SMP system hangs with OpenGL
I recently built a dual Athlon box (Tyan Tiger MPX) and my GF4 Ti4200 just hung the system when I tried an OpenGL visualization plugin through XMMS. Damn. Cursor still worked, but everything else was frozen. Anyone else have this problem? I had the same thing years ago on my dual-Celeron box with a PCI Riva TNT card. I figured by now this would be fixed.
System is Red Hat 7.3 with 2.4.18-5 kernel, SMP kernel and GLX drivers.
Any tips would be appreciated. Like is this a kernel problem, or a driver problem? Are there some BIOS settings that might help? I'm running with the 'mem=nopentium' option, by the way. Given some pointers I'll even start trying to fix this myself, I'm that desperate.
It might be that dual athlon stuff isn't that stable yet. It took a long time before normal intel cpu's were stable in SMP mode. Bug report to: firstname.lastname@example.org
Stable in SMP mode?
Maybe, although I haven't seen any other stability issues with this system. It's not overclocked, I didn't 'cheat' by using a pair of Athlon XP processors... so why doesn't this work?
My old dual Intel box could handle neither the xmms visualization plugin nor the OpenGL xscreensaver modes
I ran through a couple levels of Quake3 last night, and that worked fine. So maybe the issue is with the xmms plugin code and not with the NVdriver or the GLX stuff. Frustrating.
Thanks for your reply.
The linux kernel does not yet work stably on Athlons with AGP enabled. You can either:
a) Disable AGP by adding
Option "NvAgp" "0"
to the device section of your XF86Config-4 file, or
b) Patch your kernel with either the experimental patch from AMD. An article on it is here:
And the patch is here:
Great linkage, thanks much. Required reading for anyone with a dual Athlon system + nVidia card.
Just wanted to follow up on the SMP stability issue.
Disabling AGP sounded like a performance hit, so I read through the AMD issue report--which was awesome--and applied the patch to the 2.4.18-10 kernel, which RedHat released today.
So basically what I did was to grab "kernel-source-2.4.18-10" via up2date and apply the patch (see gbrauer's post for the link) to /usr/src/linux-2.4/arch/i386/kernel/setup.c before building a custom kernel. Red Hat's instructions were very clear although I encountered one error compiling the "transparent decompression extension" to the iso9660 filesystem. I disabled this option and was able to make bzImage/modules etc.
I rebooted into that kernel 30 minutes ago and haven't had a crash yet. During this time XMMS has been running that OpenGL plugin that quickly crashed my system under the stock 2.4.18-5 kernel.
Thanks again for the valuable information, hope this helps other people in the same situation.
Glad to be of help... it took me weeks to find that info. However....
I just noticed in the changelog for RH kernel-2.4.18-10 this change:
* Tue Jun 11 2002 Benjamin LaHaise <email@example.com>
- add linux-2.4.19-pageattr.patch to fix the cache attribute conflict problem
I *believe* this is a backport of the AGP speculative caching fix from the 2.5 kernel. The patch I linked to above is a workaround that works by disabling speculative caching. This patch may actually fix the architectural problems in the kernel. You may have "double fixed" your problem by patching RH 2.4.18-10.
AMD claims the workaround patch only adds a 2-3% performance hit, but you may want to try running the stock 2.4.18-10 kernel and see if you're still stable.
can u plz clearly explain how to apply this patch .. preet preety please :-)
You need the kernel sources, obviously. And you have to edit the filename in the patch:
--- linux/arch/i386/kernel/setup.c-pre10 Sun Jun 16 21:34:48 2002
+++ linux/arch/i386/kernel/setup.c Sun Jun 16 21:50:53 2002
I removed the "-pre10" from the first line. Then
# cd /usr/src/linux-2.4/arch/i386/kernel
# patch -p4 < /path/to/patchfile/you/just/edited
Then follow the instructions to recompile a custom kernel (edit the /usr/src/linux-2.4/Makefile and add "Extraversion=<something>" so you don't overwrite your vendor kernel).
gbrauer mentioned the RedHat 2.4.18-10 kernel seems to have a (better?) patch for this, so you may not need this, but you might want to try it just to cover all bases.
What a saga :-)
I went through hell with this, with RHL 7.2 I and
the 2880 it was ok with my dual athlon on a Tyan
S2462. Upograde to RHL 7.3 and 2.4.18-3,4,5 it was hosed. But I am now on 2.4.18-10 kernel and I seem to be ok. I never could make it through one round of Tuxracer before. Time permitting I will dig into this some more, it definelty seems to be a big performance hit with any of the open gl stuff going. ANyway, great thread with a great link.
|All times are GMT -5. The time now is 10:33 AM.|
Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2015, Jelsoft Enterprises Ltd.
Copyright ©1998 - 2014, nV News.