsigned long *cbrlo, cbrle;
> +	struct gmap *gmap;
> +	int i;
> +
> +	VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
> +	gmap = vcpu->arch.gmap;
> +	vcpu->stat.instruction_essa++;
> +	if (!kvm_enabled_cmma() || !vcpu->arch.sie_block->cbrlo)
> +		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
> +
> +	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
> +		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
> +
> +	if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
> +		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
> +
> +	/* Rewind PSW to repeat the ESSA instruction */
> +	vcpu->arch.sie_block->gpsw.addr =
> +		__rewind_psw(vcpu->arch.sie_block->gpsw, 4);
> +	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
> +	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
> +	down_read(&gmap->mm->mmap_sem);
> +	for (i = 0; i < entries; ++i) {
> +		cbrle = cbrlo[i];
> +		if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
> +			/* invalid entry */
> +			break;
> +		/* try to free backing */
> +		__gmap_zap(cbrle, gmap);
> +	}
> +	up_read(&gmap->mm->mmap_sem);
> +	if (i < entries)
> +		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
> +	return 0;
> +}
> +
>  static const intercept_handler_t b9_handlers[256] = {
>  	[0x8d] = handle_epsw,
>  	[0x9c] = handle_io_inst,
> +	[0xab] = handle_essa,
>  	[0xaf] = handle_pfmf,
>  };
> 
> diff --git a/arch/s390/mm/pgtable.c b/arch/s390/mm/pgtable.c
> index a8154a1..567f89e 100644
> --- a/arch/s390/mm/pgtable.c
> +++ b/arch/s390/mm/pgtable.c
> @@ -17,6 +17,7 @@
>  #include <linux/quicklist.h>
>  #include <linux/rcupdate.h>
>  #include <linux/slab.h>
> +#include <linux/swapops.h>
> 
>  #include <asm/pgtable.h>
>  #include <asm/pgalloc.h>
> @@ -573,6 +574,82 @@ unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
>  }
>  EXPORT_SYMBOL_GPL(gmap_fault);
> 
> +static void gmap_zap_swap_entry(swp_entry_t entry, struct mm_struct *mm)
> +{
> +	if (!non_swap_entry(entry))
> +		dec_mm_counter(mm, MM_SWAPENTS);
> +	else if (is_migration_entry(entry)) {
> +		struct page *page = migration_entry_to_page(entry);
> +
> +		if (PageAnon(page))
> +			dec_mm_counter(mm, MM_ANONPAGES);
> +		else
> +			dec_mm_counter(mm, MM_FILEPAGES);
> +	}
> +	free_swap_and_cache(entry);
> +}
> +
> +/**
> + * The mm->mmap_sem lock must be held
> + */
> +static void gmap_zap_unused(struct mm_struct *mm, unsigned long address)
> +{
> +	unsigned long ptev, pgstev;
> +	spinlock_t *ptl;
> +	pgste_t pgste;
> +	pte_t *ptep, pte;
> +
> +	ptep = get_locked_pte(mm, address, &ptl);
> +	if (unlikely(!ptep))
> +		return;
> +	pte = *ptep;
> +	if (!pte_swap(pte))
> +		goto out_pte;
> +	/* Zap unused and logically-zero pages */
> +	pgste = pgste_get_lock(ptep);
> +	pgstev = pgste_val(pgste);
> +	ptev = pte_val(pte);
> +	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
> +	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) {
> +		gmap_zap_swap_entry(pte_to_swp_entry(pte), mm);
> +		pte_clear(mm, address, ptep);
> +	}
> +	pgste_set_unlock(ptep, pgste);
> +out_pte:
> +	pte_unmap_unlock(*ptep, ptl);
> +}
> +
> +/*
> + * this function is assumed to be called with mmap_sem held
> + */
> +void __gmap_zap(unsigned long address, struct gmap *gmap)
> +{
> +	unsigned long *table, *segment_ptr;
> +	unsigned long segment, pgstev, ptev;
> +	struct gmap_pgtable *mp;
> +	struct page *page;
> +
> +	segment_ptr = gmap_table_walk(address, gmap);
> +	if (IS_ERR(segment_ptr))
> +		return;
> +	segment = *segment_ptr;
> +	if (segment & _SEGMENT_ENTRY_INV)
> +		return;
> +	page = pfn_to_page(segment >> PAGE_SHIFT);
> +	mp = (struct gmap_pgtable *) page->index;
> +	address = mp->vmaddr | (address & ~PMD_MASK);
> +	/* Page table is present */
> +	table = (unsigned long *)(segment & _SEGMENT_ENTRY_ORIGIN);
> +	table = table + ((address >> 12) & 0xff);
> +	pgstev = table[PTRS_PER_PTE];
> +	ptev = table[0];
> +	/* quick check, checked again with locks held */
> +	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
> +	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID)))
> +		gmap_zap_unused(gmap->mm, address);
> +}
> +EXPORT_SYMBOL_GPL(__gmap_zap);
> +
>  void gmap_discard(unsigned long from, unsigned long to, struct gmap *gmap)
>  {
> 

