drm-fbsd11.2-kmod-4.11g20191204 painful to look at

[Scott Bennett]

Scott Bennett <bennett at sdf.org> wrote:

> Niclas Zeising <zeising+freebsd at daemonic.se> wrote:
>
> > On 2019-12-11 11:42, Scott Bennett wrote:
> > > Jan Beich <jbeich at FreeBSD.org> wrote:
> > > 
> > >> Scott Bennett <bennett at sdf.org> writes:
> > >>
> > >>>       I just upgraded from drm-fbsd11.2-kmod-4.11g20190906, which worked
> > >>> fairly well, to drm-fbsd11.2-kmod-4.11g20190906,
> > >>
> > >> You've typed drm-fbsd11.2-kmod-4.11g20190906 twice. Which version worked fine?
> > > 
> > >       Oops!  Actually, it was a copy-and-paste error.  The September version
> > > worked fine.  The December version shown in the Subject: header
> > > (drm-fbsd11.2-kmod-4.11g20191204) is the new one that is giving me a headache.
> > >>
> > >>> which flashes, jumps, at flickers streaks across the screen every few
> > >                            ^^and
> > >>> seconds.  It is really awful.  I don't think I can sit and look at it
> > >>> for very long at a time.
> > >>
> > >> Do you use modesetting DDX or xf86-video-intel? If the latter then the
> > > 
> > >       Neither.  I use radeonkms loaded by kld_list="/boot/modules/radeonkms.ko"
> > > in /etc/rc.conf.local.  The GPU is a Radeon HD 5770.
> > > 
> >
> > Hi!
> > The latest version of the drm-fbsd11.2-kmod update only contained 
> > changes for i915kms.ko.  What other updates did you do at the same time?
>
>      I upgraded the base system from r355127 to r355521.
>      However, I now think I may have sounded an alarm too soon.  After posting that
> note, I went to sleep for a few hours.  When I woke and returned to the machine, it
> took me four or five minutes to notice--I was still half asleep--that the image on
> the screen was no longer behaving in the previous, offensive manner.  I have also
> seen a number of crashes since then, and I now believe the machine is having some
> hardware stability problems.
>      Some of the crashes have left messages in /var/log/messages like these.
>
> drmn0: ring 0 stalled for more than 10011msec                                                                                                                                                                     
> drmn0: GPU lockup (current fence id 0x0000000000029cee last fence id 0x0000000000029d02 on ring 0)                                                                                                                
> drmn0: failed to get a new IB (-11)                                                                                                                                                                               
> [drm:radeon_cs_ib_fill] Failed to get ib !                                                                                                                                                                        
> drmn0: Saved 631 dwords of commands on ring 0.
>
> The above are then followed immediately in /var/log/messages by the FreeBSD copyright
> notices that always appear first in the boot messages.  I do not know whether the GPU
> lockups are due to hardware failures or GPU software crashes, but the system always
> tries to reboot after a few seconds, during which time it is unresponsive.  No crash
> dump is produced in these crashes.
>
     After the last time the driver attempted to reboot the system--it can't, but it can
crash the system--I rebooted it with the module loaded again by the kld_list in
/etc/rc.conf.local entry shown above.  Whereas the driver would randomly try to reboot
the system in less than two days if X11 were running (sometimes in three hours or so),
this time it has remained up and running for 18 days, *but* I did not start X11 this
time and have been using the machine in vt(9) mode instead.
     In another message that I don't seem to be able to find, I stated that I thought
maybe there were a hardware problem.  Now I'm baffled.  Can someone tell me why the
error messages shown above would be issued by the driver?  Why would the GPU hang if
X11 is running, but work fine if X11 is not running?

					Scott Bennett