Differences between version 6 and predecessor to the previous major change of RadeonNotes.
Other diffs: Previous Revision, Previous Author, or view the Annotated Edit History
Newer page: | version 6 | Last edited on Wednesday, March 3, 2004 12:03:47 am | by BrendonJones | Revert |
Older page: | version 4 | Last edited on Thursday, January 8, 2004 10:14:20 pm | by SqeakyTopp | Revert |
@@ -51,9 +51,36 @@
</verbatim>
The problem is well documented - basically, you don't have correct AGPGART support. The way I solved this on [LinuxKernel2.6] was to make sure UseInternalAGPGART was set to no in the XF86Config file and rebooting - if at any stage the fglrx AGPGART has loaded, you're shit out of luck. (Ask dmesg(8)).
-If you have an NForce2; use Kernel 2.4.22 or higher for a kernel that supports the Nvidia AGPGART. (You should be using Kernel 2.4.23
or higher anyway!) It's native to 2.6.
+If you have an NForce2; use Kernel 2.4.22 or higher for a kernel that supports the Nvidia AGPGART. (You should be using Kernel 2.4.24
or higher anyway!) It's native to 2.6.
-With the NForce2 chipset there may be diffiulties
trying to have output mirrored to the TV out, unfotunately I do not currently know any solution other than to disable TV out support.
+With the NForce2 chipset there may be difficulties
trying to have output mirrored to the TV out, unfotunately I do not currently know any solution other than to disable TV out support.
The [Gentoo ATI Radeon FAQ|http://odin.prohosting.com/wedge01/gentoo-radeon-faq.html] has some great examples. If you're using a VIA KT400, start at [Rage3D's message board|http://www.rage3d.com/board/showthread.php?s=b63b7ad75931c247abc20583b648f4c4&threadid=33724523&highlight=xf86eNOMEM].
+
+!!Kernel Oops when starting X
+
+With a 2.6.3 kernel I was getting a kernel oops everytime X tried to load with the fglrx module loaded. The backtrace looked something like this:
+<verbatim>
+Feb 20 20:27:10 than kernel: Call Trace:
+Feb 20 20:27:10 than kernel: [<f89f49cb>] __ke_amd_adv_spec_cache_feature+0xb/0x80 [fglrx]
+Feb 20 20:27:10 than kernel: [<f8a031bc>] drm_agp_enable+0x0/0xc4 [fglrx]
+Feb 20 20:27:10 than kernel: [<f89fd2e8>] _r6x_CheckAGPCommand+0x130/0x178 [fglrx]
+Feb 20 20:27:10 than kernel: [<f89f4617>] __ke_agp_enable+0x17/0x40 [fglrx]
+Feb 20 20:27:10 than kernel: [<f8a03240>] drm_agp_enable+0x84/0xc4 [fglrx]
+Feb 20 20:27:10 than kernel: [<f89f513e>] firegl_ioctl+0x166/0x1a4 [fglrx]
+Feb 20 20:27:10 than kernel: [<c015b6d9>] sys_ioctl+0x219/0x270
+Feb 20 20:27:10 than kernel: [<c0108cff>] syscall_call+0x7/0xb
+</verbatim>
+
+A lot of Googling later, the fix that WorksForMe was to find the function __ke_amd_adv_spec_cache_feature in /lib/modules/fglrx/build_mod/2.6.x/firegl_public.c and edit it so that it returned immediately;
+i.e. it should end up looking like this:
+
+<verbatim>
+int __ke_amd_adv_spec_cache_feature(void)
+{
+ return 0;
+}
+</verbatim>
+Save and exit, then cd .. back up to build_mod, run ./make.sh, cd back up to fglrx and run ./make_install.sh.
+Just removing the offending function might not always be a good idea, but doing this lets me use X with hardware acceleration, and I have yet to notice any issues.