Re: [PATCH 2/2] s390/kvm: support collaborative memory management

[Christian Borntraeger]

On 25/07/13 10:54, Martin Schwidefsky wrote:
> From: Konstantin Weitz <konstantin.weitz@gmail.com>
> 
> This patch enables Collaborative Memory Management (CMM) for kvm
> on s390. CMM allows the guest to inform the host about page usage
> (see arch/s390/mm/cmm.c). The host uses this information to avoid
> swapping in unused pages in the page fault handler. Further, a CPU
> provided list of unused invalid pages is processed to reclaim swap
> space of not yet accessed unused pages.
> 
> [ Martin Schwidefsky: patch reordering and cleanup ]
> 
> Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

Two things to consider: life migration and reset

When we implement life migration, we need to add some additional magic for
userspace to query/set unused state. But this can be a followup patch, 
whenever this becomes necessary.

As of today it should be enough to add some code to the diag308 handler to
make reset save. For other kinds of reset (e.g. those for kdump) we need
to make this accessible to userspace. Again, this can be added later on
when we implement the other missing pieces for kdump and friends.

So

Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>



> ---
>  arch/s390/include/asm/kvm_host.h |    5 ++-
>  arch/s390/include/asm/pgtable.h  |   24 ++++++++++++
>  arch/s390/kvm/kvm-s390.c         |   25 +++++++++++++
>  arch/s390/kvm/kvm-s390.h         |    2 +
>  arch/s390/kvm/priv.c             |   41 ++++++++++++++++++++
>  arch/s390/mm/pgtable.c           |   77 ++++++++++++++++++++++++++++++++++++++
>  6 files changed, 173 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
> index 3238d40..de6450e 100644
> --- a/arch/s390/include/asm/kvm_host.h
> +++ b/arch/s390/include/asm/kvm_host.h
> @@ -113,7 +113,9 @@ struct kvm_s390_sie_block {
>  	__u64	gbea;			/* 0x0180 */
>  	__u8	reserved188[24];	/* 0x0188 */
>  	__u32	fac;			/* 0x01a0 */
> -	__u8	reserved1a4[92];	/* 0x01a4 */
> +	__u8	reserved1a4[20];	/* 0x01a4 */
> +	__u64	cbrlo;			/* 0x01b8 */
> +	__u8	reserved1c0[64];	/* 0x01c0 */
>  } __attribute__((packed));
> 
>  struct kvm_vcpu_stat {
> @@ -149,6 +151,7 @@ struct kvm_vcpu_stat {
>  	u32 instruction_stsi;
>  	u32 instruction_stfl;
>  	u32 instruction_tprot;
> +	u32 instruction_essa;
>  	u32 instruction_sigp_sense;
>  	u32 instruction_sigp_sense_running;
>  	u32 instruction_sigp_external_call;
> diff --git a/arch/s390/include/asm/pgtable.h b/arch/s390/include/asm/pgtable.h
> index 75fb726..65d48b8 100644
> --- a/arch/s390/include/asm/pgtable.h
> +++ b/arch/s390/include/asm/pgtable.h
> @@ -227,6 +227,7 @@ extern unsigned long MODULES_END;
>  #define _PAGE_SWR	0x008		/* SW pte referenced bit */
>  #define _PAGE_SWW	0x010		/* SW pte write bit */
>  #define _PAGE_SPECIAL	0x020		/* SW associated with special page */
> +#define _PAGE_UNUSED	0x040		/* SW bit for ptep_clear_flush() */
>  #define __HAVE_ARCH_PTE_SPECIAL
> 
>  /* Set of bits not changed in pte_modify */
> @@ -375,6 +376,12 @@ extern unsigned long MODULES_END;
> 
>  #endif /* CONFIG_64BIT */
> 
> +/* Guest Page State used for virtualization */
> +#define _PGSTE_GPS_ZERO 	0x0000000080000000UL
> +#define _PGSTE_GPS_USAGE_MASK	0x0000000003000000UL
