Hi. Whilst attempting to resolve what initially appeared to be an ati-drivers issue for another user I determined that the problem revolved around an MTRR allocation failure on the part of the kernel: fglrx: module license 'Proprietary. (C) 2002 - ATI Technologies, Starnberg, GERMANY' taints kernel. [fglrx] Maximum main memory to use for locked dma buffers: 804 MBytes. ACPI: PCI Interrupt 0000:01:00.0[A] -> GSI 16 (level, low) -> IRQ 19 [fglrx] module loaded - fglrx 8.19.10 [Nov 9 2005] on minor 0 mtrr: type mismatch for d0000000,8000000 old: write-back new: write-combining [fglrx:firegl_addmap] *ERROR* mtrr allocation failed (-22) It turns out that he was not the only person to be affected, as evidenced by this forum thread: http://forums.gentoo.org/viewtopic-t-393738.html. I had the affected user apply the patch mentioned there (and in the URL field of this bug) against his gentoo-sources-2.6.14-r2 tree, rebuild and reboot and the problem was resolved. Shaohua Li's post suggests that this issue can occur with processors with "steppings 0F33h(supporting x86) and 0F34h(supporting x86 and EM64T)". In view of this, would this not be a good candidate for genpatches? I'm not sure whether the latest kernel snapshot suffers from this problem but as far as I know this never went it.
Adding lu_zero to the CC field just in case this is an issue that has cropped up more often than not with ATI users and thus, may be of interest.
Created attachment 73473 [details, diff] Shaohua Li's CPUID workaround
Also, the user (phoenixnull) uploaded various information here http://slash.ath.cx/~phoenixnull/. The dmesg.output file shows how things were prior to the patch being applied, and the dmesg.outpu.new file was generated after.
Thank you for the notice
Fixed in gentoo-sources-2.6.14-r3 (genpatches-2.6.14-4) and in linux 2.6.14.3