2008-06-09 16:19:47

by Russ Anderson

[permalink] [raw]
Subject: [PATCH 2/3] mm: Avoid putting a bad page back on the LRU v6

Prevent a page with a physical memory error from being placed back
on the LRU. A new page flag (PG_memerror) is added if
CONFIG_PAGEFLAGS_EXTENDED is defined.

If the migration of data off the old page fails, PG_memerror is
cleared and the page placed back on the LRU. If the page gets
another correctable error another attempt will be made to
migrate it.


Signed-off-by: Russ Anderson <[email protected]>

---
include/linux/page-flags.h | 11 ++++++++++-
mm/migrate.c | 31 ++++++++++++++++++++++++++++++-
mm/page_alloc.c | 13 ++++++++-----
3 files changed, 48 insertions(+), 7 deletions(-)

Index: linux-next/mm/page_alloc.c
===================================================================
--- linux-next.orig/mm/page_alloc.c 2008-06-09 09:26:52.203578218 -0500
+++ linux-next/mm/page_alloc.c 2008-06-09 09:26:54.715891226 -0500
@@ -602,10 +602,10 @@ static int prep_new_page(struct page *pa
bad_page(page);

/*
- * For now, we report if PG_reserved was found set, but do not
- * clear it, and do not allocate the page: as a safety net.
+ * For now, we report if PG_reserved or PG_memerror was found set, but
+ * do not clear it, and do not allocate the page: as a safety net.
*/
- if (PageReserved(page))
+ if (PageReserved(page) || PageMemError(page))
return 1;

page->flags &= ~(1 << PG_uptodate | 1 << PG_error | 1 << PG_reclaim |
@@ -2476,8 +2476,11 @@ static void setup_zone_migrate_reserve(s
continue;
page = pfn_to_page(pfn);

- /* Blocks with reserved pages will never free, skip them. */
- if (PageReserved(page))
+ /*
+ * Blocks with reserved pages or memory errors will never
+ * free, skip them.
+ */
+ if (PageReserved(page) || PageMemError(page))
continue;

block_migratetype = get_pageblock_migratetype(page);
Index: linux-next/mm/migrate.c
===================================================================
--- linux-next.orig/mm/migrate.c 2008-06-09 09:26:46.226833574 -0500
+++ linux-next/mm/migrate.c 2008-06-09 09:26:54.739894217 -0500
@@ -720,7 +720,25 @@ unlock:
* restored.
*/
list_del(&page->lru);
- move_to_lru(page);
+ if (PageMemError(page))
+ if (rc == 0)
+ /*
+ * A page with a memory error that has
+ * been migrated will not be moved to
+ * the LRU.
+ */
+ goto move_newpage;
+ else
+ /*
+ * The page failed to migrate and will not
+ * be added to the bad page list. Clearing
+ * the error bit will allow another attempt
+ * to migrate if it gets another correctable
+ * error.
+ */
+ ClearPageMemError(page);
+
+ move_to_lru(page);
}

move_newpage:
@@ -790,6 +808,17 @@ int migrate_pages(struct list_head *from
}
}
}
+
+ if (rc != 0)
+ list_for_each_entry_safe(page, page2, from, lru)
+ if (PageMemError(page))
+ /*
+ * The page failed to migrate. Clearing
+ * the error bit will allow another attempt
+ * to migrate if it gets another correctable
+ * error.
+ */
+ ClearPageMemError(page);
rc = 0;
out:
if (!swapwrite)
Index: linux-next/include/linux/page-flags.h
===================================================================
--- linux-next.orig/include/linux/page-flags.h 2008-06-09 09:26:52.223580710 -0500
+++ linux-next/include/linux/page-flags.h 2008-06-09 09:26:54.763897207 -0500
@@ -84,6 +84,7 @@ enum pageflags {
PG_private, /* If pagecache, has fs-private data */
PG_writeback, /* Page is under writeback */
#ifdef CONFIG_PAGEFLAGS_EXTENDED
+ PG_memerror, /* Page has a physical memory error */
PG_head, /* A head page */
PG_tail, /* A tail page */
#else
@@ -308,9 +309,17 @@ static inline void __ClearPageTail(struc

