Skip to contentSkip to conversation

page migration: ifdef out code

PATCH New Christoph Lameter 1 participant 5 messages May 9, 2006

Subsystems (8)
lore
Share & tools
lore
#0 lore raw

PATCH 1/5 page migration: ifdef out code

ifdef around migration codeClean up various minor things.

Put #ifdef CONFIG_MIGRATION around two locations that would generate code for the non migration case.

mm/mprotect.c +2 -0
Index: linux-2.6.17-rc3-mm1/mm/mprotect.c
===================================================================
@@ -45,6 +45,7 @@ static void change_pte_range(struct mm_s
 			ptent = pte_modify(ptep_get_and_clear(mm, addr, pte), newprot);
 			set_pte_at(mm, addr, pte, ptent);
 			lazy_mmu_prot_update(ptent);
+#ifdef CONFIG_MIGRATION
 		} else if (!pte_file(oldpte)) {
 			swp_entry_t entry = pte_to_swp_entry(oldpte);
 
@@ -57,6 +58,7 @@ static void change_pte_range(struct mm_s
 				set_pte_at(mm, addr, pte,
 					swp_entry_to_pte(entry));
 			}
+#endif
 		}
 
 	} while (pte++, addr += PAGE_SIZE, addr != end);
mm/rmap.c +9 -4
Index: linux-2.6.17-rc3-mm1/mm/rmap.c
===================================================================
@@ -596,6 +596,7 @@ static int try_to_unmap_one(struct page 
 				spin_unlock(&mmlist_lock);
 			}
 			dec_mm_counter(mm, anon_rss);
+#ifdef CONFIG_MIGRATION
 		} else {
 			/*
 			 * Store the pfn of the page in a special migration
@@ -604,17 +605,21 @@ static int try_to_unmap_one(struct page 
 			 */
 			BUG_ON(!migration);
 			entry = make_migration_entry(page, pte_write(pteval));
+#endif
 		}
 		set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
 		BUG_ON(pte_file(*pte));
-	} else if (!migration)
-		dec_mm_counter(mm, file_rss);
-	else {
+	} else
+#ifdef CONFIG_MIGRATION
+	if (migration) {
 		/* Establish migration entry for a file page */
 		swp_entry_t entry;
 		entry = make_migration_entry(page, pte_write(pteval));
 		set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
-	}
+	} else
+#endif
+		dec_mm_counter(mm, file_rss);
+
 
 	page_remove_rmap(page);
 	page_cache_release(page);

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

#1 lore raw

PATCH 2/5 page migration: Update comments

Fix comments

Edit comments to be nicer better and reflect the current state of page migration.

Remove useless BUG_ON() in migrate_entry_wait() since migration_entry_to_page() already does a BUG_ON().

include/linux/swap.h +1 -1
Index: linux-2.6.17-rc3-mm1/include/linux/swap.h
===================================================================
@@ -32,7 +32,7 @@ static inline int current_is_kswapd(void
 #ifndef CONFIG_MIGRATION
 #define MAX_SWAPFILES		(1 << MAX_SWAPFILES_SHIFT)
 #else
-/* Use last entry for page migration swap entries */
+/* Use last two entries for page migration swap entries */
 #define MAX_SWAPFILES		((1 << MAX_SWAPFILES_SHIFT)-2)
 #define SWP_MIGRATION_READ	MAX_SWAPFILES
 #define SWP_MIGRATION_WRITE	(MAX_SWAPFILES + 1)
mm/migrate.c +14 -18
Index: linux-2.6.17-rc3-mm1/mm/migrate.c
===================================================================
@@ -121,8 +121,7 @@ static inline int is_swap_pte(pte_t pte)
 }
 
 /*
- * Restore a potential migration pte to a working pte entry for
- * anonymous pages.
+ * Restore a potential migration pte to a working pte entry
  */
 static void remove_migration_pte(struct vm_area_struct *vma, unsigned long addr,
 		struct page *old, struct page *new)
@@ -181,9 +180,8 @@ out:
 }
 
