[PATCH] VM dirty page balancing

- The balance_dirty_pages() logic is simply wrong.  It goes:

	if (value > threshold)
		go_and_write(value - threshold);

  which is just fine for a single process writing data.  But
  for many processes, they *all* go and bring things back into
  balance, and too much data gets written out.

- The

	go_and_write(this much)

  logic is inoperative, because I turned off the ->writeback_mapping()
  function in ext2.  So a call to writeback_unlocked_inodes(this_much)
  doesn't actually decrement and test *this_much.  It will walk every
  inode, all the time.  Silly.

So quickly fixing the above things, the amount of dirty+writeback
memory in the machine nicely stabilises at 500 megabytes across
the run.

fs/ext2/inode.c

@@ -591,6 +591,8 @@ struct address_space_operations ext2_aops = {
 	commit_write: generic_commit_write,
 	bmap: ext2_bmap,
 	direct_IO: ext2_direct_IO,
+	writeback_mapping: generic_writeback_mapping,
+	vm_writeback: generic_vm_writeback,
 };
 
 /*

fs/proc/proc_misc.c

@@ -159,7 +159,7 @@ static int meminfo_read_proc(char *page, char **start, off_t off,
 		"SwapTotal:    %8lu kB\n"
 		"SwapFree:     %8lu kB\n"
 		"Dirty:        %8lu kB\n"
-		"Locked:       %8lu kB\n",
+		"Writeback:    %8lu kB\n",
 		K(i.totalram),
 		K(i.freeram),
 		K(i.sharedram),
@@ -175,7 +175,7 @@ static int meminfo_read_proc(char *page, char **start, off_t off,
 		K(i.totalswap),
 		K(i.freeswap),
 		K(ps.nr_dirty),
-		K(ps.nr_locked)
+		K(ps.nr_writeback)
 		);
 
 	return proc_calc_metrics(page, start, off, count, eof, len);

include/linux/page-flags.h

@@ -13,7 +13,7 @@
  *
  * The PG_private bitflag is set if page->private contains a valid value.
  *
- * During disk I/O, PG_locked_dontuse is used. This bit is set before I/O and
+ * During disk I/O, PG_locked is used. This bit is set before I/O and
  * reset when I/O completes. page_waitqueue(page) is a wait queue of all tasks
  * waiting for the I/O on this page to complete.
  *
@@ -28,7 +28,7 @@
  *
  * Note that the referenced bit, the page->lru list_head and the active,
  * inactive_dirty and inactive_clean lists are protected by the
- * pagemap_lru_lock, and *NOT* by the usual PG_locked_dontuse bit!
+ * pagemap_lru_lock, and *NOT* by the usual PG_locked bit!
  *
  * PG_error is set to indicate that an I/O error occurred on this page.
  *
@@ -47,7 +47,7 @@
  * locked- and dirty-page accounting.  The top eight bits of page->flags are
  * used for page->zone, so putting flag bits there doesn't work.
  */
-#define PG_locked_dontuse	 0	/* Page is locked. Don't touch. */
+#define PG_locked	 0	/* Page is locked. Don't touch. */
 #define PG_error		 1
 #define PG_referenced		 2
 #define PG_uptodate		 3
@@ -71,7 +71,7 @@
  */
 extern struct page_state {
 	unsigned long nr_dirty;
-	unsigned long nr_locked;
+	unsigned long nr_writeback;
 	unsigned long nr_pagecache;
 } ____cacheline_aligned_in_smp page_states[NR_CPUS];
 
@@ -91,37 +91,16 @@ extern void get_page_state(struct page_state *ret);
 /*
  * Manipulation of page state flags
  */
-#define PageLocked(page)	test_bit(PG_locked_dontuse, &(page)->flags)
+#define PageLocked(page)		\
+		test_bit(PG_locked, &(page)->flags)
 #define SetPageLocked(page)		\
-	do {								\
-		if (!test_and_set_bit(PG_locked_dontuse,		\
-				&(page)->flags))			\
-			inc_page_state(nr_locked);			\
-	} while (0)
+		set_bit(PG_locked, &(page)->flags)
 #define TestSetPageLocked(page)		\
-	({								\
-		int ret;						\
-		ret = test_and_set_bit(PG_locked_dontuse,		\
-					&(page)->flags);		\
-		if (!ret)						\
-			inc_page_state(nr_locked);			\
-		ret;							\
-	})
+		test_and_set_bit(PG_locked, &(page)->flags)
 #define ClearPageLocked(page)		\
-	do {								\
-		if (test_and_clear_bit(PG_locked_dontuse,		\
-				&(page)->flags))			\
-			dec_page_state(nr_locked);			\
-	} while (0)
+		clear_bit(PG_locked, &(page)->flags)
 #define TestClearPageLocked(page)	\
-	({								\
-		int ret;						\
-		ret = test_and_clear_bit(PG_locked_dontuse,		\
-				&(page)->flags);			\
-		if (ret)						\
-			dec_page_state(nr_locked);			\
-		ret;							\
-	})
+		test_and_clear_bit(PG_locked, &(page)->flags)
 
 #define PageError(page)		test_bit(PG_error, &(page)->flags)
 #define SetPageError(page)	set_bit(PG_error, &(page)->flags)
@@ -201,12 +180,36 @@ extern void get_page_state(struct page_state *ret);
 #define PagePrivate(page)	test_bit(PG_private, &(page)->flags)
 
