ChangeSet 1.1521.20.2, 2005/01/20 14:29:18-06:00, [EMAIL PROTECTED]
JFS: fix livelock waiting for stale metapage
Several waitors were keeping mp->count from going to zero, so we
would never release the page. Simplify the logic by doing a busy
wait without locking the metapage.
Signed-off-by: Dave Kleikamp <[EMAIL PROTECTED]>
jfs_metapage.c | 23 ++++++-----------------
1 files changed, 6 insertions(+), 17 deletions(-)
diff -Nru a/fs/jfs/jfs_metapage.c b/fs/jfs/jfs_metapage.c
--- a/fs/jfs/jfs_metapage.c 2005-02-23 10:05:50 -08:00
+++ b/fs/jfs/jfs_metapage.c 2005-02-23 10:05:50 -08:00
@@ -298,14 +298,14 @@
mp = search_hash(hash_ptr, mapping, lblock);
if (mp) {
page_found:
- mp->count++;
- lock_metapage(mp);
- spin_unlock(&meta_lock);
if (test_bit(META_stale, &mp->flag)) {
- release_metapage(mp);
- yield(); /* Let other waiters release it, too */
+ spin_unlock(&meta_lock);
+ yield();
goto again;
}
+ mp->count++;
+ lock_metapage(mp);
+ spin_unlock(&meta_lock);
if (test_bit(META_discard, &mp->flag)) {
if (!new) {
jfs_error(inode->i_sb,
@@ -518,7 +518,6 @@
}
if (mp->page) {
- /* Releasing spinlock, we have to check mp->count later */
set_bit(META_stale, &mp->flag);
spin_unlock(&meta_lock);
kunmap(mp->page);
@@ -555,12 +554,6 @@
list_del(&mp->synclist);
LOGSYNC_UNLOCK(log);
}
- if (mp->count) {
- /* Someone else is trying to get this metpage */
- unlock_metapage(mp);
- spin_unlock(&meta_lock);
- return;
- }
remove_from_hash(mp, meta_hash(mp->mapping, mp->index));
spin_unlock(&meta_lock);
@@ -589,12 +582,8 @@
mp = search_hash(hash_ptr, mapping, lblock);
if (mp) {
if (test_bit(META_stale, &mp->flag)) {
- /* Racing with release_metapage */
- mp->count++;
- lock_metapage(mp);
spin_unlock(&meta_lock);
- /* racing release_metapage should be done now */
- release_metapage(mp);
+ yield();
goto again;
}
-
To unsubscribe from this list: send the line "unsubscribe bk-commits-24" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
