James Cloos cloos at
Fri Aug 28 19:25:09 PDT 2009

>>>>> "Roland" == Roland Scheidegger <sroland at> writes:

Roland> Would the attached patch help?

Roland> diff --git a/src/radeon_textured_videofuncs.c b/src/radeon_textured_videofuncs.c
Roland> -	BEGIN_ACCEL(23);
Roland> +	BEGIN_ACCEL_RELOC(23, 3);
Roland> -	BEGIN_ACCEL(9);
Roland> +	BEGIN_ACCEL_RELOC(9, 1);
Roland> -	BEGIN_ACCEL(36);
Roland> -	BEGIN_ACCEL(24);
Roland> +	BEGIN_ACCEL_RELOC(24, 1);

I'm now at git master of everything as of earlier today (tag v2.6.31-rc8
for the kernel, 39dfac41 for video-ati, 19d6fadf for libdrm) with the
above patch added to video-ati.

XVideo is still not usable, but the dmesg errors which prompted said
patch are indeed gone.

I'm starting xcompmgr -a from my .xinitrc, and so am not seeing
corruption elsewhere; but that may still be due to xcompmgr....

I can say that the X server is *much* better behaved when using
xcompmgr -a on top of KMS than w/o KMS.  During my last boot the server
was frequently unresponsive, using 99% CPU.  I didn't see that w/o
xcompmgr and have not seen it with KMS.

IOW, MKS makes xcompmgr actually usable on this hardware.

Because of the bug where glxinfo caused the server to quit, I now have
glx disabled in my xorg.conf.  I've also not tried switching to and from
the text consoles; I don't know whether those two bugs are fixed.

It is also nice that X is using substantially less VM and RSS than w/o KMS.

James Cloos <cloos at>         OpenPGP: 1024D/ED7DAEA6

More information about the xorg-driver-ati mailing list