Alex Deucher and I came up with the current patch in
https://bugs.freedesktop.org/show_bug.cgi?id=1912
It is supposed to reestablish the clock setup code from 6.7.0, which so
far have worked pretty well. When 6.8.0 introduced the dynamic clocks
feature we started seeing hard lockups on radeon 7000s, typically on x86
smp boxes. I tracked it down to the dynamic clocks feature, and it
triggers wether or not you enable dynamic clocks. So for our FC-3 6.8.1
RPMs I added this patch
https://bugs.freedesktop.org/attachment.cgi?id=1432
to revert to the previous clock initialization code, which worked pretty
well. When 6.8.2 came out it had a slightly different patch applied
Well, the patch that was in RH was sort-of mixing up the dynamic clocks
code and the "old style" clock forcing. The end result craps out on
rv350 at least. I don't like the big hammer of just disabling the
dynamic clocks code, this is a significant loss on laptops.
On the other hands, I have posted fixes for the Dynamic Clocks "off"
case which indeed was bogus, plus workarounds for chip erratas around
PLL accesses that triggered, among others, lockups on rv250, in addition
to the fact that we don't touch dynamic clocks on non-mobility chips. So
I suppose that my patch set should be tried first, and if the problem
still happen, get me in touch with the user so I can try to find out
precisely what bit of dynamic clocking is causing the problem.
https://bugs.freedesktop.org/attachment.cgi?id=1433
which didn't put the 6.7.0 clock code back in, but it only invoked the
dynamic clock code for mobility chips. When we updated our RPMs to the
6.8.2 release and dropped the first patch, we got the regression bug
mentioned earlier. Apparently some radeon 7000s (I can't reproduce the
regression on my radeon 7000 box) require the 6.7.0 clock setup to work,
so I put back the patch we used with 6.8.1, which is what we have now.
David, thanks for building the RPMs with Bens patches and posting it in
the bug, I've been meaning to do that. If these patches work well,
we'll included them in FC5, but I'm reluctant to change this much for FC4.
We may want to force some clocks on some desktop 7000's. It may be that
the BIOS is indeed improperly enabling some dynamic clcoks. If you find
somebody with a repro case, please point that person to me and I'll
chase the problem with him/her.
Ben.