 /*
- * Note that remove_file_migration_ptes will only work on regular mappings
- * specialized other mappings will simply be unmapped and do not use
- * migration entries.
+ * Note that remove_file_migration_ptes will only work on regular mappings,
+ * Nonlinear mappings do not use migration entries.
  */
 static void remove_file_migration_ptes(struct page *old, struct page *new)
 {
@@ -269,9 +267,6 @@ void migration_entry_wait(struct mm_stru
 
 	page = migration_entry_to_page(entry);
 
-	/* Pages with migration entries are always locked */
-	BUG_ON(!PageLocked(page));
-
 	get_page(page);
 	pte_unmap_unlock(ptep, ptl);
 	wait_on_page_locked(page);
@@ -282,7 +277,7 @@ out:
 }
 
 /*
- * Remove or replace the page in the mapping.
+ * Replace the page in the mapping.
  *
  * The number of remaining references must be:
  * 1 for anonymous pages without a mapping
@@ -466,19 +461,20 @@ int buffer_migrate_page(struct address_s
 }
 EXPORT_SYMBOL(buffer_migrate_page);
 
+/*
+ * Default handling if a filesystem does not provide a migration function.
+ */
 static int fallback_migrate_page(struct address_space *mapping,
 	struct page *newpage, struct page *page)
 {
-	/*
-	 * Default handling if a filesystem does not provide
-	 * a migration function. We can only migrate clean
-	 * pages so try to write out any dirty pages first.
-	 */
 	if (PageDirty(page)) {
 		/*
-		 * Remove the migration entries because pageout() may
-		 * unlock which may result in migration entries pointing
-		 * to unlocked pages.
+		 * A dirty page may imply that the underlying filesystem has
+		 * the page on some queue. So the page must be clean for
+		 * migration. Writeout may mean we loose the lock and the
+		 * page state is no longer what we checked for earlier.
+		 * At this point we know that the migration attempt cannot
+		 * be successful.
 		 */
 		remove_migration_ptes(page, page);
 
@@ -490,7 +486,7 @@ static int fallback_migrate_page(struct 
 	}
 
 	/*
-	 * Buffers are managed in a filesystem specific way.
+	 * Buffers may be managed in a filesystem specific way.
 	 * We must have no buffers or drop them.
 	 */
 	if (page_has_buffers(page) &&
Documentation/vm/page_migration +33 -58

131 lines not on this page · raw message

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

#2 lore raw

PATCH 3/5 page migration: Remove useless mapping checks

In reply to Christoph Lameter

Remove another check for mapping

Page migration still checked for mapping being NULL after taking the tree_lock. However the mapping never changes for a locked page. Remove two more checks for mapping being NULL.

mm/migrate.c +2 -3
Index: linux-2.6.17-rc3/mm/migrate.c
===================================================================
@@ -291,7 +291,7 @@
 
 	if (!mapping) {
 		/* Anonymous page */
-		if (page_count(page) != 1 || !page->mapping)
+		if (page_count(page) != 1)
 			return -EAGAIN;
 		return 0;
 	}
@@ -302,8 +302,7 @@
 						&mapping->page_tree,
 						page_index(page));
 
-	if (!page_mapping(page) ||
-			page_count(page) != 2 + !!PagePrivate(page) ||
+	if (page_count(page) != 2 + !!PagePrivate(page) ||
 			*radix_pointer != page) {
 		write_unlock_irq(&mapping->tree_lock);
 		return -EAGAIN;

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

#3 lore raw

PATCH 4/5 page migration: Fix up remove_migration_ptes()

In reply to Christoph Lameter

Fix up remove_migration_ptes()

Add the update_mmu/lazy_mmu_update() calls that most arches need and that IA64 needs for executable pages.

