Re: [PATCH] VFS: Go through the LRU list of inode from head

[Jan Kara <jack@suse.cz>]

On Tue 26-06-12 03:35:46, mc@linux.vnet.ibm.com wrote:
> Quoting Jan Kara <jack@suse.cz>:
> 
> >On Thu 21-06-12 17:00:27, Cong Meng wrote:
> >>Go through the LRU list of inode from head.
> >>
> >>(I'm not sure whether there is any trick here I doesn't get.  If yes,
> >>any one could explain it)
> >  Look at inode_lru_list_add(). It adds at the head of the list. So you
> >should take from the tail to get the least recently used element...
> 
> I still have a quetion about the subsequent code and comment:
> 
> inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
> /*
>  * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
>  * so use a trylock. If we fail to get the lock, just move the
>  * inode to the back of the list so we don't spin on it.
>  */
> if (!spin_trylock(&inode->i_lock)) {
>         list_move_tail(&inode->i_lru, &sb->s_inode_lru);
>         continue;
> }
> 
> Shouldn't the inode be moved to the head to avoid spin on it?
  Yes, it should.

> I note that list_move was replaced by list_move_tail purposely in a commit.
  Right, you are speaking about Christoph's commit 62a3ddef? I agree that
commit looks bogus and should be reverted AFAICT. Christoph?

> and below piece of code (at the bottom of prune_icache_sb()):
> 
> if (inode != list_entry(sb->s_inode_lru.next,
>                         struct inode, i_lru))
>         continue;       /* wrong inode or list_empty */
> 
> Should the inode be compared against to the tail of the list other
> than the head
> after re-get the lru lock?
  And you seem to be right here as well. Thanks for having a look!

								Honza

> >>Signed-off-by: Cong Meng <mc@linux.vnet.ibm.com>
> >>---
> >> fs/inode.c |    2 +-
> >> 1 files changed, 1 insertions(+), 1 deletions(-)
> >>
> >>diff --git a/fs/inode.c b/fs/inode.c
> >>index 775cbab..aac8449 100644
> >>--- a/fs/inode.c
> >>+++ b/fs/inode.c
> >>@@ -704,7 +704,7 @@ void prune_icache_sb(struct super_block *sb,
> >>int nr_to_scan)
> >> 		if (list_empty(&sb->s_inode_lru))
> >> 			break;
> >>
> >>-		inode = list_entry(sb->s_inode_lru.prev, struct inode, i_lru);
> >>+		inode = list_entry(sb->s_inode_lru.next, struct inode, i_lru);
> >>
> >> 		/*
> >> 		 * we are inverting the sb->s_inode_lru_lock/inode->i_lock here,
> >>--
> >>1.7.5.4
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR