Most recent kernel where this bug did not occur: Unknown, it's happened as long as I've had this machine, a year or so. Distribution: Fedora Core 5 (64bit) Hardware Environment: Single proc AMD64 3500 with 1GBRAM. AGP Radeon 9200SE using Xorg (7.0) radeon driver. Software Environment: Kernel 2.6.18, Xorg 7.0 (latest update from FC 5) Problem Description: When using agp and DRI the X will lock up after a few minutes, Logging in over the network and trying to kill/restart X generally results in the machine hanging. Steps to reproduce: With the above configuration X will lockup within minutes. This problem has been discussed on LKML and has a Xorg bugzilla entry. http://www.ussg.iu.edu/hypermail/linux/kernel/0609.2/1706.html https://bugs.freedesktop.org/show_bug.cgi?id=6111 Today I tried the BusType PCI workaround and that seems to have stopped the hangs. I'll attach my current working xorg.conf and kernel .config If I was to comment out the Option "BusType" "PCI" I would get the hangs. Really just putting this here to help track this issue... Cheers, Andrew
Created attachment 9171 [details] working xorg.conf
Created attachment 9172 [details] 2.6.18 kernel config
Can you get http://www.codemonkey.org.uk/projects/agp/testgart.c and run that as root, and paste the output ?
Sure, here you go. version: 0.101 bridge id: 0x2821106 agp_mode: 0x1f000a09 aper_base: 0xe4000000 aper_size: 64 pg_total: 240896 pg_system: 240896 pg_used: 0 Setting up MTRR at base=0xe4000000 for 64MB. entry.key : 0 entry.key : 1 Allocated 8 megs of GART memory MemoryBenchmark: 1064 MB/s MemoryBenchmark: 1008 MB/s MemoryBenchmark: 1080 MB/s Average speed: 1050 MB/s Testing data integrity (1st pass): passed on first pass. Testing data integrity (2nd pass): passed on second pass.
looks sane, punting to DRI
Created attachment 9173 [details] MSI K8T Master2-Far (VIA K8T8000) testgart output (dual Opteron 244s)
$ cat /proc/mtrr reg00: base=0x00000000 ( 0MB), size=2048MB: write-back, count=1 reg01: base=0x7ff00000 (2047MB), size= 1MB: uncachable, count=1 reg02: base=0xf8000000 (3968MB), size= 32MB: write-combining, count=1 reg03: base=0xe8000000 (3712MB), size= 128MB: write-combining, count=2
The debian 2.6.18 kernel gave me a nice message yesterday stating that the AGP aperture was too small. So, I went into the BIOS and made the aperture 64MB. X now works with DRI over AGP. Yay. -s
Dave, is there something we can go here which is a bit more user-friendly than just hanging up? How come Debian manages to print that message and we don't?
I've revisited this. Although a few things have changed since I last tested, including updated (from Fedora Core 6 updates) libdrm, Mesa, xorg-x11-server-Xorg, xorg-x11-drv-ati, the kernel and my card has gone from a 9200 SE to a 9200 PRO. Unfortunately I can't remember exactly when I last tested and which of the above components would have been in use. The good news is that the current setup (with the AGP aperture size at 64MB, the same as before), seems to be stable. AGP/DRI/Compiz, the 2.6.20-rc6 kernel seemed OK and am currently testing the 2.6.19-1.2895.fc6 Fedora kernel (as that's what most of the workstations are using around here).
OK, thanks. Can you please close off the bug if/when you think that's appropriate?
Sure, I'll give it to the end of the day.
OK this seems stable.