On Sun, Mar 27, 2005 at 12:26:11AM +0100, Manuel Bouyer wrote:
> This would probablt be a good idea, but isn't __insn_barrier() a
> gcc builtin ?

Should have checked again before posting ...
it's defined in sys/cdef.h, but isn't preceeded by a #ifndef __insn_barrier.
It's easy to change if we want to.

We would probably want to define a MD __insn_barrier on other systems too.

BTW, shouldn't the bus_space functions have such barriers too ?
The esiop(4) scheduler could be affected by this, for example.

Hum, I wonder if intructions reordering could be the reason of a strange
behavior I noticed with pciide on some sparc64 systems ... I should
probably start using bus_space_barrier() in my drivers :)

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 26 ans d'experience feront toujours la difference
--