> +#define _PGSTE_GPS_USAGE_STABLE 0x0000000000000000UL
> +#define _PGSTE_GPS_USAGE_UNUSED 0x0000000001000000UL
> +
>  /*
>   * A user page table pointer has the space-switch-event bit, the
>   * private-space-control bit and the storage-alteration-event-control
> @@ -590,6 +597,12 @@ static inline int pte_file(pte_t pte)
>  	return (pte_val(pte) & mask) == _PAGE_TYPE_FILE;
>  }
> 
> +static inline int pte_swap(pte_t pte)
> +{
> +	unsigned long mask = _PAGE_RO | _PAGE_INVALID | _PAGE_SWT | _PAGE_SWX;
> +	return (pte_val(pte) & mask) == _PAGE_TYPE_SWAP;
> +}
> +
>  static inline int pte_special(pte_t pte)
>  {
>  	return (pte_val(pte) & _PAGE_SPECIAL);
> @@ -794,6 +807,7 @@ unsigned long gmap_translate(unsigned long address, struct gmap *);
>  unsigned long __gmap_fault(unsigned long address, struct gmap *);
>  unsigned long gmap_fault(unsigned long address, struct gmap *);
>  void gmap_discard(unsigned long from, unsigned long to, struct gmap *);
> +void __gmap_zap(unsigned long address, struct gmap *);
> 
>  void gmap_register_ipte_notifier(struct gmap_notifier *);
>  void gmap_unregister_ipte_notifier(struct gmap_notifier *);
> @@ -825,6 +839,7 @@ static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
> 
>  	if (mm_has_pgste(mm)) {
>  		pgste = pgste_get_lock(ptep);
> +		pgste_val(pgste) &= ~_PGSTE_GPS_ZERO;
>  		pgste_set_key(ptep, pgste, entry);
>  		pgste_set_pte(ptep, entry);
>  		pgste_set_unlock(ptep, pgste);
> @@ -858,6 +873,12 @@ static inline int pte_young(pte_t pte)
>  	return 0;
>  }
> 
> +#define __HAVE_ARCH_PTE_UNUSED
> +static inline int pte_unused(pte_t pte)
> +{
> +	return pte_val(pte) & _PAGE_UNUSED;
> +}
> +
>  /*
>   * pgd/pmd/pte modification functions
>   */
> @@ -1142,6 +1163,9 @@ static inline pte_t ptep_clear_flush(struct vm_area_struct *vma,
>  	pte_val(*ptep) = _PAGE_TYPE_EMPTY;
> 
>  	if (mm_has_pgste(vma->vm_mm)) {
> +		if ((pgste_val(pgste) & _PGSTE_GPS_USAGE_MASK) ==
> +		    _PGSTE_GPS_USAGE_UNUSED)
> +			pte_val(pte) |= _PAGE_UNUSED;
>  		pgste = pgste_update_all(&pte, pgste);
>  		pgste_set_unlock(ptep, pgste);
>  	}
> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
> index ba694d2..052a773 100644
> --- a/arch/s390/kvm/kvm-s390.c
> +++ b/arch/s390/kvm/kvm-s390.c
> @@ -67,6 +67,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
>  	{ "instruction_storage_key", VCPU_STAT(instruction_storage_key) },
>  	{ "instruction_stsch", VCPU_STAT(instruction_stsch) },
>  	{ "instruction_chsc", VCPU_STAT(instruction_chsc) },
> +	{ "instruction_essa", VCPU_STAT(instruction_essa) },
>  	{ "instruction_stsi", VCPU_STAT(instruction_stsi) },
>  	{ "instruction_stfl", VCPU_STAT(instruction_stfl) },
>  	{ "instruction_tprot", VCPU_STAT(instruction_tprot) },
> @@ -276,7 +277,11 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
>  	if (kvm_is_ucontrol(vcpu->kvm))
>  		gmap_free(vcpu->arch.gmap);
> 
> +	if (vcpu->arch.sie_block->cbrlo)
> +		__free_page(__pfn_to_page(
> +				vcpu->arch.sie_block->cbrlo >> PAGE_SHIFT));
>  	free_page((unsigned long)(vcpu->arch.sie_block));
> +
>  	kvm_vcpu_uninit(vcpu);
>  	kmem_cache_free(kvm_vcpu_cache, vcpu);
>  }
> @@ -380,6 +385,8 @@ int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
> 
