HD4550 DRI issues

Robert Noland rnoland at FreeBSD.org
Thu Oct 1 10:16:40 UTC 2009


On Wed, 2009-09-30 at 21:18 -0500, Richard Kolkovich wrote:
> On Sat, Sep 26, 2009 at 04:47:01PM -0500, Robert Noland wrote:
> > 
> > Ok, that eliminates everything to do with the card and X.  Let me talk
> > to some folks and see if we can figure out where to go from here...  I'm
> > wondering if this might be BIOS or CPU related now...
> > 
> 
> Any ideas yet?  For some reason, I now need to disable mtrr for my nvidia-driver to work (the card
> my 4550 is hopefully replacing...I'm having other issues with nvidia-driver and would prefer an
> open-source solution).  I don't think I've tried the radeon with mtrr disabled.
> 
> My 30 day return window on this card is almost up, so I may end up returning it if we can't figure
> this out.  Apparently, a different card wouldn't help, either. :-\

All of the evidence suggests that this has nothing to do with the card.
Basically, with the test program the only card specific things that
happen are those that happen at driver load time.  We haven't even told
the card about the memory that we have allocated yet.  It has been
suggested that it might be a bug in drm's mmap routine, but I can't find
it at this point.  It has also been suggested that you run memcheck to
verify the ram, but since things seem to be correct from the kernel
perspective, I don't think that is the issue.

The mtrr issue, might be a clue though.  It sets the caching mode for
the memory allocation.  I'm mostly using PAT now, (particularly with the
patch that I gave you).  Did you send me a "memcontrol list", I don't
recall at this point.  The x58 appears to deal with memory rather
differently than traditional chipsets, so I wonder if it might not be a
chipset or BIOS bug.

As for nvidia, you can always use nouveau.  That will (*should*) get you
EXA and Xv acceleration, but not 3d at this point.  Follow the
instructions in the xf86-video-nouveau for getting my kernel patch how
to set things up.  However, that driver also uses scatter-gather memory,
perhaps even more so than the radeon driver.  So, given what we have
seen so far, I don't know that it will not produce corruption or
possibly worse.

I've spent a fair amount of time digging through the mmap and memory
allocation bits in the kernel, but haven't found anything that looks
wrong.  What I don't understand is, why this doesn't effect all or at
least all i386 systems.  So, I think it almost has to be chipset
specific. .... Well, I just went back and found one of the prior reports
of this.  It looks like that (assuming it is the same cause) was on an
intel 945 chipset running 7.2-RELEASE i386.  Have you considered running
amd64?  Your i7 is certainly worthy of it and I would be curious to see
if the problem exists when running amd64 also.

robert.
 
> 
> Thanks,
> 
-- 
Robert Noland <rnoland at FreeBSD.org>
FreeBSD



More information about the freebsd-x11 mailing list