OK ... I'm want to speed up the I/O initated by brw_page() ... without
a condition for the `wake_up' of the buffer wait queue slows down this I/O
is really slow for the case of nr of buffer heads equals zero.
This can be tested very easily: run a `time e2fsck -f' on a big partition
in a loop and count the long check cycles. With a wake_up() call you
have to run e2fsck very often to get a speed up due to the same sized buffers.
[...]
> Also, your amended patch leaves out a much more important change pointed
> out by Mark Hemment -- after the call to try_to_free_buffer in
> find_candidate, there needs to be a test
>
> if (!bh)
> break;
Hmmm ... this null pointer detection is already included in
pre-patch-2.0.31-2. My patches are always against pre-patch-2.0.31-2 because
of the fixes in the net and isdn layer.
Werner