>  int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
>  {
> +	struct page *cbrl;
> +
>  	atomic_set(&vcpu->arch.sie_block->cpuflags, CPUSTAT_ZARCH |
>  						    CPUSTAT_SM |
>  						    CPUSTAT_STOPPED |
> @@ -388,6 +395,14 @@ int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
>  	vcpu->arch.sie_block->ecb2  = 8;
>  	vcpu->arch.sie_block->eca   = 0xC1002001U;
>  	vcpu->arch.sie_block->fac   = (int) (long) facilities;
> +	if (kvm_enabled_cmma()) {
> +		cbrl = alloc_page(GFP_KERNEL | __GFP_ZERO);
> +		if (cbrl) {
> +			vcpu->arch.sie_block->ecb2 |= 0x80;
> +			vcpu->arch.sie_block->ecb2 &= ~0x08;
> +			vcpu->arch.sie_block->cbrlo = page_to_phys(cbrl);
> +		}
> +	}
>  	hrtimer_init(&vcpu->arch.ckc_timer, CLOCK_REALTIME, HRTIMER_MODE_ABS);
>  	tasklet_init(&vcpu->arch.tasklet, kvm_s390_tasklet,
>  		     (unsigned long) vcpu);
> @@ -682,6 +697,16 @@ static int kvm_s390_handle_requests(struct kvm_vcpu *vcpu)
>  	return 0;
>  }
> 
> +bool kvm_enabled_cmma(void)
> +{
> +	if (!MACHINE_IS_LPAR)
> +		return false;
> +	/* only enable for z10 and later */
> +	if (!MACHINE_HAS_EDAT1)
> +		return false;
> +	return true;
> +}
> +
>  static int __vcpu_run(struct kvm_vcpu *vcpu)
>  {
>  	int rc;
> diff --git a/arch/s390/kvm/kvm-s390.h b/arch/s390/kvm/kvm-s390.h
> index 028ca9f..ea3d5f7 100644
> --- a/arch/s390/kvm/kvm-s390.h
> +++ b/arch/s390/kvm/kvm-s390.h
> @@ -145,6 +145,8 @@ void s390_vcpu_block(struct kvm_vcpu *vcpu);
>  void s390_vcpu_unblock(struct kvm_vcpu *vcpu);
>  void exit_sie(struct kvm_vcpu *vcpu);
>  void exit_sie_sync(struct kvm_vcpu *vcpu);
> +/* are we going to support cmma? */
> +bool kvm_enabled_cmma(void);
>  /* implemented in diag.c */
>  int kvm_s390_handle_diag(struct kvm_vcpu *vcpu);
> 
> diff --git a/arch/s390/kvm/priv.c b/arch/s390/kvm/priv.c
> index 0da3e6e..e0703db 100644
> --- a/arch/s390/kvm/priv.c
> +++ b/arch/s390/kvm/priv.c
> @@ -581,9 +581,50 @@ static int handle_pfmf(struct kvm_vcpu *vcpu)
>  	return 0;
>  }
> 
> +static int handle_essa(struct kvm_vcpu *vcpu)
> +{
> +	/* entries expected to be 1FF */
> +	int entries = (vcpu->arch.sie_block->cbrlo & ~PAGE_MASK) >> 3;
> +	unsigned long *cbrlo, cbrle;
> +	struct gmap *gmap;
> +	int i;
> +
> +	VCPU_EVENT(vcpu, 5, "cmma release %d pages", entries);
> +	gmap = vcpu->arch.gmap;
> +	vcpu->stat.instruction_essa++;
> +	if (!kvm_enabled_cmma() || !vcpu->arch.sie_block->cbrlo)
> +		return kvm_s390_inject_program_int(vcpu, PGM_OPERATION);
> +
> +	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
> +		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
> +
> +	if (((vcpu->arch.sie_block->ipb & 0xf0000000) >> 28) > 6)
> +		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
> +
> +	/* Rewind PSW to repeat the ESSA instruction */
> +	vcpu->arch.sie_block->gpsw.addr =
> +		__rewind_psw(vcpu->arch.sie_block->gpsw, 4);
> +	vcpu->arch.sie_block->cbrlo &= PAGE_MASK;	/* reset nceo */
> +	cbrlo = phys_to_virt(vcpu->arch.sie_block->cbrlo);
> +	down_read(&gmap->mm->mmap_sem);
> +	for (i = 0; i < entries; ++i) {
> +		cbrle = cbrlo[i];
> +		if (unlikely(cbrle & ~PAGE_MASK || cbrle < 2 * PAGE_SIZE))
> +			/* invalid entry */
> +			break;
> +		/* try to free backing */
> +		__gmap_zap(cbrle, gmap);
> +	}
> +	up_read(&gmap->mm->mmap_sem);
> +	if (i < entries)
