Bug 1654

Summary: [ATI/radeon] Bus error in radeon_drv on PCI 9200 + Power Mac
Product: xorg Reporter: Ian Romanick <idr>
Component: Driver/RadeonAssignee: Xorg Project Team <xorg-team>
Status: RESOLVED FIXED QA Contact:
Severity: major    
Priority: high    
Version: git   
Hardware: PowerPC   
OS: Linux (All)   
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
The offending config file
none
The resulting log file none

Description Ian Romanick 2004-10-17 00:06:38 UTC
X.org tree from current CVS built with no host.def on Power Mac G4 generates
signal 7 (bus error) during start-up.  The last message in /var/log/Xorg.0.log
is "Module vgahw: ...".  There are no warnings or errors in the log.  Everything
looks pretty normal until the signal.  Just for giggles, I tried disabling DRI,
but that didn't help.

I tried starting in GDB, but could not get a backtrace at all.  I'll rebuild
with "-O0 -gdb3" and see if that helps.

The card in question is a PCI Radeon 9200 128MB.  I removed the working Rage 128
AGP from the system, put the PCI 9200 in, added a device entry, and started
seeing the problem.  That seems to point the finger at something in the radeon
driver.
Comment 1 Ian Romanick 2004-10-19 19:18:14 UTC
Created attachment 1135 [details]
The offending config file
Comment 2 Ian Romanick 2004-10-19 19:19:11 UTC
Created attachment 1136 [details]
The resulting log file
Comment 3 Ian Romanick 2004-10-20 13:58:24 UTC
I talked with Michel Daenzer last night on IRC.  He believes that the problem is
that the MMIO routines for PowerPC are hardcoded for PCI bus ID 1, but my card
is on bus 2.  I tried moving the card to a different slot to no avail.  All 3
PCI slots on the Power Mac give a 0001:2:XX.0 bus ID.

Since this is a known issue, drivers should at least log a message and
gracefully terminate.  Death by bus error isn't very user friendly. :(
Comment 4 Michel Dänzer 2004-10-22 11:08:30 UTC
It's neither MMIO nor driver code - it's the port I/O code used by the vgaHW module.

I'm not sure, but implementing PCI domain support for PPC might fix this?

Workarounds you could try:

* Don't use vgaHW with UseFBDev
* Change the bus number passed to the __NR_pciconfig_iobase syscall in lnx_video.c
Comment 5 Alex Deucher 2004-12-19 14:47:07 UTC
the vgahw part should be fixed in cvs (bug 2064)
Comment 6 Benjamin Herrenschmidt 2005-10-24 16:29:55 UTC
The option "VGAAccess" "No" fixes this and is now enabled by default on ppc, so
unless you can reproduce, I'll mark it fixed

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.