Comment 4 for bug 1579374

Behavior with 4.6.0-040600rc7-generic appears to be even saner than with 4.5.0-050400, although the AMD GPU is still not usable ("DRI_PRIME=1 glxgears" opens a black window).

Only after reporting this bug, I found that 4.5.0-050400 presented issues with sound output in the HDMI port (sound output at a lower sampling rate, maybe half, than it was supposed to). These issues are fixed in 4.6.0-040600rc7-generic.

Also, 4.6.0-040600rc7-generic does not present any errors in dmesg output when trying to use the AMD GPU (again "DRI_PRIME=1 glxgears"):

---[cut]---
[ 251.019823] [drm] PCIE GART of 2048M enabled (table at 0x0000000000040000).
[ 251.258535] [drm] ring test on 0 succeeded in 15 usecs
[ 251.258748] [drm] ring test on 1 succeeded in 14 usecs
[ 251.258787] [drm] ring test on 2 succeeded in 13 usecs
[ 251.258797] [drm] ring test on 3 succeeded in 4 usecs
[ 251.258812] [drm] ring test on 4 succeeded in 2 usecs
[ 251.258818] [drm] ring test on 5 succeeded in 2 usecs
[ 251.258824] [drm] ring test on 6 succeeded in 2 usecs
[ 251.258830] [drm] ring test on 7 succeeded in 2 usecs
[ 251.258838] [drm] ring test on 8 succeeded in 3 usecs
[ 251.258865] [drm] ring test on 9 succeeded in 6 usecs
[ 251.258878] [drm] ring test on 10 succeeded in 6 usecs
---[cut]---

Whereas 4.5.0-050400 used to output the same kind of error message presented by xenial's 4.4 kernel ("*ERROR* amdgpu: cp failed to lock ring N (-2)"), although the NULL pointer dereference was already fixed in 4.5.0-050400.

In short, the NULL pointer dereference is fixed upstream. With upstream kernel, the system appears to run pretty stably using the Intel GPU as the renderer, whereas xenial's 4.4 kernel would constantly crash. However, the AMD GPU is still not usable even with upstream kernel, as it renders empty (black) output.