> +		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
> +	return 0;
> +}
> +
>  static const intercept_handler_t b9_handlers[256] = {
>  	[0x8d] = handle_epsw,
>  	[0x9c] = handle_io_inst,
> +	[0xab] = handle_essa,
>  	[0xaf] = handle_pfmf,
>  };
> 
> diff --git a/arch/s390/mm/pgtable.c b/arch/s390/mm/pgtable.c
> index a8154a1..567f89e 100644
> --- a/arch/s390/mm/pgtable.c
> +++ b/arch/s390/mm/pgtable.c
> @@ -17,6 +17,7 @@
>  #include <linux/quicklist.h>
>  #include <linux/rcupdate.h>
>  #include <linux/slab.h>
> +#include <linux/swapops.h>
> 
>  #include <asm/pgtable.h>
>  #include <asm/pgalloc.h>
> @@ -573,6 +574,82 @@ unsigned long gmap_fault(unsigned long address, struct gmap *gmap)
>  }
>  EXPORT_SYMBOL_GPL(gmap_fault);
> 
> +static void gmap_zap_swap_entry(swp_entry_t entry, struct mm_struct *mm)
> +{
> +	if (!non_swap_entry(entry))
> +		dec_mm_counter(mm, MM_SWAPENTS);
> +	else if (is_migration_entry(entry)) {
> +		struct page *page = migration_entry_to_page(entry);
> +
> +		if (PageAnon(page))
> +			dec_mm_counter(mm, MM_ANONPAGES);
> +		else
> +			dec_mm_counter(mm, MM_FILEPAGES);
> +	}
> +	free_swap_and_cache(entry);
> +}
> +
> +/**
> + * The mm->mmap_sem lock must be held
> + */
> +static void gmap_zap_unused(struct mm_struct *mm, unsigned long address)
> +{
> +	unsigned long ptev, pgstev;
> +	spinlock_t *ptl;
> +	pgste_t pgste;
> +	pte_t *ptep, pte;
> +
> +	ptep = get_locked_pte(mm, address, &ptl);
> +	if (unlikely(!ptep))
> +		return;
> +	pte = *ptep;
> +	if (!pte_swap(pte))
> +		goto out_pte;
> +	/* Zap unused and logically-zero pages */
> +	pgste = pgste_get_lock(ptep);
> +	pgstev = pgste_val(pgste);
> +	ptev = pte_val(pte);
> +	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
> +	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) {
> +		gmap_zap_swap_entry(pte_to_swp_entry(pte), mm);
> +		pte_clear(mm, address, ptep);
> +	}
> +	pgste_set_unlock(ptep, pgste);
> +out_pte:
> +	pte_unmap_unlock(*ptep, ptl);
> +}
> +
> +/*
> + * this function is assumed to be called with mmap_sem held
> + */
> +void __gmap_zap(unsigned long address, struct gmap *gmap)
> +{
> +	unsigned long *table, *segment_ptr;
> +	unsigned long segment, pgstev, ptev;
> +	struct gmap_pgtable *mp;
> +	struct page *page;
> +
> +	segment_ptr = gmap_table_walk(address, gmap);
> +	if (IS_ERR(segment_ptr))
> +		return;
> +	segment = *segment_ptr;
> +	if (segment & _SEGMENT_ENTRY_INV)
> +		return;
> +	page = pfn_to_page(segment >> PAGE_SHIFT);
> +	mp = (struct gmap_pgtable *) page->index;
> +	address = mp->vmaddr | (address & ~PMD_MASK);
> +	/* Page table is present */
> +	table = (unsigned long *)(segment & _SEGMENT_ENTRY_ORIGIN);
> +	table = table + ((address >> 12) & 0xff);
> +	pgstev = table[PTRS_PER_PTE];
> +	ptev = table[0];
> +	/* quick check, checked again with locks held */
> +	if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) ||
> +	    ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID)))
> +		gmap_zap_unused(gmap->mm, address);
> +}
> +EXPORT_SYMBOL_GPL(__gmap_zap);
> +
>  void gmap_discard(unsigned long from, unsigned long to, struct gmap *gmap)
>  {
> 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [RFC][PATCH 0/2] s390/kvm: add kvm support for guest page hinting v2