#endif /* !PAGEFLAGS_EXTENDED */

-#define PAGE_FLAGS (1 << PG_lru | 1 << PG_private | 1 << PG_locked | \
+#define PAGE_FLAGS_BASE (1 << PG_lru | 1 << PG_private | 1 << PG_locked | \
1 << PG_buddy | 1 << PG_writeback | \
1 << PG_slab | 1 << PG_swapcache | 1 << PG_active)
+
+#ifdef CONFIG_PAGEFLAGS_EXTENDED
+PAGEFLAG(MemError, memerror)
+#define PAGE_FLAGS (PAGE_FLAGS_BASE | 1UL << PG_memerror)
+#else
+PAGEFLAG_FALSE(MemError)
+#define PAGE_FLAGS (PAGE_FLAGS_BASE)
+#endif

/*
* Flags checked in bad_page(). Pages on the free list should not have
--
Russ Anderson, OS RAS/Partitioning Project Lead
SGI - Silicon Graphics Inc [email protected]


2008-06-09 18:28:11

by Christoph Lameter

[permalink] [raw]
Subject: Re: [PATCH 2/3] mm: Avoid putting a bad page back on the LRU v6

On Mon, 9 Jun 2008, Russ Anderson wrote:

> -#define PAGE_FLAGS (1 << PG_lru | 1 << PG_private | 1 << PG_locked | \
> +#define PAGE_FLAGS_BASE (1 << PG_lru | 1 << PG_private | 1 << PG_locked | \
> 1 << PG_buddy | 1 << PG_writeback | \
> 1 << PG_slab | 1 << PG_swapcache | 1 << PG_active)
> +
> +#ifdef CONFIG_PAGEFLAGS_EXTENDED
> +PAGEFLAG(MemError, memerror)
> +#define PAGE_FLAGS (PAGE_FLAGS_BASE | 1UL << PG_memerror)
> +#else
> +PAGEFLAG_FALSE(MemError)
> +#define PAGE_FLAGS (PAGE_FLAGS_BASE)
> +#endif

Hmmmm... That doesnt look nice. It would be good if we had some definition
that allows the simple oring with a flag mask. That mask should be
zero if the flag is always false.

#ifdef CONFIG_PAGEFLAGS_EXTENDED
PAGEFLAG(MemError, memerror)
#define MemError_mask (1 << PG_memerror)
#else
PAGEFLAG_FALSE(MemError)
#define MemError_mask 0
#endif

#define PAGE_FLAGS (PAGE_FLAGS_BASE | MemError_mask)

Maybe we could have the PAGEFLAG and the PAGEFLAG_FALSE macros generate
these definitions?

2008-06-09 21:03:56

by Russ Anderson

[permalink] [raw]
Subject: Re: [PATCH 2/3] mm: Avoid putting a bad page back on the LRU v6

