Red Hat Bugzilla – Bug 509428
X fails to start, segfaults
Last modified: 2012-08-08 18:37:58 EDT
Description of problem:
After an update yesterday, X fails to start in the morning with a segfault
at address 0x3c8.
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. Remove xorg.conf
Here's the backtrace:
0: /usr/bin/X(xorg_backtrace+0x28) [0x45f178]
1: /usr/bin/X [0x4627c9]
2: /lib64/libc.so.6 [0x3cba633370]
3: /usr/lib64/xorg/modules/drivers/intel_drv.so [0x7faab0090b36]
4: /usr/bin/X(xf86DeleteScreen+0x80) [0x475da0]
5: /usr/bin/X(InitOutput+0x57a) [0x4707ca]
6: /usr/bin/X [0x42272a]
7: /lib64/libc.so.6(__libc_start_main+0xfd) [0x3cba61ea2d]
8: /usr/bin/X [0x422479]
Segmentation fault at address 0x3c8
Since this is a Havendale box (e.g. not supported), it usually uses VESA.
Obviously yesterday's update switched it from vesa to intel, with the
result above. Actually I welcome this, because vesa gets wrong resolution.
David Airlie told me that Kristian is supposed to fix Auburndale/Havendale,
but krh blogged a few days ago that he had a son. And now this!
Coincidentially, forcing vesa with xorg.conf does not work either,
because the same update broke vesa as well (segfault in pixman).
What a scew-up.
Created attachment 350321 [details]
Updated to xorg-x11-drv-intel-2.8.0-0.3.fc12.x86_64, no more crash at
bogus MMIO access with the offset of VGA I/O register, but no work either.
Now it looks like this:
(EE) open /dev/fb0: No such file or directory
FATAL: Module fbcon not found
(EE) intel(0): No modes.
Evidently it must have KMS to function, but that is missing.
When kernel boots, it stays in text mode. Version:
kernel-2.6.31-0.94.rc4.fc12.x86_64 came out, has changelog from Kristian
about "Add drm-page-flip.patch", but no help to the lack of /dev/fb0.
This may be fixed in kernel-2.6.31-0.125.rc5.git2.fc12.x86_64. The kernel
goes into framebuffer mode on boot. Unfortunately, startx hits bug 515785,
so I cannot tell for sure if X works. I'll close once I know that it works.
Created attachment 356403 [details]
Note that i915 still complains about being unable to read the EDID.
This is possibly because I have a DisplayLink connector. Nonetheless,
the resolution appears correct visually. Maybe the fb preserves the
old modes, I dunno.
I still see some issues, like not restoring display if DPMS strikes,
but that's for a different bug. Closing.