Also move the call to page_address_in_vma into remove_migrate_pte() and check for the possible -EFAULT return code.

mm/migrate.c +11 -4
Index: linux-2.6.17-rc3-mm1/mm/migrate.c
===================================================================
@@ -123,7 +123,7 @@ static inline int is_swap_pte(pte_t pte)
 /*
  * Restore a potential migration pte to a working pte entry
  */
-static void remove_migration_pte(struct vm_area_struct *vma, unsigned long addr,
+static void remove_migration_pte(struct vm_area_struct *vma,
 		struct page *old, struct page *new)
 {
 	struct mm_struct *mm = vma->vm_mm;
@@ -133,6 +133,10 @@ static void remove_migration_pte(struct 
  	pmd_t *pmd;
 	pte_t *ptep, pte;
  	spinlock_t *ptl;
+	unsigned long addr = page_address_in_vma(new, vma);
+
+	if (addr == -EFAULT)
+		return;
 
  	pgd = pgd_offset(mm, addr);
 	if (!pgd_present(*pgd))
@@ -175,6 +179,10 @@ static void remove_migration_pte(struct 
 	else
 		page_add_file_rmap(new);
 
+	/* No need to invalidate - it was non-present before */
+	update_mmu_cache(vma, addr, pte);
+	lazy_mmu_prot_update(pte);
+
 out:
 	pte_unmap_unlock(ptep, ptl);
 }
@@ -196,7 +204,7 @@ static void remove_file_migration_ptes(s
 	spin_lock(&mapping->i_mmap_lock);
 
 	vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff)
-		remove_migration_pte(vma, page_address_in_vma(new, vma), old, new);
+		remove_migration_pte(vma, old, new);
 
 	spin_unlock(&mapping->i_mmap_lock);
 }
@@ -223,8 +231,7 @@ static void remove_anon_migration_ptes(s
 	spin_lock(&anon_vma->lock);
 
 	list_for_each_entry(vma, &anon_vma->head, anon_vma_node)
-		remove_migration_pte(vma, page_address_in_vma(new, vma),
-					old, new);
+		remove_migration_pte(vma, old, new);
 
 	spin_unlock(&anon_vma->lock);
 }

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

#4 lore raw

PATCH 5/5 page migration: Replace call to pageout() with writepage()

In reply to Christoph Lameter

page migration: Do not use pageout() but writepage() for fallback.

Migration cannot use pageout for fallback since the migration entries have to be removed before calling writepage. writepage (and therefore pageout) may drop the lock and expose migration entries. Removing migration entries in turn increases the mapcount which results in pageout() not writing out the page. sigh.

This problem was re-introduced with the use of migration entries for file backed pages.

Implement our own writeout() function (this approach was posted already last week but not included in the patch reorg) and undo the export of pageout() since page migration was the only user of pageout().

Also remove a definition for remove_vma_swap() that was somehow left over from earlier changes.

mm/migrate.c +48 -17
Index: linux-2.6.17-rc3-mm1/mm/migrate.c
===================================================================
@@ -24,6 +24,7 @@
 #include <linux/topology.h>
 #include <linux/cpu.h>
 #include <linux/cpuset.h>
+#include <linux/writeback.h>
 
 #include "internal.h"
 
@@ -468,28 +469,58 @@ int buffer_migrate_page(struct address_s
 EXPORT_SYMBOL(buffer_migrate_page);
 
 /*
- * Default handling if a filesystem does not provide a migration function.
+ * Writeback a page to clean the dirty state
  */
-static int fallback_migrate_page(struct address_space *mapping,
-	struct page *newpage, struct page *page)
+static int writeout(struct address_space *mapping, struct page *page)
 {
-	if (PageDirty(page)) {
-		/*
-		 * A dirty page may imply that the underlying filesystem has
-		 * the page on some queue. So the page must be clean for
-		 * migration. Writeout may mean we loose the lock and the
-		 * page state is no longer what we checked for earlier.
-		 * At this point we know that the migration attempt cannot
-		 * be successful.
-		 */
-		remove_migration_ptes(page, page);
+	struct writeback_control wbc = {
+		.sync_mode = WB_SYNC_NONE,
+		.nr_to_write = 1,
+		.range_start = 0,
+		.range_end = LLONG_MAX,
+		.nonblocking = 1,
+		.for_reclaim = 1
+	};
+	int rc;
 