[Christian Borntraeger]

On 25/07/13 10:54, Martin Schwidefsky wrote:
> v1->v2:
>  - found a way to simplify the common code patch
> 
> Linux on s390 as a guest under z/VM has been using the guest page
> hinting interface (alias collaborative memory management) for a long
> time. The full version with volatile states has been deemed to be too
> complicated (see the old discussion about guest page hinting e.g. on
> http://marc.info/?l=linux-mm&m=123816662017742&w=2).
> What is currently implemented for the guest is the unused and stable
> states to mark unallocated pages as freely available to the host.
> This works just fine with z/VM as the host.
> 
> The two patches in this series implement the guest page hinting
> interface for the unused and stable states in the KVM host.
> Most of the code specific to s390 but there is a common memory
> management part as well, see patch #1.
> 
> The code is working stable now, from my point of view this is ready
> for prime-time.
> 
> Konstantin Weitz (2):
>   mm: add support for discard of unused ptes
>   s390/kvm: support collaborative memory management

Can you also add the patch from our tree that reset the usage states
on reboot (diag 308 subcode 3 and 4)?

Re: [RFC][PATCH 0/2] s390/kvm: add kvm support for guest page hinting v2

[Christian Borntraeger]

On 25/07/13 10:54, Martin Schwidefsky wrote:
> v1->v2:
>  - found a way to simplify the common code patch
> 
> Linux on s390 as a guest under z/VM has been using the guest page
> hinting interface (alias collaborative memory management) for a long
> time. The full version with volatile states has been deemed to be too
> complicated (see the old discussion about guest page hinting e.g. on
> http://marc.info/?l=linux-mm&m=123816662017742&w=2).
> What is currently implemented for the guest is the unused and stable
> states to mark unallocated pages as freely available to the host.
> This works just fine with z/VM as the host.
> 
> The two patches in this series implement the guest page hinting
> interface for the unused and stable states in the KVM host.
> Most of the code specific to s390 but there is a common memory
> management part as well, see patch #1.
> 
> The code is working stable now, from my point of view this is ready
> for prime-time.
> 
> Konstantin Weitz (2):
>   mm: add support for discard of unused ptes
>   s390/kvm: support collaborative memory management

Can you also add the patch from our tree that reset the usage states
on reboot (diag 308 subcode 3 and 4)?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

[RFC][PATCH 0/2] s390/kvm: add kvm support for guest page hinting

[Martin Schwidefsky]

Linux on s390 as a guest under z/VM has been using the guest page
hinting interface (alias collaborative memory management) for a long
time. The full version with volatile states has been deemed to be too
complicated (see the old discussion about guest page hinting e.g. on
http://marc.info/?l=linux-mm&m=123816662017742&w=2).
What is currently implemented for the guest is the unused and stable
states to mark unallocated pages as freely available to the host.
This works just fine with z/VM as the host.

The two patches in this series implement the guest page hinting
interface for the unused and stable states in the KVM host.
Most of the code specific to s390 but there is a common memory
management part as well, see patch #1.

The circus is back ;-)

Konstantin Weitz (2):
  mm: add support for discard of unused ptes
  s390/kvm: support collaborative memory management

 arch/s390/include/asm/kvm_host.h |    8 +++-
 arch/s390/include/asm/pgtable.h  |   24 ++++++++++++
 arch/s390/kvm/kvm-s390.c         |   24 ++++++++++++
 arch/s390/kvm/kvm-s390.h         |    2 +
 arch/s390/kvm/priv.c             |   37 ++++++++++++++++++
 arch/s390/mm/pgtable.c           |   77 ++++++++++++++++++++++++++++++++++++++
 include/asm-generic/pgtable.h    |   13 +++++++
 include/linux/rmap.h             |    1 +
 mm/rmap.c                        |   28 +++++++++++++-
 mm/vmscan.c                      |    3 ++
 10 files changed, 214 insertions(+), 3 deletions(-)

-- 
1.7.9.5

[PATCH 1/2] mm: add support for discard of unused ptes

[Martin Schwidefsky]

From: Konstantin Weitz <konstantin.weitz@gmail.com>

In a virtualized environment and given an appropriate interface the guest
can mark pages as unused while they are free (for the s390 implementation
see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
the unused state is a property of the pte.

This patch adds the primitive 'pte_unused' and code to the host swap out
handler so that pages marked as unused by all mappers are not swapped out
but discarded instead, thus saving one IO for swap out and potentially
another one for swap in.

