hi,

> Paul suggested eliminating the AGE queue some time ago.  I disagreed with him
> then, but looking at things more closely I think that if we keep the current
> use of B_AGE, but replace enqueueing on BQ_AGE with releasing the buffer's
> memory back to the system, it's the right thing to do.
> 
> I have a first cut at this, but clearly I've hosed the locking somehow; the
> system gets to the point where it is copying in init and hangs.  I'd really
> appreciate it if others could look at the patch below and see if they can
> find my mistake.

you should brelvp() and bremhash() if needed before free a struct buf.

YAMAMOTO Takashi