 #define PageWriteback(page)	test_bit(PG_writeback, &(page)->flags)
-#define SetPageWriteback(page)	set_bit(PG_writeback, &(page)->flags)
-#define ClearPageWriteback(page) clear_bit(PG_writeback, &(page)->flags)
+#define SetPageWriteback(page)						\
+	do {								\
+		if (!test_and_set_bit(PG_writeback,			\
+				&(page)->flags))			\
+			inc_page_state(nr_writeback);			\
+	} while (0)
 #define TestSetPageWriteback(page)					\
-	test_and_set_bit(PG_writeback, &(page)->flags)
+	({								\
+		int ret;						\
+		ret = test_and_set_bit(PG_writeback,			\
+					&(page)->flags);		\
+		if (!ret)						\
+			inc_page_state(nr_writeback);			\
+		ret;							\
+	})
+#define ClearPageWriteback(page)					\
+	do {								\
+		if (test_and_clear_bit(PG_writeback,			\
+				&(page)->flags))			\
+			dec_page_state(nr_writeback);			\
+	} while (0)
 #define TestClearPageWriteback(page)					\
-	test_and_clear_bit(PG_writeback, &(page)->flags)
+	({								\
+		int ret;						\
+		ret = test_and_clear_bit(PG_writeback,			\
+				&(page)->flags);			\
+		if (ret)						\
+			dec_page_state(nr_writeback);			\
+		ret;							\
+	})
 
 /*
  * The PageSwapCache predicate doesn't use a PG_flag at this time,

mm/filemap.c

@@ -628,7 +628,7 @@ static void wait_on_page_bit(struct page *page, int bit_nr)
  */
 void ___wait_on_page_locked(struct page *page)
 {
-	wait_on_page_bit(page, PG_locked_dontuse);
+	wait_on_page_bit(page, PG_locked);
 }
 EXPORT_SYMBOL(___wait_on_page_locked);
 

mm/page-writeback.c

@@ -29,12 +29,12 @@
 /*
  * Start background writeback (via pdflush) at this level
  */
-static int dirty_background_ratio = 30;
+static int dirty_background_ratio = 40;
 
 /*
  * The generator of dirty data starts async writeback at this level
  */
-static int dirty_async_ratio = 45;
+static int dirty_async_ratio = 50;
 
 /*
  * The generator of dirty data performs sync writeout at this level
@@ -62,25 +62,28 @@ void balance_dirty_pages(struct address_space *mapping)
 	int async_thresh;
 	int sync_thresh;
 	int wake_pdflush = 0;
-	unsigned long dirty_and_locked;
+	unsigned long dirty_and_writeback;
 
 	get_page_state(&ps);
-	dirty_and_locked = ps.nr_dirty + ps.nr_locked;
+	dirty_and_writeback = ps.nr_dirty + ps.nr_writeback;
 
 	background_thresh = (dirty_background_ratio * tot) / 100;
 	async_thresh = (dirty_async_ratio * tot) / 100;
 	sync_thresh = (dirty_sync_ratio * tot) / 100;
 
-	if (dirty_and_locked > sync_thresh) {
-		int nr_to_write = dirty_and_locked - async_thresh;
+	if (dirty_and_writeback > sync_thresh) {
+		int nr_to_write = 1500;
 
+		printk("sync thresh\n");
 		writeback_unlocked_inodes(&nr_to_write, WB_SYNC_LAST, NULL);
+		get_page_state(&ps);
+		dirty_and_writeback = ps.nr_dirty + ps.nr_writeback;
 		wake_pdflush = 1;
-	} else if (dirty_and_locked > async_thresh) {
-		int nr_to_write = dirty_and_locked - async_thresh;
+	} else if (dirty_and_writeback > async_thresh) {
+		int nr_to_write = 1500;
 
 		writeback_unlocked_inodes(&nr_to_write, WB_SYNC_NONE, NULL);
-	} else if (dirty_and_locked > background_thresh) {
+	} else if (dirty_and_writeback > background_thresh) {
 		wake_pdflush = 1;
 	}
 
@@ -88,9 +91,8 @@ void balance_dirty_pages(struct address_space *mapping)
 		/*
 		 * There is no flush thread against this device. Start one now.
 		 */
-		get_page_state(&ps);
-		if (ps.nr_dirty > 0) {
-			pdflush_flush(ps.nr_dirty);
+		if (dirty_and_writeback > async_thresh) {
+			pdflush_flush(dirty_and_writeback - async_thresh);
 			yield();
 		}
 	}
@@ -109,7 +111,7 @@ void balance_dirty_pages_ratelimited(struct address_space *mapping)
 
 	preempt_disable();
 	cpu = smp_processor_id();
-	if (ratelimits[cpu].count++ >= 32) {
+	if (ratelimits[cpu].count++ >= 1000) {
 		ratelimits[cpu].count = 0;
 		preempt_enable();
 		balance_dirty_pages(mapping);

mm/page_alloc.c

@@ -584,7 +584,7 @@ void get_page_state(struct page_state *ret)
 	int pcpu;
 
 	ret->nr_dirty = 0;
-	ret->nr_locked = 0;
+	ret->nr_writeback = 0;
 	ret->nr_pagecache = 0;
 
 	for (pcpu = 0; pcpu < smp_num_cpus; pcpu++) {
@@ -592,7 +592,7 @@ void get_page_state(struct page_state *ret)
 
 		ps = &page_states[cpu_logical_map(pcpu)];
 		ret->nr_dirty += ps->nr_dirty;
-		ret->nr_locked += ps->nr_locked;
+		ret->nr_writeback += ps->nr_writeback;
 		ret->nr_pagecache += ps->nr_pagecache;
 	}
 }