[ Martin Schwidefsky: patch reordering and cleanup ]

Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
---
 include/asm-generic/pgtable.h |   13 +++++++++++++
 include/linux/rmap.h          |    1 +
 mm/rmap.c                     |   28 +++++++++++++++++++++++++++-
 mm/vmscan.c                   |    3 +++
 4 files changed, 44 insertions(+), 1 deletion(-)

diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
index b183698..aae349a 100644
--- a/include/asm-generic/pgtable.h
+++ b/include/asm-generic/pgtable.h
@@ -192,6 +192,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
 }
 #endif
 
+#ifndef __HAVE_ARCH_PTE_UNUSED
+/*
+ * Some architectures provide facilities to virtualization guests
+ * so that they can flag allocated pages as unused. This allows the
+ * host to transparently reclaim unused pages. This function returns
+ * whether the pte's page is unused.
+ */
+static inline int pte_unused(pte_t pte)
+{
+	return 0;
+}
+#endif
+
 #ifndef __HAVE_ARCH_PMD_SAME
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
diff --git a/include/linux/rmap.h b/include/linux/rmap.h
index 6dacb93..915e5c6 100644
--- a/include/linux/rmap.h
+++ b/include/linux/rmap.h
@@ -272,5 +272,6 @@ static inline int page_mkclean(struct page *page)
 #define SWAP_AGAIN	1
 #define SWAP_FAIL	2
 #define SWAP_MLOCK	3
+#define SWAP_FREE	4
 
 #endif	/* _LINUX_RMAP_H */
diff --git a/mm/rmap.c b/mm/rmap.c
index 6280da8..be2788d 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1233,6 +1233,10 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 		}
 		set_pte_at(mm, address, pte,
 			   swp_entry_to_pte(make_hwpoison_entry(page)));
+	} else if (pte_unused(pteval) && PageAnon(page)) {
+		pte_clear(mm, address, pte);
+		dec_mm_counter(mm, MM_ANONPAGES);
+		ret = SWAP_FREE;
 	} else if (PageAnon(page)) {
 		swp_entry_t entry = { .val = page_private(page) };
 
@@ -1455,6 +1459,7 @@ static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
 	pgoff_t pgoff;
 	struct anon_vma_chain *avc;
 	int ret = SWAP_AGAIN;
+	int used = 0;
 
 	anon_vma = page_lock_anon_vma_read(page);
 	if (!anon_vma)
@@ -1479,10 +1484,31 @@ static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
 
 		address = vma_address(page, vma);
 		ret = try_to_unmap_one(page, vma, address, flags);
+
+		/*
+		 * If SWAP_FREE was returned, we know that the page
+		 * is not used (as indicated by pte_unused()) by this
+		 * mapper. If only one of the mappers used the page,
+		 * it is considered used.
+		 */
+		if (ret == SWAP_FREE)
+			ret = SWAP_AGAIN;
+		else
+			used = 1;
+
 		if (ret != SWAP_AGAIN || !page_mapped(page))
 			break;
 	}
 
+	/*
+	 * If none of the mappers use the page, clear the dirty bit
+	 * so that the caller of try_to_unmap_anon() will free its mapping.
+	 */
+	if (!used && page_swapcount(page) == 0) {
+		ClearPageDirty(page);
+		ret = SWAP_FREE;
+	}
+
 	page_unlock_anon_vma_read(anon_vma);
 	return ret;
 }
@@ -1625,7 +1651,7 @@ int try_to_unmap(struct page *page, enum ttu_flags flags)
 		ret = try_to_unmap_anon(page, flags);
 	else
 		ret = try_to_unmap_file(page, flags);
-	if (ret != SWAP_MLOCK && !page_mapped(page))
+	if (ret != SWAP_FREE && ret != SWAP_MLOCK && !page_mapped(page))
 		ret = SWAP_SUCCESS;
 	return ret;
 }
diff --git a/mm/vmscan.c b/mm/vmscan.c
index fa6a853..093c1d7 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -800,6 +800,9 @@ static unsigned long shrink_page_list(struct list_head *page_list,
 				goto keep_locked;
 			case SWAP_MLOCK:
 				goto cull_mlocked;
+			case SWAP_FREE:
+				if (PageSwapCache(page))
+					try_to_free_swap(page);
 			case SWAP_SUCCESS:
 				; /* try to free the page below */
 			}
-- 
1.7.9.5

[PATCH 1/2] mm: add support for discard of unused ptes

[Martin Schwidefsky]

From: Konstantin Weitz <konstantin.weitz@gmail.com>

In a virtualized environment and given an appropriate interface the guest
can mark pages as unused while they are free (for the s390 implementation
see git commit 45e576b1c3d00206 "guest page hinting light"). For the host
the unused state is a property of the pte.

This patch adds the primitive 'pte_unused' and code to the host swap out
handler so that pages marked as unused by all mappers are not swapped out
but discarded instead, thus saving one IO for swap out and potentially
another one for swap in.

