Fixing R300 hangs

[Mike A. Harris]

John Thacker wrote:
> Mike A. Harris wrote:
> 
> 
>>Doing a bit of investigation has shown that it isn't just the
>>r300 DRI driver that is instable, but even just loading the X server
>>DRI module with many r300 or newer cards causes the system to crash,
>>even if DRI is actually disabled, and even if the r300 DRI driver is
>>not even present on the system.
> 
> 
> I believe that this is almost surely this problem:
> http://lists.freedesktop.org/archives/xorg/2005-December/011678.html
> https://bugs.freedesktop.org/show_bug.cgi?id=4847
> http://webcvs.freedesktop.org/xorg/driver/xf86-video-ati/ChangeLog?rev=1.22&view=markup
> 
> There are some important memory map and other fixes for Radeon cards
> (especially R300 and above) that hit the xorg CVS tree on February
> 17, after 7.0.  (The author considered them a little too experimental
> for 7.0.)  Apparently without the fixes many people saw lockups using 
> R300 cards without DRI enabled or the r300 DRI drive present.  However,
> with these fixes the instability goes away.
> 
> Perhaps these upstream CVS patches could be integrated into the Fedora
> Core patches.

The great thing about modular X, is that it makes it possible for X.Org
to release new versions of individual components much more easily.  This
means that problems of this nature which get fixed, can be integrated
into the next stable bugfix release of the particular component.

Any patches that get committed to the CVS head of any given component in
X.Org, show commitment that they'll be present in the next major release
of the X Window System - 7.1.  If the upstream developers feel confident
about a particular patch being stable for the existing release, they can
commit it to the stable branch of the particular component and release
an update of that specific component, without having to release the
entire window system.

Once upstream has integrated the patches into the stable branch, and has
released an updated set of packages with the fixes present, we will
consider releasing a Fedora Core 5 update to the new packages.

> If so, the r300 driver could be left in FC5.  Of course,
> we are running low on time, but I think it's a relatively critical bug.
> Especially if the recent Mesa change doesn't solve the crashes for
> everyone.

My current plan for the r300 dri driver, is to leave it out of FC5, and
once the various issues are resolved upstream we will re-enable it in
rawhide post-FC5.  This will allow the driver to get the level of
testing that is needed, and hopefully to stabilize sooner.  Then, once
the major problems are resolved upstream by X.Org/DRI developers, and
new upstream stable packages have been released with the fixes, we will
probably include them in rawhide.

If the r300 DRI support matures upstream enough that it is not going to
be a major support burden, then we will consider enabling it in a future
FC5 update.





-- 
Mike A. Harris  *  Open Source Advocate  *  http://mharris.ca
                       Proud Canadian.