-		if (pageout(page, mapping) == PAGE_SUCCESS)
-			/* unlocked. Relock */
-			lock_page(page);
+	if (!mapping->a_ops->writepage)
+		/* No write method for the address space */
+		return -EINVAL;
 
+	if (!clear_page_dirty_for_io(page))
+		/* Someone else already triggered a write */
 		return -EAGAIN;
-	}
+
+	/*
+	 * A dirty page may imply that the underlying filesystem has
+	 * the page on some queue. So the page must be clean for
+	 * migration. Writeout may mean we loose the lock and the
+	 * page state is no longer what we checked for earlier.
+	 * At this point we know that the migration attempt cannot
+	 * be successful.
+	 */
+	remove_migration_ptes(page, page);
+
+	rc = mapping->a_ops->writepage(page, &wbc);
+	if (rc < 0)
+		/* I/O Error writing */
+		return -EIO;
+
+	if (rc != AOP_WRITEPAGE_ACTIVATE)
+		/* unlocked. Relock */
+		lock_page(page);
+
+	return -EAGAIN;
+}
+
+/*
+ * Default handling if a filesystem does not provide a migration function.
+ */
+static int fallback_migrate_page(struct address_space *mapping,
+	struct page *newpage, struct page *page)
+{
+	if (PageDirty(page))
+		return writeout(mapping, page);
 
 	/*
 	 * Buffers may be managed in a filesystem specific way.
include/linux/swap.h +0 -15
Index: linux-2.6.17-rc3-mm1/include/linux/swap.h
===================================================================
@@ -188,20 +188,6 @@ extern unsigned long shrink_all_memory(u
 extern int vm_swappiness;
 extern int remove_mapping(struct address_space *mapping, struct page *page);
 
-/* possible outcome of pageout() */
-typedef enum {
-	/* failed to write page out, page is locked */
-	PAGE_KEEP,
-	/* move page to the active list, page is locked */
-	PAGE_ACTIVATE,
-	/* page has been sent to the disk successfully, page is unlocked */
-	PAGE_SUCCESS,
-	/* page is clean and locked */
-	PAGE_CLEAN,
-} pageout_t;
-
-extern pageout_t pageout(struct page *page, struct address_space *mapping);
-
 #ifdef CONFIG_NUMA
 extern int zone_reclaim_mode;
 extern int zone_reclaim_interval;
@@ -264,7 +250,6 @@ extern int remove_exclusive_swap_page(st
 struct backing_dev_info;
 
 extern spinlock_t swap_lock;
-extern int remove_vma_swap(struct vm_area_struct *vma, struct page *page);
 
 /* linux/mm/thrash.c */
 extern struct mm_struct * swap_token_mm;
mm/vmscan.c +13 -1
Index: linux-2.6.17-rc3-mm1/mm/vmscan.c
===================================================================
@@ -291,11 +291,23 @@ static void handle_write_error(struct ad
 	unlock_page(page);
 }
 
+/* possible outcome of pageout() */
+typedef enum {
+	/* failed to write page out, page is locked */
+	PAGE_KEEP,
+	/* move page to the active list, page is locked */
+	PAGE_ACTIVATE,
+	/* page has been sent to the disk successfully, page is unlocked */
+	PAGE_SUCCESS,
+	/* page is clean and locked */
+	PAGE_CLEAN,
+} pageout_t;
+
 /*
  * pageout is called by shrink_page_list() for each dirty page.
  * Calls ->writepage().
  */
-pageout_t pageout(struct page *page, struct address_space *mapping)
+static pageout_t pageout(struct page *page, struct address_space *mapping)
 {
 	/*
 	 * If the page is dirty, only perform writeback if that write

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