On Sun, Sep 10, 2006 at 08:42:17PM +0900, YAMAMOTO Takashi wrote:
> 
> are you sure if it works in the case of PGO_LOCKED?
> at a glance, it seems broken.

does it _need_ to?  Or are the comments at the top of uvn_get()
incorrect - according to those PGO_LOCKED only happens on pages going
out.  I was not intending checking pages being written.

> have you tested it with LOCKDEBUG or MULTIPROCESSOR?
>

I have now.  I get a panic, effectively, in vfs_composefh_alloc()
saying a simple lock is held when it is trying to allocate some wired
memory.  It seems, perhaps, that fileassoc should pre-allocate some
fhandle_t sized buffers for use in the lookup routines to avoid this
lock contention.

> (i thought i made the same comment on an old version of this while ago...)
> 

You probably did but I think they were overshadowed by the fact that I
previously had the per-page checks in genfs - shifting that resulted
in a lot of work that was then simplified a lot by fileassoc... I lost
track of the other issues.  my bad.

-- 
Brett Lymn