re: kern/56714: Radeon acceleration not working with SAPPHIRE AMD FIREPRO 2270 x1

[matthew green <mrg%eterna.com.au@localhost>]

The following reply was made to PR kern/56714; it has been noted by GNATS.

From: matthew green <mrg%eterna.com.au@localhost>
To: gnats-bugs%netbsd.org@localhost
Cc: kern-bug-people%netbsd.org@localhost, gnats-admin%netbsd.org@localhost,
    netbsd-bugs%netbsd.org@localhost
Subject: re: kern/56714: Radeon acceleration not working with SAPPHIRE AMD FIREPRO 2270 x1
Date: Tue, 06 Dec 2022 13:18:55 +1100

 this is also the same problem that triggers PR#57059, which has
 a teardown issue that causes a panic afterwards.
 
 
 > kern error: [drm:(/hurz/src/sys/external/bsd/drm2/dist/drm/radeon/radeon=
 _r600_dma.c:274)r600_dma_ring_test] *ERROR* radeon: ring 3 test failed (0x=
 CAFEDEAD)
 
 we've recently made some progresss in understanding this problem.
 
 it appears to happen on UEFI boots only.  at least 3 people have
 confirmed this fixes (or breaks) their systems.
 
 in confirmed i was able to break it on my ryzen 3600 system by
 turning on uefi.  this systems normally runs with a serial console
 (which is not supported by the uefi on the host, hence using bios
 normally), so as part of the effort to debug this i forced the
 kernel to use serial console using these options:
 
 options         CONS_OVERRIDE
 options         CONSDEVNAME=3D"\"com\""
 options         CONSPEED=3D115200
 options         CONADDR=3D0x3f8
 
 (note that previously, CONS_OVERRIDE was not required.)
 
 however, this kernel boots from UEFI and *does* have working
 radeon acceleration.  eg:
 
 Dec  2 08:11:57 the-blessing /netbsd: [  24.6873761] {drm:netbsd:r600_dma_=
 ring_test+0x22c} *ERROR* radeon: ring 3 test failed (0xCAFEDEAD) after 100=
 000 usecs
 
 vs
 
 Dec  2 08:53:42 the-blessing /netbsd: [ 2057.8790796] [drm] ring test on 3=
  succeeded in 2 usecs
 
 
 hopefully this data point can help isolate the cause.
 
 
 .mrg.