[Spice-devel] investigation into spice kms crash

[David Mansfield]

Hi All,

I've been poking around trying to find the cause of an intermittent 
spice kms crash, which I think is the same as these three bugzilla bugs:

https://bugzilla.redhat.com/show_bug.cgi?id=1038275
https://bugzilla.redhat.com/show_bug.cgi?id=1050034
https://bugzilla.redhat.com/show_bug.cgi?id=1039063

and an email I sent on 11 Sep. 2014 (Subject: qxl dri related kernel oops).

The code in question seems to be looping for about 150-300ms (in 
qxl_ttm.c:qxl_sync_obj_wait) and if I've understood the architecture 
correctly, it's waiting for the "host" side of QXL to release some objects.

Also, in qxl_fence.c, there is this comment:

    For some reason every so often qxl hw fails to release, things go wrong.

In particular, my colleague hits this semi-regularly when using spice 
via a mobile-tether: i.e. high jitter, high latency and periodic network 
dropouts.

Does the "qxl hw" have a discard policy when the client is "missing" and 
is the 150-300ms timing loop in the guest based on anything?


-- 
Thanks,
David Mansfield
Cobite, INC.