[REGRESSION] 5.18-rc1+ blank screen on AMD RMB with OLED panel

Blank screen when bootup on AMD RMB with OLED panel. 5.17 is good, first bad commit: [1b08dfb889b2c584b444538c9500af24ba0a6dc7] drm/amdgpu: remove gart.ready flag

log attached: 5.19-rc1-amd-blankscreen-gart.ready.dmesg

Hardware description:

• CPU: AMD Ryzen 5 PRO 6650U with Radeon Graphics
• GPU: 62:00.0 VGA compatible controller [0300]: Advanced Micro Devices, Inc. [AMD/ATI] Rembrandt [1002:1681] (rev d2)
• Display(s): SDC ATNA33AA02-0
• Type of Display Connection: eDP OLED

System information:

• Distro name and Version: Ubuntu 22.04
• Kernel version: 5.18-rc1
• Custom kernel: 5.19-rc1

Is this just a default brightness problem? If you SSH in while the blank screen is active, can you check the brightness from sysfs? If you change it (either with sysfs or hotkeys) does it come back?

Also if it's a default brightness problem, perhaps picking up the OLED brightness calculation fix might help.

I don't see why that change would effect the display hardware unless there is something strange going on with display from system memory. @ckoenig for awareness.

Yeah; can we see the bisect log if you still have it? AFAIK there's two other changes (29ba7b16b907a1f915aab8b83ef901e209146938, f9130b81aea2de3fb6d356e9495a384b2d35b1d1) on top of that to the gart touching stuff. If all 3 are reverted on top of 5.19-rc does it "fix it"?

After revert the 3 patches and "f9130b81aea2 drm/amdgpu: drop WARN_ON in amdgpu_gart_bind/unbind"

The screen is blank too. There are many errors in log: revert-gart.ready4.dmesg

Sorry I don't have the bisect log. But I double confirmed that reset to the commit 1b08dfb889b2c584b~ , the screen is OK.

Well that means that your bisect is somehow bogus. Please try to run that again and keep the bisect log around so that we can double check it.

$ git bisect log git bisect start '--' 'drivers/gpu/drm/amd'

bad: [3123109284176b1532874591f7c81f3837bbdc17] Linux 5.18-rc1

git bisect bad 3123109284176b1532874591f7c81f3837bbdc17

good: [f443e374ae131c168a065ea1748feac6b2e76613] Linux 5.17

git bisect good f443e374ae131c168a065ea1748feac6b2e76613

bad: [7f161df1a513e2961f4e3c96a8355c8ce93ad175] drm/amdkfd: replace err by dbg print at svm vram migration

git bisect bad 7f161df1a513e2961f4e3c96a8355c8ce93ad175

bad: [243c719e872a1322b22efccff80776353357b296] drm/amdgpu: handle BACO synchronization with secondary funcs

git bisect bad 243c719e872a1322b22efccff80776353357b296

good: [4efdddbce7c1329f00c458e85dcaf105aebdc0ed] Merge tag 'amd-drm-next-5.17-2022-01-12' of https://gitlab.freedesktop.org/agd5f/linux into drm-next

git bisect good 4efdddbce7c1329f00c458e85dcaf105aebdc0ed

good: [86700a402694db56030a74481d09f35520332736] drm/amdgpu: modify a pair of functions for the pcie port wreg/rreg

git bisect good 86700a402694db56030a74481d09f35520332736

bad: [04022982fc5ddac6cc783d66846f2464fe4985fb] drm/amdgpu: switch to common helper to read bios from rom

git bisect bad 04022982fc5ddac6cc783d66846f2464fe4985fb

bad: [1f2cf08aa010594036ccfb19d207e5b80b5bb7a0] drm/amd/pm: drop unneeded feature->mutex

git bisect bad 1f2cf08aa010594036ccfb19d207e5b80b5bb7a0

bad: [a357dca964e0c77c479075dd65ef86199078d82f] drm/amdgpu: fix the page fault caused by uninitialized variables

git bisect bad a357dca964e0c77c479075dd65ef86199078d82f

bad: [590e86fe3462da81f1cbc4fc8d4cbf8b16b4f968] drm/amdgpu: fix broken debug sdma vram access function

git bisect bad 590e86fe3462da81f1cbc4fc8d4cbf8b16b4f968

good: [33cd016e600ac3417aff7c85f59b9a4b70a947e9] drm/amdgpu: remove unused variable

git bisect good 33cd016e600ac3417aff7c85f59b9a4b70a947e9

bad: [1b08dfb889b2c584b444538c9500af24ba0a6dc7] drm/amdgpu: remove gart.ready flag

git bisect bad 1b08dfb889b2c584b444538c9500af24ba0a6dc7

good: [5904e4135f3b3e6cc7bed46bda71118d55a56681] drm/amdgpu: remove unused variable warning

git bisect good 5904e4135f3b3e6cc7bed46bda71118d55a56681

first bad commit: [1b08dfb889b2c584b444538c9500af24ba0a6dc7] drm/amdgpu: remove gart.ready flag

I bisected again with the same result.

While on a kernel with the "bad" state, if you leave it idle at the login screen a minute or two and then touch the touchpad/keyboard, does display come back? If so; is it possible that your bisect didn't catch immediately?

Additional to that please also don't bisect like this "$ git bisect log git bisect start '--' 'drivers/gpu/drm/amd'"

The problem can also be completely outside of the amdgpu driver, e.g. in the DRM subsystem for example. And your bisect result still doesn't seem to make any sense.

After wait for a while, it cannot be back normal. But screen off/on can bring it back. log attached:amd-rmb-sdcoled-5.18-blank-screen-back.log

the system can be login by ssh on 5.18-r1+, no affects after set the brightness.

I got one of these Rembrandt machines that could repro this and ran a bisect myself. I came up with more sensible results.

$ git bisect log
git bisect start '--term-new=broken' '--term-old=fixed'
# broken: [4b0986a3613c92f4ec1bdc7f60ec66fea135991f] Linux 5.18
git bisect broken 4b0986a3613c92f4ec1bdc7f60ec66fea135991f
# fixed: [f443e374ae131c168a065ea1748feac6b2e76613] Linux 5.17
git bisect fixed f443e374ae131c168a065ea1748feac6b2e76613
# fixed: [25fd2d41b505d0640bdfe67aa77c549de2d3c18a] selftests: kselftest framework: provide "finished" helper
git bisect fixed 25fd2d41b505d0640bdfe67aa77c549de2d3c18a
# broken: [02e2af20f4f9f2aa0c84e9a30a35c02f0fbb7daa] Merge tag 'char-misc-5.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc
git bisect broken 02e2af20f4f9f2aa0c84e9a30a35c02f0fbb7daa
# broken: [9a8b3d5f71eb74b1b95927bd320b1070866a119a] Merge tag 'mips_5.18' of git://git.kernel.org/pub/scm/linux/kernel/git/mips/linux
git bisect broken 9a8b3d5f71eb74b1b95927bd320b1070866a119a
# broken: [54f43c17d681f6d9523fcfaeefc9df77993802e1] Merge tag 'drm-misc-next-2022-02-23' of git://anongit.freedesktop.org/drm/drm-misc into drm-next
git bisect broken 54f43c17d681f6d9523fcfaeefc9df77993802e1
# fixed: [30424ebae8df0f786835e7a31ad790fa00764f35] Merge tag 'drm-intel-gt-next-2022-02-17' of git://anongit.freedesktop.org/drm/drm-intel into drm-intel-n
ext
git bisect fixed 30424ebae8df0f786835e7a31ad790fa00764f35
# fixed: [b65007590911ad558999073dbffa413cf4861df3] drm/amdgpu/display/dc: do blocked MST topology discovery at resume from S3/S4
git bisect fixed b65007590911ad558999073dbffa413cf4861df3
# fixed: [476e4063022787b5720758239ee4c22fa2495e82] drm: implement top-down allocation method
git bisect fixed 476e4063022787b5720758239ee4c22fa2495e82
# broken: [dcb0bd00586bb991eae66a23091e037c2734f8fe] drm/radeon: Fix backlight control on iMac 12,1
git bisect broken dcb0bd00586bb991eae66a23091e037c2734f8fe
# broken: [b06b48d7ddae440f3393e5c1ddda4d2537b3a7cc] amdgpu/pm: Implement emit_clk_levels for navi10
git bisect broken b06b48d7ddae440f3393e5c1ddda4d2537b3a7cc
# broken: [b1c87b0874a4a2045dc78a8d3d3a77ef61536711] drm/amdkfd: use unmap all queues for poison consumption
git bisect broken b1c87b0874a4a2045dc78a8d3d3a77ef61536711
# broken: [b6fba4ecf3554c515aa5354c54dfdf70d7526ff1] drm/amdgpu: reserve the pd while cleaning up PRTs
git bisect broken b6fba4ecf3554c515aa5354c54dfdf70d7526ff1
# broken: [39da460fd4c0f8e7290dcc9cbfc9375de9d0eeca] drm/amd/display: Fix DP LT sequence on EQ fail
git bisect broken 39da460fd4c0f8e7290dcc9cbfc9375de9d0eeca
# fixed: [98ea24e6c563241814abcb1d22ec207a7ec2fd9b] drm/amd/display: change fastboot timing validation
git bisect fixed 98ea24e6c563241814abcb1d22ec207a7ec2fd9b
# broken: [559e2655220d56f939a9fbc6212992345025392c] drm/amd/display: keep eDP Vdd on when eDP stream is already enabled
git bisect broken 559e2655220d56f939a9fbc6212992345025392c
# first broken commit: [559e2655220d56f939a9fbc6212992345025392c] drm/amd/display: keep eDP Vdd on when eDP stream is already enabled

Can you have a try with this revert? 0001-drm-amd-Revert-drm-amd-display-keep-eDP-Vdd-on-when-.patch

Ah, yes that looks much more reasonable. @hwentland, @kazlaus can you guys take a look here as well? Thanks in advance.

I did bisect again, the same as @superm1 . Reverted patch559e2655220d56f9 did make it work.

Thanks for confirming. It seems odd to me though, the patch is trying to keep VDD on. I experimented with some other modifications where it explicitly tried to ignore that panel was already on and backlight already on and try to force it, but I couldn't get proper result still.

So I posted the revert https://patchwork.freedesktop.org/patch/489661/ if that ends up being the right direction to go for this issue.

added DC label

mentioned in commit agd5f/linux@65c3bb54

mentioned in commit agd5f/linux@4dd905fd

mentioned in commit agd5f/linux@937e24b7

It's been reverted for 5.19-rc4: https://github.com/torvalds/linux/commit/937e24b7f5595566a64e0f856ebab9147f2e4d1b

closed

mentioned in commit nouveau@fb688de6