[Savage40] Savage IX : DRI and APM -> frozen X.org

[Phil Stopford]

Alex Deucher wrote:
> 
> --- Phil Stopford <phil at ldex.terica.net> wrote:
> 
>>Hi there,
>>
>>I'm someone might be able to help with this and perhaps suggest a way
>>to 
>>avoid the problem. Simply put, if I don't enable DRI in xorg.conf, I
>>can 
>>quite happily suspend and resume my IBM T21 without problems (other
>>than 
>>sound, but that's not likely to be relevant here). As soon as DRI is 
>>enabled, though, the resumed system shows a scrambled X session and
>>no 
>>amount of keyboard prodding will get a response from the system. All
>>I 
>>can do is remove the power source. I have tried all available
>>snaphots 
>>from http://dri.freedesktop.org/snapshots/ through to the mid-June 
>>release without finding one that appears to work reliably.
>>
>>DRI does work other than that, showing a 300 fps output for glxgears 
>>compared to ~70 for non-DRI, and is proving useful enough that I'd
>>like 
>>to be able to persuade APM and DRI to cooperate.
>>
>>In connection with this, is there a way to trigger debug output (to a
>>
>>log file) from the DRI system so that crash bugs can be reported more
>>
>>effective? I couldn't find any information about this during a brief 
>>search. Since DRI crash bugs seem to lock up X, getting a logged
>>chunk 
>>of debug output would obviously be handy in reporting failures.
>>
>>Any comments/help would be appreciated.
>>
>>Phil Stopford.
> 
> Generally speaking the DRI and suspend/resume do not work well
> together.  There are several issues with various subsystems.  Not all
> agp drivers, for example, have suspend and resume routines.  If your
> problem is related to agp, ou might be able to get it working by
> forcing the dri to PCI mode.  this will cost you some performance
> however.  
> Option "BusType" "PCI"
> in your xorg.conf
> 
> Does your whole system actually hang or is just is the GPU just wedged?
> You can possibly debug it if you can get a serial console.
> 
> Alex
> 

I have only just arranged for access to another machine to test the 
state of the system after resume. I should know more tomorrow - I simply 
didn't have time today to look into it.

More problematic currently, especially if I cannot access the machine at 
all, I can reproducibly lock the system up with DRI enabled when using a 
specific, reproducible set of operations in a program using OpenGL. In 
software, this same set of operations works without issue. I'd love to 
get a debug log that could be forwarded to the developers, but there 
doesn't appear to be any way to do this. I was expecting to be able to 
set an environment variable to active logging or similar, but 
dri.sourceforge.net hasn't thrown anything up in this respect. This 
makes bug reports very tricky to file.

Phil