Ooops, ignore my earlier XAA log: I'd done a bunch of troubleshooting and then generated logs in a hurry. I mistyped the option (AppendMethod) and so it was also using EXA.

I discovered another detail about the failure. From a clean reboot, I started with XAA (`xlog-xaa`). This worked. I then tried to start X with EXA, logged in `xlog`. This failed but returned me to a working console. Then I tried to restart X with XAA again (`xlog-xaa2`) and the screen garbled and locked up hard. I had to shell in and reboot.

Hm, this is weird. The patch applied to agpgart in 2.6.27.5 is only related to G45-class hardware, your 845G should not be affected by this. Could you test this again with the libdrm package from extra? The one in testing will not move to extra when we move xorg.

Note that once you get such a lockup, you can't recover from it without rebooting.

Could you attach your xorg.conf?

Attached are my xorg.conf as well as the verbose log from running EXA with libdrm-2.3.1-1-i686

libdrm 2.3.1-1
mesa 7.2-1
intel-dri 7.2-1
xf86-video-intel 2.4.2-2
xorg-server 1.5.3-2

Xorg starts and the desktop initializes, but there are some worrying issues in the log.
(WW) intel(0): Register 0x70024 (PIPEASTAT) changed from 0x80000203 to 0x00000000
(WW) intel(0): PIPEASTAT before: status: FIFO_UNDERRUN VSYNC_INT_STATUS VBLANK_INT_STATUS OREG_UPDATE_STATUS
(WW) intel(0): PIPEASTAT after: status:

This was the first run of X after reboot.

So libdrm 2.4.x is causing these issues? As it's not a requirement for xf86-video-intel 2.4, I'll remove it from testing soon. Until we have GEM in the kernel, I see no reason to switch to the newer libdrm version.

I'm not so sure it's libdrm after all. I was able to trigger the crash with libdrm-2.3.1-1 and kernel26-2.26.27.4-1. I'm going to revert all the xorg packages I upgraded yesterday, and then try to install as few as possible at a time to see when the error starts.

Alright, I did some troubleshooting. First, I backed out upgrades, I downgraded to:
* xorg-server-utils 7.3-1
* xorg-xkb-utils 7.3-1
* kernel26 2.6.27.4-1
* libdrm 2.3.1-1
* libgl 7.0.3-2
* intel-dri 7.0.3
* mesa 7.0.3-1
* xf86-input-evdev 2.0.1-1
* libpciaccess REMOVED
* xf86-video-intel 2.3.2-1
* xorg-server 1.4.2-2

The output of `pacman -Q` is available in pacman-0. After a reboot to pick up the new kernel, a good EXA start was obtained, logged in xlog-0. There's a crash when the server exits, but this appears to be unrelated to the issue of this bug.

I upgraded all packages EXCEPT libdrm and kernel26.
* libgl 7.2-1
* intel-dri 7.2-1
* mesa 7.2-1
* xf86-video-evdev 2.0.7-1
* libpciaccess 0.10.5-1
* xf86-video-intel 2.4.2-2
* xorg-server-utils 7.4-1
* xorg-server 1.5.3-2
* xorg-xkb-utils 7.4-1

A reboot was performed. Packages logged in packages-1. An attempt to start using EXA failed, this is logged in xlog-1. Green show was observed on the display, a horizontal band about 60 pixels high 60% down the screen. I don't know if this is diagnostic of anything, but there you have it. ;)

Another reboot. I have a good XAA start, this is logged in xlog-2. Without rebooting, I tried a second XAA start, this cleanly started and terminated, but an MTRR error was observed. This is logged in xlog-3.

I upgraded the kernel, 2.6.27.5-1, and rebooted. Packages logged to packages-2. I had a GOOD EXA start, logged to xlog-4, but with the same MTRR error on termination. I've logged the loaded modules (lsmod) to lsmod-1 in case there's anything interesting in there.

A second attempt at an EXA start fialed, logged in xlog-5. I rebooted, started in XAA, and am now posting this information.

After filing the previous report, I upgraded libdrm to 2.4.1-1, and restarted.

I had a good XAA start, logged in xlog-6, there was the now familiar MTRR error at the end.

I tried a second XAA start, the cursor came up on a black screen, but the userspace (WM, etc) wouldn't start. I ctrl-alt-backspaced. Log in xlog-7.

Finally, I did a third XAA start. The userspace came up, logged in xlog-8.

Packages dumped in xlog-8.

I restarted, and tried an EXA start. It failed, xlog-9.

It seems that something other than libdrm is responsible for breaking EXA. XAA works with libdrm 2.4.1 from testing. I'll stay with XAA for now, maybe the problem will be fixed upstream.

That MTRR error I've been having appears to have something to do with the kernel's i915 module:
Observed in dmesg from start to termination of X:

[drm] Initialized i915 1.6.0 20060119 on minor 0
X:2168 conflicting memory types e8000000-f0000000 write-combining<->uncached-minus
reserve_memtype failed 0xe8000000-0xf0000000, track write-combining, req write-combining
X:2168 conflicting memory types e8000000-f0000000 write-combining<->uncached-minus
reserve_memtype failed 0xe8000000-0xf0000000, track write-combining, req write-combining
X:2195 freeing invalid memtype e8000000-f0000000
X:2168 conflicting memory types e8000000-f0000000 write-combining<->uncached-minus
reserve_memtype failed 0xe8000000-0xf0000000, track write-combining, req write-combining
X:2196 freeing invalid memtype e8000000-f0000000
X:2168 freeing invalid memtype e8000000-f0000000
mtrr: no MTRR for e8000000,8000000 found

contents of /proc/mtrr:
reg00: base=0x00000000 ( 0MB), size= 512MB: write-back, count=1
reg01: base=0x20000000 ( 512MB), size= 256MB: write-back, count=1
reg02: base=0x2ff00000 ( 767MB), size= 1MB: uncachable, count=1

Is this information useful? I'm going to try setting my BIOS to an 8 MB video buffer and see what effect it has.