Re: [PATCH] drm/radeon: Avoid double gpu reset by adding a timeout on IB ring tests.

[Bruno Kant <bkant@cloppy.net>]

Le 2016-02-08 15:51, Matthew Dawson a écrit :
>> Yeah, 1s works as well. But that 100ms shouldn't be enough sounds like
>> another bug to me.
> Probably, but it has a tendency to crash my entire system, so I'm not 
> sure how
> to debug it further.  Netconsole maybe?  I probably won't be able to 
> dig into
> this due to time since 1s does work.
> 

Hello,

netconsole won't help if the box gets killed. IP, USB, everything should 
be dead on a hang. I have a netconsole running, but I see nothing on it 
when my box hangs.

I'm now trying to use crash dumps, they could help you also. It dumps a 
core, plus a short text file with the main kernel panic message. If your 
box panics, you would get at least the text file (sort of main cause).

The core dump itself must later be read with kernel debug symbols, this 
can be more difficult to set up. But try kdump, this was easy to set up 
on my Fedora box, look at this tuto:

https://fedoraproject.org/wiki/How_to_use_kdump_to_debug_kernel_crashes

You may also later need a custom built kernel, or a debug kernel (which 
could slow down your box). I have noticed some kernel debug options 
which generate panics for different kind of kernel locks or oops.

Best regards
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel