Benjamin Herrenschmidt wrote:
> On Mon, 2004-11-08 at 17:06 +0800, Antonino A. Daplas wrote:
>
> >
> > How about this patch? This is almost the original macro in riva_hw.h,
> > with the __force annotation.
>
> I don't like it neither. It lacks barriers. the rivafb driver
> notoriously lacks barriers, except in a few places where it was so bad
> that it actually broke all the time, where we added some. This
> originates from the X "nv" driver written by Mark Vojkovich who didn't
> want to hear about barriers for perfs reasons I think.
Could this be the reason why in 2.6.7 I get solid lockups when
switching to nv-driven X and rivafb-driven console? Is there
something I could test to see if this is the reason?
--
Giuseppe "Oblomov" Bilotta
Can't you see
It all makes perfect sense
Expressed in dollar and cents
Pounds shillings and pence
(Roger Waters)
On Tue, Nov 09, 2004 at 12:07:58AM +0100, Giuseppe Bilotta wrote:
> Benjamin Herrenschmidt wrote:
> > On Mon, 2004-11-08 at 17:06 +0800, Antonino A. Daplas wrote:
> >
> > >
> > > How about this patch? This is almost the original macro in riva_hw.h,
> > > with the __force annotation.
> >
> > I don't like it neither. It lacks barriers. the rivafb driver
> > notoriously lacks barriers, except in a few places where it was so bad
> > that it actually broke all the time, where we added some. This
> > originates from the X "nv" driver written by Mark Vojkovich who didn't
> > want to hear about barriers for perfs reasons I think.
>
> Could this be the reason why in 2.6.7 I get solid lockups when
> switching to nv-driven X and rivafb-driven console? Is there
> something I could test to see if this is the reason?
Please try 2.6.10-rc1-mm5. It should work there.
-- Guido