Re: VM in 2.4.7-pre hurts...

[Andrea Arcangeli <andrea@suse.de>]

On Tue, Jul 10, 2001 at 07:43:15AM +0200, Andrea Arcangeli wrote:
> My first arguments about "overkill" were for async I/O and kiobufs, where
> the race cannot trigger. Mainly for the kiobufs I/O I'm still not very
> convinced.

another issue is that I don't see any value in defining the
unlock_buffer() with the get_bh/put_bh in it. The get_bh over there is
just useless because we don't have the memory barriers in the get_bh and
try_to_free_buffer paths, that's what party fooled me last night in
thinking we didn't need to get_bh in the implicit synchronization point
but that using unlock_buffer (with the get_bh in it was enough), otherwise
unlock_buffer would been pointless (and infact now it sorted out it is
as far I can tell ;).

So I'd kill unlock_buffer and replace it with __unlock_buffer.

Andrea