[ Martin Schwidefsky: patch reordering and cleanup ]

Signed-off-by: Konstantin Weitz <konstantin.weitz@gmail.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
---
 include/asm-generic/pgtable.h |   13 +++++++++++++
 include/linux/rmap.h          |    1 +
 mm/rmap.c                     |   28 +++++++++++++++++++++++++++-
 mm/vmscan.c                   |    3 +++
 4 files changed, 44 insertions(+), 1 deletion(-)

diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
index b183698..aae349a 100644
--- a/include/asm-generic/pgtable.h
+++ b/include/asm-generic/pgtable.h
@@ -192,6 +192,19 @@ static inline int pte_same(pte_t pte_a, pte_t pte_b)
 }
 #endif
 
+#ifndef __HAVE_ARCH_PTE_UNUSED
+/*
+ * Some architectures provide facilities to virtualization guests
+ * so that they can flag allocated pages as unused. This allows the
+ * host to transparently reclaim unused pages. This function returns
+ * whether the pte's page is unused.
+ */
+static inline int pte_unused(pte_t pte)
+{
+	return 0;
+}
+#endif
+
 #ifndef __HAVE_ARCH_PMD_SAME
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
diff --git a/include/linux/rmap.h b/include/linux/rmap.h
index 6dacb93..915e5c6 100644
--- a/include/linux/rmap.h
+++ b/include/linux/rmap.h
@@ -272,5 +272,6 @@ static inline int page_mkclean(struct page *page)
 #define SWAP_AGAIN	1
 #define SWAP_FAIL	2
 #define SWAP_MLOCK	3
+#define SWAP_FREE	4
 
 #endif	/* _LINUX_RMAP_H */
diff --git a/mm/rmap.c b/mm/rmap.c
index 6280da8..be2788d 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1233,6 +1233,10 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 		}
 		set_pte_at(mm, address, pte,
 			   swp_entry_to_pte(make_hwpoison_entry(page)));
+	} else if (pte_unused(pteval) && PageAnon(page)) {
+		pte_clear(mm, address, pte);
+		dec_mm_counter(mm, MM_ANONPAGES);
+		ret = SWAP_FREE;
 	} else if (PageAnon(page)) {
 		swp_entry_t entry = { .val = page_private(page) };
 
@@ -1455,6 +1459,7 @@ static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
 	pgoff_t pgoff;
 	struct anon_vma_chain *avc;
 	int ret = SWAP_AGAIN;
+	int used = 0;
 
 	anon_vma = page_lock_anon_vma_read(page);
 	if (!anon_vma)
@@ -1479,10 +1484,31 @@ static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
 
 		address = vma_address(page, vma);
 		ret = try_to_unmap_one(page, vma, address, flags);
+
+		/*
+		 * If SWAP_FREE was returned, we know that the page
+		 * is not used (as indicated by pte_unused()) by this
+		 * mapper. If only one of the mappers used the page,
+		 * it is considered used.
+		 */
+		if (ret == SWAP_FREE)
+			ret = SWAP_AGAIN;
+		else
+			used = 1;
+
 		if (ret != SWAP_AGAIN || !page_mapped(page))
 			break;
 	}
 
+	/*
+	 * If none of the mappers use the page, clear the dirty bit
+	 * so that the caller of try_to_unmap_anon() will free its mapping.
+	 */
+	if (!used && page_swapcount(page) == 0) {
+		ClearPageDirty(page);
+		ret = SWAP_FREE;
+	}
+
 	page_unlock_anon_vma_read(anon_vma);
 	return ret;
 }
@@ -1625,7 +1651,7 @@ int try_to_unmap(struct page *page, enum ttu_flags flags)
 		ret = try_to_unmap_anon(page, flags);
 	else
 		ret = try_to_unmap_file(page, flags);
-	if (ret != SWAP_MLOCK && !page_mapped(page))
+	if (ret != SWAP_FREE && ret != SWAP_MLOCK && !page_mapped(page))
 		ret = SWAP_SUCCESS;
 	return ret;
 }
diff --git a/mm/vmscan.c b/mm/vmscan.c
index fa6a853..093c1d7 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -800,6 +800,9 @@ static unsigned long shrink_page_list(struct list_head *page_list,
 				goto keep_locked;
 			case SWAP_MLOCK:
 				goto cull_mlocked;
+			case SWAP_FREE:
+				if (PageSwapCache(page))
+					try_to_free_swap(page);
 			case SWAP_SUCCESS:
 				; /* try to free the page below */
 			}
-- 
1.7.9.5

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>