This (In reply to Luke from comment #0)
>=20
> Then the log repeats a few hundred times the non-utf-8 character "=
;\00"
> before it cuts off.
> 

Hello, just some naive comments:

I also get some "^@" character after reboot sometimes, this h=
appen because the
kernel is not functional enough to properly write all messages on disk. One
solution is to read these messages just before it freezes (does it freeze b=
tw
?). For example ssh from another machine and type: "dmesg -w" bef=
ore to start
the test. This will read from the ring buffer directly.

Though the real problem here is "radeon 0000:07:00.0: ring 4 stalled f=
or more
than 10024msec" so it is not really worth to get more messages in this=
 case.
Do you require to hard reboot after these gpu reset failures ? Alternatively
you can try radeon.hard_reset=3D1 as a kernel param to do a pci reset inste=
ad of
a soft reset.

Also maybe you could generate an apitrace (apt-get install apitrace, apitra=
ce
trace /path/to/application ).  Then if you reproduce the issue with apitrace
replay application.trace you can attach it to this bug.

About the amdgpu issue maybe you could try to build this kernel
https://cgit.freedesktop.org/~agd5f/linux/log/?h=3Damd-staging-4.9 =
(set
CONFIG_DRM_AMDGPU_SI=3Dy in the .config) which contains some up to date fix=
es for
amdgpu (and radeon).