On Mon, 26 Aug 2002, Mike Wlodarczyk wrote:

> While I'm running 1.6_RC1/i386, I think this may affect current
> too.
>
> I've been experiencing some silent lockups since moving to the
> 1.6 branch when running XFree86 4.2 from CVS and using the aperture
> driver on a K6-3. The lockup occurs randomly, but at least once a
> day. I normally notice it when I'm moving the mouse and it just
> stops. No pings, no response from the serial terminal, nothing.
>
> I've removed "options MTRR" and "options PERFCTRS" from the kernel
> config and have not seen the problem, but I only did this yesterday.

Pretty wierd, since the MTRR code is only called when the Xserver is
started, and PERFCTRS wouldn't do a thing unless you run "pmc", and
that doesn't seem to work on K6.

> While the MTRR support seems limited to providing the system calls
> to use those registers, is it possible there's some conflict anyway ?

I find that to be pretty unlikely. The aperture and kernel mtrr's use
a completely different interface. Besides, I've played with the mtrr's
on my K6-2, and setting them wrong doesn't seem to have any
consequences. [Setting them right doesn't buy you much either -- I got
a more noticable boost in video playback performance by building the
kernel with -march=k6.]

By the way, which aperture driver? The shell archive in xsrc/xfree is
broken (the bits of the mask are inverted). Even if you use the one in
pkgsrc, which has that fixed, I think the X server will always use the
kernel to set the mtrr's if built on a system that supports it. What
does it say about "mtrr" in "/var/log/XFree86.0.log"?

I suspect that without kernel MTRR support, you're not setting the
mtrr's at all, or trying and failing, and maybe that triggers an X
server bug?

> Or perhaps the K6-3 is part of it ?

AMD says the UWCCCR implementation is the same for K6-2 and K6-3.


Frederick