Bug#1042818: firmware-amd-graphics: Random display freezes on certain AMD GPUs due to "Error waiting for DMUB idle: status=3"
Gregor Riepl
Version: 20230515-3
Severity: important
Tags: upstream
Forwarded: https://gitlab.freedesktop.org/drm/amd/-/issues/1887
X-Debbugs-Cc: oni...@gmail.com
Dear Maintainer,
The current AMDGPU firmware in Debian has compatibility issues with 6.3+
kernels.
These errors manifest themselves with kernel messages like these:
[ +0.226777] [drm:dc_dmub_srv_wait_idle [amdgpu]] *ERROR* Error waiting for
DMUB idle: status=3
[ +4.020959] [drm:dc_dmub_setup_subvp_dmub_command [amdgpu]] *ERROR* Error
waiting for DMUB idle: status=3
Furthermore, they cause sudden display freezes and even GPU lock-ups that
require power-cycling the system.
It's not clear why these problems occur, but they might have to do with certain
optimizations that AMD had to do to reduce power consumption at idle on RX 7000
series GPUs. See https://gitlab.freedesktop.org/drm/amd/-/issues/2315 for more
information about this issue.
As a temporary workaround, it's possible to avoid the power management
optimizations by reducing the overall pixel clock rate or creating modelines
with a longer blanking delay, as long as the display supports this. For
example, reducing the refresh rate from 120Hz to 60Hz has helped in one case
for me.
Another workaround is to disable optimizations in the affected firmware with
the kernel option drm.vblankoffdelay=0 .
As stated in https://gitlab.freedesktop.org/drm/amd/-/issues/1887#note_1993615
, the problematic firmware changes were reverted in
https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-
firmware.git/commit/?id=d3f66064cf43bd7338a79174bd0ff60c4ecbdf6d , and there
have been several amdgpu firmware commits since.
Please update linux-firmware as soon a release containing the revert or a
permanent fix is available.
Thanks.
-- System Information:
Debian Release: trixie/sid
APT prefers testing
APT policy: (990, 'testing'), (500, 'unstable-debug'), (500, 'testing-debug'), (500, 'stable-debug'), (500, 'proposed-updates-debug'), (300, 'unstable'), (1, 'experimental-debug'), (1, 'experimental')
Architecture: amd64 (x86_64)
Foreign Architectures: i386
Kernel: Linux 6.4.0-1-amd64 (SMP w/24 CPU threads; PREEMPT)
Locale: LANG=en_GB.UTF-8, LC_CTYPE=en_GB.UTF-8 (charmap=UTF-8), LANGUAGE=en_GB
Shell: /bin/sh linked to /usr/bin/dash
Init: systemd (via /run/systemd/system)
LSM: AppArmor: enabled
firmware-amd-graphics depends on no packages.
firmware-amd-graphics recommends no packages.
Versions of packages firmware-amd-graphics suggests:
ii initramfs-tools 0.142
-- no debconf information
Gregor Riepl
Version: 20230515-3
Tags: fixed-upstream, upstream
Followup-For: Bug #1042818
X-Debbugs-Cc: oni...@gmail.com
linux-firmware 20230804 has been released and contains the mentioned reverts
for amdgpu firmware.
This is not a permanent fix of the underlying problem, but it will at least
allow systems to function normally again.
Please update as soon as possible.
This should also fix #1040185, but for that issue, a backport to bookworm may
be required.
Diederik de Haas
Thanks for looking into this, but I have some questions.
On Tuesday, 1 August 2023 12:55:55 CEST Gregor Riepl wrote:
> Forwarded: https://gitlab.freedesktop.org/drm/amd/-/issues/1887
On Saturday, 5 August 2023 10:55:10 CEST Gregor Riepl wrote:
> Version: 20230515-3
> Tags: fixed-upstream, upstream
>
> for amdgpu firmware.
Partially revert "amdgpu: DMCUB updates for DCN 3.1.4 and 3.1.5"
That does indeed seem relevant to the error msgs you posted in this bug.
That commit links to https://gitlab.freedesktop.org/drm/amd/-/issues/2666
which is rather different from the 'Forwarded' URL/issue.
Was the initial Forwarded incorrect or am I missing something?
"This partially reverts commit ade163aaaeae0c1ad20cb3dd8ce878bf61c91b3a" is
also part of that commit msg, but that commit was NOT part of version
20230515, but 20230625 which has not (yet?) been uploaded to Debian.
Which leads me to conclude that you shouldn't have been affected as Debian's
firmware package (version) does not include the commit which caused the issue.
Can you shed some light on this discrepancy?
> This should also fix #1040185, but for that issue, a backport to bookworm
> may be required.
Raphaël Gomès
Followup-For: Bug #1042818
X-Debbugs-Cc: alph...@gmail.com
Dear Maintainer,
I think the most relevant new information is that after installing
20230625-1 from 20230515-3 on kernel version 6.0.0-2, I had the random
freezes mentioned above.
Also, only my DP monitor would output anything, and my wayland compositor
would try to make sense of them in a loop, leading to a very unresponsive
system (freezes, missed keystrokes, repeated keystrokes) and other things
like keyboard settings not being applied because the config reload process
got interrupted.
I tried upgrading to linux 6.5.0-5, which didn't do anything. Still on that
version, I manually installed 20230515-4 (instead of the original -3, but
I don't think this makes a difference) and everything is working fine.
The workaround for drm kernel parameters did not work for me unfortunately.
Thanks,
Raphaël Gomès
-- System Information:
Debian Release: trixie/sid
APT prefers testing
APT policy: (500, 'testing'), (200, 'unstable')
Architecture: amd64 (x86_64)
Kernel: Linux 6.5.0-5-amd64 (SMP w/32 CPU threads; PREEMPT)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8) (ignored: LC_ALL set to en_US.UTF-8), LANGUAGE=en_US:en
Shell: /bin/sh linked to /usr/bin/dash
Init: systemd (via /run/systemd/system)
LSM: AppArmor: enabled
firmware-amd-graphics depends on no packages.
firmware-amd-graphics recommends no packages.
Versions of packages firmware-amd-graphics suggests:
ii initramfs-tools 0.142
-- debconf-show failed