Red Hat Bugzilla – Bug 1008760
Starting with kernel-PAE-3.12.0-0.rc0.git24.1.fc21.i686 using radeon.agpmode=-1 blocks kms which blocks X
Last modified: 2013-09-29 23:13:04 EDT
Description of problem: When booting with kernel-PAE-3.12.0-0.rc0.git25.1.fc21.i686 or kernel-PAE-3.12.0-0.rc0.git26.2.fc21.i686 (nodebug repo) the text size does not switch during the boot process as is normal and neither kdm nor lxdm will start a graphical session. When I boot using kernel-PAE-3.12.0-0.rc0.git22.2.fc21.i686 things work as expected. I tried downgrading dracut and rebuilding the initramfs for git26, but that didn't help. My video card is an rv280 based card. The attached monitor doesn't do EDID properly and I an xorg.conf file to set a nice mode. I also have an x86_64 system with an rv530 based card that runs kernel-3.12.0-0.rc0.git26.2.fc21.x86_64 without any video related issues.
The x86_64 system is really f20 with a nodebug rawhide kernel, the i686 system is a normal rawhide system.
I am still seeing this with kernel-PAE-3.12.0-0.rc1.git0.1.fc21.i686. I am using systemd-207-3 now, but that doesn't seem to make a difference either way.
The problem existed with kernel-PAE-3.12.0-0.rc0.git24.1.fc21. git23 failed to build in koji, so I can test that one. The problem still exists with kernel-PAE-3.12.0-0.rc1.git2.1.fc21. I'm going to attach some logs from X and kdm that might shed a bit more light on things.
Created attachment 799724 [details] kdm log output
Created attachment 799725 [details] X log output
In comment 3, that should have been that I can't test the git23 build.
It looks like this is related to using radeon.agpmode=-1 as a boot parameter. I need this because of bug 662312, which seems to still affect me.
When I booted without radeon.agpmode=-1 I didn't have the problem with kms not working (blocking X from started), but shortly after logging into to xfce, all I could do was move the mouse pointer around which was a symptom of bug 662312. I hadn't tested that bug in quite a while, but it does still seem to be a problem.
I ran git whatchanged d5d04bb..3711d86 which I think should list the changes between git22 and git24 and I didn't see any that looked like radeon graphics driver changes. There were some PCI changes and maybe one of those broken using the radeon driver in PCI mode as a side effect.
I filed an upstream bug for this (https://bugzilla.kernel.org/show_bug.cgi?id=61811).
Another error message that seems to be related that is showing up in dmesg output is "radeon: `-1' invalid for parameter `agpmode'". Also, after a bit of learning, I am working at doing a bisect to find the commit that changed the behavior. If I don't run into any problems, I'll probably be done on Monday.
Turns out this is an argument parsing bug, not a video driver bug. At least one other person reported a similar problem with a totally different type of device. I tested a proposed fix against 3.12-rc2 and it works. I'll close this once I see the fix show up in Fedora. I think we can let it flow in normally unless other people start having trouble as well. (The issue was with a change from strict_strto*() to kstrto*() that had some mistakes in it. The fix was posted on a powerpc list. You can look at the upstream bug for more info.)
Commit 3a126f85e015701e56240884f27f97543580d5f7, which fixes this, is now in Linus' tree.
This is fixed in kernel-PAE-3.12.0-0.rc2.git4.1.fc21.i686.