mm: Avoid putting a bad page back on the LRU v5

Prevent a page with a physical memory error from being placed back on the LRU. A new page flag (PG_memerror) is added on 64 bit architectures.

---
 include/linux/migrate.h    |    1 +
 include/linux/page-flags.h |   11 ++++++++++-
 mm/migrate.c               |   12 ++++++++++--
 mm/page_alloc.c            |    8 ++++----
 4 files changed, 25 insertions(+), 7 deletions(-)

Index: linus/mm/page_alloc.c
=================================--- linus.orig/mm/page_alloc.c 2008-05-15 12:59:03.334381011 -0500
+++ linus/mm/page_alloc.c 2008-05-15 12:59:04.958581070 -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 |
@@ -2474,7 +2474,7 @@ static void setup_zone_migrate_reserve(s
page = pfn_to_page(pfn);

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

block_migratetype = get_pageblock_migratetype(page);
Index: linus/mm/migrate.c
=================================--- linus.orig/mm/migrate.c 2008-05-15 12:58:25.165679802 -0500
+++ linus/mm/migrate.c 2008-05-16 12:25:16.145557095 -0500
@@ -35,6 +35,8 @@

#define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru))

+unsigned int totalbad_pages;
+
/*
* Isolate one page from the LRU lists. If successful put it onto
* the indicated list with elevated page count.
@@ -110,6 +112,8 @@ int putback_lru_pages(struct list_head *

list_for_each_entry_safe(page, page2, l, lru) {
list_del(&page->lru);
+ if (unlikely(PageMemError(page)))
+ continue;
move_to_lru(page);
count++;
}
@@ -717,10 +721,14 @@ unlock:
* A page that has been migrated has all references
* removed and will be freed. A page that has not been
* migrated will have kepts its references and be
- * restored.
+ * restored. A page with a memory error will not
+ * be moved to the LRU.
*/
list_del(&page->lru);
- move_to_lru(page);
+ if (PageMemError(page))
+ totalbad_pages++;
+ else
+ move_to_lru(page);
}

move_newpage:
Index: linus/include/linux/page-flags.h
=================================--- linus.orig/include/linux/page-flags.h 2008-05-15 12:59:03.362384460 -0500
+++ linus/include/linux/page-flags.h 2008-05-15 12:59:05.002586491 -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
@@ -307,9 +308,17 @@ 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
Index: linus/include/linux/migrate.h
=================================--- linus.orig/include/linux/migrate.h 2008-05-15 12:58:25.165679802 -0500
+++ linus/include/linux/migrate.h 2008-05-15 12:59:05.026589447 -0500
@@ -38,6 +38,7 @@ extern int migrate_prep(void);
extern int migrate_vmas(struct mm_struct *mm,
const nodemask_t *from, const nodemask_t *to, unsigned long flags);
+extern unsigned int totalbad_pages;
#else
static inline int vma_migratable(struct vm_area_struct *vma)
{ return 0; }

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

Not 64 bit. Its added if CONFIG_PAGEFLAGS_EXTENDED is defined. That is the case when we are not in a configuration that uses lots ofpageflags for other purposes (like regular sparsemem).

That was a problem. If migration failed the page would end up with an extra page reference. The new patch fixes that problem. If the page fails to migrate it will be put back on the LRU and the MemError bit cleared. If the page gets another corrected error it will try to migrate the page again.

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