Re: 2.6.5-rc1-mm2 and direct_read_under and wb

[Andrew Morton <akpm@osdl.org>]

Daniel McNeil <daniel@osdl.org> wrote:
>
> The test has been running over 5 hours now without seeing uninitialized
>  data.  I'll let it run overnight, but it looks like the small 
>  __block_write_full_page patch fixes the problem.

OK, thanks.  It does cause gargantuan amounts of CPU consumption as
expected.  The below version fixes that up.  I'm not 100% happy with it,
but it seems OK on the profiles.


diff -puN fs/buffer.c~block_write_full_page-redirty fs/buffer.c
--- 25/fs/buffer.c~block_write_full_page-redirty	2004-03-23 01:06:59.281253176 -0800
+++ 25-akpm/fs/buffer.c	2004-03-23 01:09:11.313181272 -0800
@@ -1810,14 +1810,18 @@ static int __block_write_full_page(struc
 		get_bh(bh);
 		if (!buffer_mapped(bh))
 			continue;
-		if (wbc->sync_mode != WB_SYNC_NONE) {
+		/*
+		 * If it's a fully non-blocking write attempt and we cannot
+		 * lock the buffer then redirty the page.  Note that this can
+		 * potentially cause a busy-wait loop from pdflush and kswapd
+		 * activity, but those code paths have their own higher-level
+		 * throttling.
+		 */
+		if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
 			lock_buffer(bh);
-		} else {
-			if (test_set_buffer_locked(bh)) {
-				if (buffer_dirty(bh))
-					__set_page_dirty_nobuffers(page);
-				continue;
-			}
+		} else if (test_set_buffer_locked(bh)) {
+			__set_page_dirty_nobuffers(page);
+			continue;
 		}
 		if (test_clear_buffer_dirty(bh)) {
 			if (!buffer_uptodate(bh))

_