Thanks for reporting and bisecting. Certainly there's a bug in shmem
(that we ought to have caught long before it reached 36-rc1: sorry),
and it is probably the cause of your crashes; but it's possible there's
a similar bug elsewhere too, something else messing up the percpu_counters
list - so please check if the patch below really does fix it for you.
I notice your dmesg also showed fs/inode.c:692 unlock_new_inode()
warnings from get_cramfs_inode(): those would be unrelated, and
probably from Al Viro's recent changes in cramfs/inode.c - Cc'ed.
Thanks,
Hugh
[PATCH] shmem: put_super must percpu_counter_destroy
list_add() corruption messages reported from shmem_fill_super()'s recently
introduced percpu_counter_init(): shmem_put_super() needs to remember to
percpu_counter_destroy(). And also check error from percpu_counter_init().
Reported-and-bisected-by: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Signed-off-by: Hugh Dickins <hughd@google.com>
---
mm/shmem.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
--- 2.6.36-rc1/mm/shmem.c 2010-08-16 00:18:01.000000000 -0700
+++ linux/mm/shmem.c 2010-08-17 14:42:56.000000000 -0700
@@ -2325,7 +2325,10 @@ static int shmem_show_options(struct seq
static void shmem_put_super(struct super_block *sb)
{
- kfree(sb->s_fs_info);
+ struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
+
+ percpu_counter_destroy(&sbinfo->used_blocks);
+ kfree(sbinfo);
sb->s_fs_info = NULL;
}
@@ -2367,7 +2370,8 @@ int shmem_fill_super(struct super_block
#endif
spin_lock_init(&sbinfo->stat_lock);
- percpu_counter_init(&sbinfo->used_blocks, 0);
+ if (percpu_counter_init(&sbinfo->used_blocks, 0))
+ goto failed;
sbinfo->free_inodes = sbinfo->max_inodes;
sb->s_maxbytes = SHMEM_MAX_BYTES;
--
