i've tested the new drm 5.6 branch extensively across a number
of chipsets and drivers.  amdgpu is still problematic, though
it does work some what in some cases now.


[X88] -- tested with rebase to 9.99.88.


ati/amd:

Radeon 9000M [*1] [*10]
Radeon x1010 Silent [*9]
Radeon x1550
Radeon HD 3650
Radeon HD 4350
Radeon HD 4670 [*4] [*10]
Radeon HD 5450 [*12] [X88]
Radeon HD 6450
Radeon HD 7750
Radeon R2 Graphics (A10-7110 APU) [*6] [*7] [*10]
Radeon R7 240 [*6] [*8]
Radeon RX 550 [*2] [*5] [X88]
Radeon Vega 8 [*5] [*11]

sometimes many-glx-gears stutters on some systems, or has tearing.


nvidia:

GeForce GT 440 [*3]
GeForce GT 610
GeForce GTX 550i
GeForce GT 710 [*3]
GeForce GTX 750 [*3] [X88]
Quadro K620 [*3] [X88]


intel:

HD Graphics P630 (kabylake) [*10]



notes:

[*1] defaults to not attaching like prior drm2.  when enabled, the
     console now works properly, but starting X hangs with:

     [ 553.4570729] radeon0: error: ring 0 stalled for more than 10000msec
     [ 553.4570729] radeon0: warn: GPU lockup (current fence id 0x0000000000000001 last fence id 0x0000000000000002 on ring 0)

     and never recovers

[*2] mostly works, sometimes gets wrong-colour pallete type errors,
easily triggered with "full screen" mpv, but can also be triggered
without mpv.  corruption is strange, as the weird-image shows both
active windows that should be underneath the playing movie as well
as the movie itself.  requires restaring X to fix.

[*3] there is some weird scheduling issue.  if i run two glxgears
such that they are displaying on the same place (ie one is hidden),
then they show about best combined performance.  (about 20% more
than a single instance.)  however, if i run 2 - 12 of them at the
same time all being displayed, the total throughput is massively
reduced.  from ~3600 FPS with one client, to ~30 FPS * 2 or ~5 FPS
* 12.  if this is left alone after several minutes, the performance
jumps to "normal", and for the 2 or 12 glxgears case, combines to
~4500 FPS.  playing a movie with this occuring is also quite slow
(but fine alone, or after the slow period.)

[*4] HDMI output works again.  drm1 worked with it, but drm2 crashed
when either booting with hdmi connected to the laptop, or connecting
it afterwards, so this is an older system having bugs fixed again.

[*5] tested with amdgpu only

[*6] tested with radeon and amdgpu

[*7] hangs when switching to amdgpu wsdisplay, no serial available,
     works fine with radeon

[*8] fails during amdgpu startup, triggering amdgpu_irq_disable_all()
     crash with half-attach, works fine with radeon

[*9] mostly works, did hard reset about 3 hours into running 12
     copies of glxgears raw speed, no clues why.

[*10] in a laptop

[*11] ring tests fail during start up, lockdebug panic during the
      following teardown and would be followed by cnopen panic due
      to no wsdisplay attaching.

[*12] .88 rebase kernel crashed.  see
      https://github.com/riastradh/netbsd-src/issues/35