Re: dentry bloat.

[viro]

On Sat, May 08, 2004 at 03:11:59AM -0700, Andrew Morton wrote:
> Andrew Morton <akpm@xxxxxxxx> wrote:
> >
> > Here be another patch.
> 
> 
> Can't help myself!
> 
> 
> - d_vfs_flags can be removed - just use d_flags.  It looks like someone
>   added d_vfs_flags with the intent of doing bitops on it, but all
>   modifications are under dcache_lock.  4 bytes saved.

Bzzert.  ->d_vfs_flags modifications are under dcache_lock; ->d_flags ones
are *not* - they are up to whatever filesystem code happens to use them.
 
> - Pack things so that dentry.d_name.len and dentry.d_flags occupy the same
>   word.  4 bytes saved.

d_name.len is accessed on very hot paths.  In particular, we are looking
at it while traversing hash chains and we do that without dcache_lock
(see callers of ->d_compare()).

If we are going to hold dcache_lock in __d_lookup(), we can get rid of
a _lot_ more than 2 bytes.  32 bytes that came from RCU are there pretty
much for a single reason - to avoid dcache_lock on that path.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/