On Mon, Jun 09, 2008 at 11:28:00AM -0700, Christoph Lameter wrote:
> On Mon, 9 Jun 2008, Russ Anderson wrote:
>
> > -#define PAGE_FLAGS (1 << PG_lru | 1 << PG_private | 1 << PG_locked | \
> > +#define PAGE_FLAGS_BASE (1 << PG_lru | 1 << PG_private | 1 << PG_locked | \
> > 1 << PG_buddy | 1 << PG_writeback | \
> > 1 << PG_slab | 1 << PG_swapcache | 1 << PG_active)
> > +
> > +#ifdef CONFIG_PAGEFLAGS_EXTENDED
> > +PAGEFLAG(MemError, memerror)
> > +#define PAGE_FLAGS (PAGE_FLAGS_BASE | 1UL << PG_memerror)
> > +#else
> > +PAGEFLAG_FALSE(MemError)
> > +#define PAGE_FLAGS (PAGE_FLAGS_BASE)
> > +#endif
>
> Hmmmm... That doesnt look nice. It would be good if we had some definition
> that allows the simple oring with a flag mask. That mask should be
> zero if the flag is always false.
>
> #ifdef CONFIG_PAGEFLAGS_EXTENDED
> PAGEFLAG(MemError, memerror)
> #define MemError_mask (1 << PG_memerror)
> #else
> PAGEFLAG_FALSE(MemError)
> #define MemError_mask 0
> #endif
>
> #define PAGE_FLAGS (PAGE_FLAGS_BASE | MemError_mask)
>
> Maybe we could have the PAGEFLAG and the PAGEFLAG_FALSE macros generate
> these definitions?

Does this look better?


Index: linus/include/linux/page-flags.h
===================================================================
--- linus.orig/include/linux/page-flags.h 2008-06-09 13:48:10.072171477 -0500
+++ linus/include/linux/page-flags.h 2008-06-09 15:58:27.957729371 -0500
@@ -84,6 +84,7 @@ enum pageflags {
PG_private, /* If pagecache, has fs-private data */
PG_writeback, /* Page is under writeback */
#ifdef CONFIG_PAGEFLAGS_EXTENDED
+ PG_memerror, /* Page has a physical memory error */
PG_head, /* A head page */
PG_tail, /* A tail page */
#else
@@ -132,15 +133,21 @@ static inline int TestSetPage##uname(str
static inline int TestClearPage##uname(struct page *page) \
{ return test_and_clear_bit(PG_##lname, &page->flags); }

+#define PAGEFLAGMASK(uname, lname) \
+static inline int PAGEMASK_##uname(void) \
+ { return (1 << PG_##lname); }

#define PAGEFLAG(uname, lname) TESTPAGEFLAG(uname, lname) \
- SETPAGEFLAG(uname, lname) CLEARPAGEFLAG(uname, lname)
+ SETPAGEFLAG(uname, lname) CLEARPAGEFLAG(uname, lname) \
+ PAGEFLAGMASK(uname, lname)

#define __PAGEFLAG(uname, lname) TESTPAGEFLAG(uname, lname) \
__SETPAGEFLAG(uname, lname) __CLEARPAGEFLAG(uname, lname)

#define PAGEFLAG_FALSE(uname) \
static inline int Page##uname(struct page *page) \
+ { return 0; } \
+static inline int PAGEMASK_##uname(void) \
{ return 0; }

#define TESTSCFLAG(uname, lname) \
@@ -307,9 +314,16 @@ static inline void __ClearPageTail(struc

#endif /* !PAGEFLAGS_EXTENDED */

+#ifdef CONFIG_PAGEFLAGS_EXTENDED
+PAGEFLAG(MemError, memerror)
+#else
+PAGEFLAG_FALSE(MemError)
+#endif
+
#define PAGE_FLAGS (1 << PG_lru | 1 << PG_private | 1 << PG_locked | \
1 << PG_buddy | 1 << PG_writeback | \
- 1 << PG_slab | 1 << PG_swapcache | 1 << PG_active)
+ 1 << PG_slab | 1 << PG_swapcache | 1 << PG_active) |\
+ PAGEMASK_MemError()

/*
* Flags checked in bad_page(). Pages on the free list should not have

--
Russ Anderson, OS RAS/Partitioning Project Lead
SGI - Silicon Graphics Inc [email protected]

2008-06-09 21:15:41

by Christoph Lameter

[permalink] [raw]
Subject: Re: [PATCH 2/3] mm: Avoid putting a bad page back on the LRU v6

On Mon, 9 Jun 2008, Russ Anderson wrote:

> Does this look better?

Excellent.

Reviewed-by: Christoph Lameter <[email protected]>