KMS selects unsupported video mode, causing blank screen after boot

[Ing. Josua Mayer]

Dear Maintainers and readers,

I have come across what I think is a kernel bug regarding how during 
system boot the video-mode for the console is being selected.
Inspection of logs with debugging enabled show that in my particular 
case kernel selects 4k resolution at 144Hz. The supported maximum of my 
GPU however is 4k at 60Hz only, because DisplayPort 1.4 is available, 
but DSC is not.

I will provide hardware details and software versions below, with 
detailed findings at the end. If you have any ideas how to resolve this, 
I am happy to try a patch and report back.

# Hardware Info:
SolidRun HoneyComb Workstation
- LX2160A (16x Cortex-A72)
- 32GB DDR4-3000 memory
- ACPI
ADATA SX8200 NVME
Radeon Pro WX2100
- DisplayPort 1.4
- 1x full-size port, 2x mini ports
EVE Spectrum ES07D03 display (max. 4k 144Hz)

# Software Info:
openSUSE Tumbleweed
Linux 5.15.7-1-vanilla aarch64
Linux 5.15.7-1-default aarch64
Linux 5.15.5-1-default aarch64
Mesa 21.3

# Work-Arounds:
I *can* get to a functional plasma wayland session by forcing a 
supported video mode for both the console and X (because sddm doesn't do 
wayland):

- kernel cmdline: video=DP-3:3840x2160 at 60

- xorg.conf.d:
Section "Monitor"
    Identifier "DisplayPort-2"
    ModeLine "3840x2160__60.0"  594.00  3840 4016 4104 4400  2160 2168 
2178 2250 +hsync +vsync
    Option "PreferredMode" "3840x2160__60.0"
EndSection

# Details
The issue manifested itself first as soon as Linux takes over the 
graphics card from UEFI. That is both UEFI and grub menus show up fine, 
but then once Linux starts the display turns black - but stays on for a 
while. The backlight makes it shine slightly - till eventually a NO 
SIGNAL message shows up before it enters a standby mode.

poking around /sys showed that Linux recognizes a display is connected, 
and that it knows what modes that one supports:
cat /sys/class/drm/card0-DP-3/status
connected
cat /sys/class/drm/card0-DP-3/modes
3840x2160
...
2560x1440
2560x1080
...

Then I tried to force a video-mode that I expected to work, and booted 
with video=DP-3:3840x2160 at 60 to choose what I believe is the maximum 
supported by this particular GPU. Now, console messages of the boot 
process show up properly, and after a few seconds the display switches 
to the 4k resolution - eventually leading to the console login prompt.
--> This is where I know there is a mode that works, and Linux does not 
pick it by default.

So I enabled debug output for the drm module by booting with 
amdgpu.dcdebugmask=0x4 drm.debug=0x104 <-- I figured that would be all 
the interesting pieces. And sure enough I can see all the modes being 
printed, one being picked - and a rather sane error message:
[   10.186899] [drm:drm_mode_debug_printmodeline [drm]] Modeline 
"3840x2160": 144 1296000 3840 3888 3920 4000 2160 2168 2178 2250 0x48 0x9
[   10.186959] [drm:drm_mode_debug_printmodeline [drm]] Modeline 
"3840x2160": 60 594000 3840 4016 4104 4400 2160 2168 2178 2250 0x48 0x5
[   10.197301] [drm:drm_client_modeset_probe [drm]] desired mode 
3840x2160 set on crtc 51 (0,0)
[   10.242536] [drm:create_validate_stream_for_sink [amdgpu]] Mode 
3840x2160 (clk 1296000) failed DC validation with error 10 (No DP link 
bandwidth)
[   10.244183] [drm:dce112_validate_bandwidth [amdgpu]] 
dce112_validate_bandwidth: Bandwidth validation failed!
--> Clearly some subsystem knows the chosen video mode is not supported 
here. But it never gets around to trying another mode, e.g. the *second* 
mode listed in edid (which would work).

I am attaching to this mail the dmesg produced with
log_level=9 drm.debug=0x11F video=DP-1:d video=DP-2:d
(unused ports forced disabled, and most drm debug messages enabled)

I have also tried swapping the connectors on the card: I tried both DP-3 
(the full-size connector) and DP-1 (one of the two mini connectors) but 
the behaviour is the same.

Further Xorg also auto-selects the 144Hz mode according to its log-file.

Finally I also attach output of the drm_info tool from
https://github.com/ascent12/drm_info

Yours sincerely
Josua Mayer
-------------- next part --------------
A non-text attachment was scrubbed...
Name: dmesg.log
Type: text/x-log
Size: 208891 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20211217/e9ba3bfc/attachment-0002.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: drminfo.log
Type: text/x-log
Size: 29231 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20211217/e9ba3bfc/attachment-0003.bin>