Most recent kernel where this bug did not occur: None. This happens in distribution, suspend2 and vanilla kernel, with or without preemption. I am currently using suspend2, since it seems it doesn't matter. I can test it in vanilla kernel should it be necessary. Distribution: Gentoo Linux 2006.1 Hardware Environment: Gigabyte GA-8IPE775-G (BIOS Version F7, latest from http://www.gigabyte.de/Support/Motherboard/BIOS_Model.aspx?ProductID=1808) Graphics Card: Connect 3D Radeon 9500 Pro lspci -vv follows (see attachment) Software Environment: Xorg 7.1, xorg-server-1.1.1-r1, xf86-video-ati-6.6.3 Problem Description: I've compiled http://www.codemonkey.org.uk/projects/agp/testgart.c and executed it. A following `startx' will crash the system (monitors switching off), and I have to press the reset button. If testgart is executed in a xserver session, nothing bad will happen until I want to exit the xserver (same symptoms). It would not bother me if I had no other problems with X, but when using AGP mode (that is, not using `Bustype PCI' option) together with GLX the system freezes randomly (only mouse moves, cannot ssh in). `Bustype PCI' and GLX work fine, and AGP works as long as GLX module is not loaded. Both ATI proprietory and ATI open source (r300) drivers show this freezing problem. Normally, I would classify this as a xserver problem, but after dealing with testgart, I wonder whether there is anything wrong with AGP kernel implementation for my board? Steps to reproduce: Execute testgart, start x server. I'll provide logs and hardware information as attachments.
Created attachment 9413 [details] lspci -vv PCI listing.
Created attachment 9414 [details] MTRRs before and after testgart MTRRs before and after testgart.
Created attachment 9415 [details] testgart output Testgart output.
Created attachment 9416 [details] Xorg log file Xorg.0.log file (using AGP, no GLX).
Created attachment 9417 [details] xorg.conf Xorg configuration file.
Created attachment 9418 [details] dmesg
Created attachment 9419 [details] CPU Info
Created attachment 9420 [details] Kernel config file Kernel configuration file.
hmm, I wonder if this has something to do with this.. http://lists.freedesktop.org/archives/xorg-commit/2006-November/009330.html
hrm, then again, that's for onboard video, whereas this bug is for an add-in ATI card.
I thought testgart might change something regarding the agp interface and does not set it back correctly, but I have no idea how to test this (and additionally, not enough knowledge how to write a program which can do that). Maybe I'll try a nvidia card to find out whether this is a specific graphics card problem or not.
BTW: The issue in comment #9 refers to an i965 chipset while my motherboard has an i865 (without onboard). I don't know if it is of importance, but the mtrr allocation during x server usage is (currently) a bit different: reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1 reg01: base=0xe0000000 (3584MB), size= 128MB: write-combining, count=2
This is after starting agp with glx module enabled: reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1 reg01: base=0xe0000000 (3584MB), size= 128MB: write-combining, count=3 reg02: base=0xd8000000 (3456MB), size= 128MB: write-combining, count=1 ...and after exiting the x server: reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1 reg01: base=0xe0000000 (3584MB), size= 128MB: write-combining, count=1 reg02: base=0xd8000000 (3456MB), size= 128MB: write-combining, count=1
Just for completeness' sake: Starting X with AGP/nodri again: reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1 reg01: base=0xe0000000 (3584MB), size= 128MB: write-combining, count=2 reg02: base=0xd8000000 (3456MB), size= 128MB: write-combining, count=1 Starting X with PCI/GLX: reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1 reg01: base=0xe0000000 (3584MB), size= 128MB: write-combining, count=2 reg02: base=0xd8000000 (3456MB), size= 128MB: write-combining, count=1 After exiting X with PCI/GLX: reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1 reg01: base=0xe0000000 (3584MB), size= 128MB: write-combining, count=1 reg02: base=0xd8000000 (3456MB), size= 128MB: write-combining, count=1 X with PCI/no GLX: reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1 reg01: base=0xe0000000 (3584MB), size= 128MB: write-combining, count=1 reg02: base=0xd8000000 (3456MB), size= 128MB: write-combining, count=1 After exiting X with PCI/no GLX: reg00: base=0x00000000 ( 0MB), size=1024MB: write-back, count=1 reg01: base=0xe0000000 (3584MB), size= 128MB: write-combining, count=1 reg02: base=0xd8000000 (3456MB), size= 128MB: write-combining, count=1 I don't think it helps much, though.
As to comment #14 >> Starting X with AGP/nodri again: should be AGP/no GLX, as I have DRI always enabled.
X also doesn't start (after running testgart) using `BusType PCI' and disabling DRI, GLX and Composite extension. So might it not be related to AGP at all? But then, testgart doesn't use anything else than AGP, does it?
Seems to have been a hardware failure. Thanks for your help.