Comment 5 for bug 174367

@Leann: As I pointed to the original bug report where the commit was discussed, I'd appreciate if you could simply check whether reverting a14a5742c371dc55ca55bfc23bec892dfea26b9b one time more would be the fix. I just git pulled the hardy kernel, and:
% git diff v2.6.24-rc5 drivers/char/agp/intel-agp.c
--- a/drivers/char/agp/intel-agp.c
+++ b/drivers/char/agp/intel-agp.c
@@ -508,6 +508,11 @@ static void intel_i830_init_gtt_entries(void)
                        break;
                }
        } else {
+ /* G33's GTT stolen memory is separate from gfx data
+ * stolen memory.
+ */
+ if (IS_G33)
+ size = 0;
                switch (gmch_ctrl & I855_GMCH_GMS_MASK) {
                case I855_GMCH_GMS_STOLEN_1M:
                        gtt_entries = MB(1) - KB(size);

So the only difference between the Ubuntu tree and the upstream tree on the file which was changed and reverted 4 or 5 times in gutsy is the exact same code -- and I bet upstream has a working G33. I checked the xserver-xorg-video-intel tree and saw that the change in gutsy was not kept over for hardy, and Kyle told me that the change had to be in both the driver and kernel or none of them.

I think reverting a14a5742c371dc55ca55bfc23bec892dfea26b9b is the way to go; could you check whether that makes sense?