Running 188.8.131.52 + drm-radeon-testing from a few days ago my laptop with a Radeon Mobility HD 2600 (m76, rv630) suspends and resumes perfectly using KMS. Running 184.108.40.206 + drm-radeon-testing HEAD at 6476cbf90ee28fda7eeee0efd9e6c9047e331d36 now causes the suspend process to hang. Machine needs a hard reset, Magic SysRq won't function and the fan ramps up to full speed suggesting the CPU is fully loaded.
git bisect shows the offending commit to be ca262a9998d46196750bb19a9dc4bd465b170ff7 "drm/ttm: Rework validation & memory space allocation (V3)"
I'm unable to test a kernel with this reverted as there are too many conflicts that I am unable to resolve successfully.
It seems this bug only occurs when I use a 64-bit kernel. The 32-bit kernels built from the same source suspend fine. I'm using the same 32-bit userspace for both.
I just tried the vanilla 2.6.33-rc2 kernel and the bug is in there too.
I dont have 32bit userspace, so cannot tell how it works there, but got identical problem with T60p/FireGL V5200(r520) and drm-radeon-testing.
s2ram works w/o kms, with kms always fails(w/o xorg too).
I made patch for r520 yesterday with manny printk's, but not sure how to catch them.
My T60 (X1400, r500) on x86-64 shows the same symptoms.
When going into suspend, the machine goes back to Linux VT with a black screen and blinking cursor in the upper left corner (as usual), but instead of suspending after a second, it stays there with the fan spinning up. Looks like it gets stuck in an endless loop somewhere.
I am currently reinstalling my second Linux machine, so I can soon try to capture some SysRq-Tracebacks via netconsole, if it helps.
I can also confirm this bug.
2.6.33-rc3, x86_64, dri2proto + mesa + libdrm + xf86-video-ati from git, RV630.
I don't have 32-bit userspace.
Created attachment 32603 [details] [review]
Patch from Dave Airlie via irc to fix kernel bug 15035
This must be the same bug as http://bugzilla.kernel.org/show_bug.cgi?id=15035 as the patch provided by airlied on irc to fix that works for me.
Can close this once a proper fix is added to the git tree.
I can confirm this fixes the bug for me too.