Bug 2532947 - amd-gpu-firmware-20260910 ships a regressed yellow_carp DMCUB (0x0400004C -> 0x0400004A); PSP rejects it, display stays black on Rembrandt/Radeon 680M
Summary: amd-gpu-firmware-20260910 ships a regressed yellow_carp DMCUB (0x0400004C -> ...
Keywords:
Status: CLOSED ERRATA
Alias: None
Product: Fedora
Classification: Fedora
Component: linux-firmware
Version: 44
Hardware: x86_64
OS: Unspecified
unspecified
high
Target Milestone: ---
Assignee: David Woodhouse
QA Contact: Fedora Extras Quality Assurance
URL:
Whiteboard:
: 2533250 2533325 2533632 2533809 2536023 (view as bug list)
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2026-09-14 07:53 UTC by undercover_black_hat
Modified: 2026-09-19 15:47 UTC (History)
19 users (show)

Fixed In Version: linux-firmware-20260916-1.fc45 linux-firmware-20260916-1.fc44 linux-firmware-20260916-1.fc43
Clone Of:
Environment:
Last Closed: 2026-09-18 00:18:07 UTC
Type: Bug
Embargoed:


Attachments (Terms of Use)


Links
System ID Private Priority Status Summary Last Updated
Gitlab kernel-firmware linux-firmware merge_requests 1241 0 None merged Partially revert "amdgpu: DMCUB updates for various ASICs" 2026-09-14 20:07:45 UTC
Launchpad.net ubuntu/+source/linux-firmware-amd-graphics/+bug/2167279 0 None None None 2026-09-14 20:07:18 UTC
freedesktop.org Gitlab drm/amd/-/work_items/5803 0 None None None 2026-09-14 20:07:18 UTC

Description undercover_black_hat 2026-09-14 07:53:29 UTC
## Description

The `amd-gpu-firmware` update from `20260810-1.fc44` to `20260910-1.fc44` ships a
**lower** DMCUB microcode version for YELLOW_CARP (DCN 3.1.2) than the package it
replaces:

| Package                      | `yellow_carp_dmcub.bin` ucode_version |
|------------------------------|---------------------------------------|
| amd-gpu-firmware-20260309-1  | `0x04000047`                          |
| amd-gpu-firmware-20260810-1  | `0x0400004C`  (works)                 |
| amd-gpu-firmware-20260910-1  | `0x0400004A`  (**regression**)        |

Verified by decompressing the shipped blob and reading `ucode_version` at offset
0x10 of the common firmware header:

    # amd-gpu-firmware-20260910-1.fc44 (installed)
    00000010  4a 00 00 04 10 97 03 00  00 01 00 00 1a fc f8 2a  |J..............*|
    sha256: 5cdaaa2c159b0742d38cb655311550a5c4dfefe9bdf12c0fc02d8f14fce781a2

    # amd-gpu-firmware-20260810-1.fc44 (previous)
    00000010  4c 00 00 04 10 99 03 00  00 01 00 00 79 00 f3 aa  |L...........y...|
    sha256: 3f871acbdb93e5c9aac03805abd597c59bd220388f3e37dc9ccc2cf21c884b6f

The PSP refuses to authenticate the 0x0400004A image, DMUB never starts, and the
eDP link to the internal panel fails to train. The machine boots completely
(networking, gdm, and the GNOME Shell Wayland greeter all start normally) but the
panel stays dark, so it presents to the user as a boot failure.

## Actual results

    amdgpu 0000:06:00.0: [drm] Loading DMUB firmware via PSP: version=0x0400004A
    amdgpu 0000:06:00.0: failed to load ucode DMCUB(0x3F)
    amdgpu 0000:06:00.0: psp gfx command LOAD_IP_FW(0x6) failed and response status is (0xFFFF0008)
    amdgpu 0000:06:00.0: [drm] Wait for DMUB auto-load failed: 3
    amdgpu 0000:06:00.0: [drm] *ERROR* Error getting DMUB FW meta info: 4
    amdgpu 0000:06:00.0: [drm] *ERROR* dc_dmub_srv_log_diagnostic_data: DMCUB error - collecting diagnostic data
      ... (repeats 126-231 times per boot) ...
    amdgpu 0000:06:00.0: [drm] enabling link 0 failed: 15

## Expected results

DMUB loads and the internal panel lights up, as with 20260810-1:

    amdgpu 0000:06:00.0: [drm] Loading DMUB firmware via PSP: version=0x0400004C
    amdgpu 0000:06:00.0: [drm] DMUB hardware initialized: version=0x0400004C

(zero DMCUB errors, no link-enable failure)

## Steps to Reproduce

1. Rembrandt / Radeon 680M system (DCN 3.1.2, YELLOW_CARP) running
   amd-gpu-firmware-20260810-1.fc44.
2. `sudo dnf update` -> amd-gpu-firmware-20260910-1.fc44 installed, initramfs regenerated.
3. Reboot. Panel stays black; the system is otherwise up and reachable.

## Hardware

- ASUSTeK ROG Zephyrus G15 GA503RS, BIOS GA503RS.317 (02/27/2024)
- AMD Ryzen 9 6900HS with Radeon Graphics
- iGPU: Rembrandt [Radeon 680M] [1002:1681] rev c7, subsystem [1043:13dd] (amdgpu)
- dGPU: NVIDIA GA104M [RTX 3080 Mobile] [10de:249c] (nvidia, blacklisted nouveau/nova_core)
- UEFI boot, Secure Boot disabled

## Additional info

**This is not kernel-specific.** It reproduces purely from the firmware change;
the same regression is reported on Arch and CachyOS across unrelated kernels:

- https://bbs.archlinux.org/viewtopic.php?pid=2309459#p2309459
  (independently reports the identical version pair: 20260910-1 = 0x0400004A broken,
  20260810-2 = 0x0400004C working)
- https://discuss.cachyos.org/t/regression-linux-firmware-amdgpu-20260910-1-dmcub-fails-to-load-on-amd-radeon-680m-rembrandt-causing-slow-boot-and-visual-glitches/35623

Other hardware reported affected in those threads: Radeon 9070 XT, 9060 XT, 7700S,
and Raphael. Severity varies -- some users see only slow boot and visual glitches,
others (as here) a completely dark panel.

**The regression can be masked on the previously-installed kernel.** An older kernel
whose initramfs predates the firmware update still carries the 0x0400004C blob and
boots fine, which makes this look like a kernel regression rather than a firmware
one. Regenerating that initramfs (`dracut -f`) breaks the older kernel too.

**Workaround** (avoids downgrading the whole version-locked firmware package set,
which would also revert amd-ucode-firmware/CPU microcode):

    sudo install -D -m 0644 yellow_carp_dmcub.bin.xz \
      /lib/firmware/updates/amdgpu/yellow_carp_dmcub.bin.xz   # from 20260810-1
    sudo restorecon -Rv /lib/firmware/updates/amdgpu
    sudo dracut -f --regenerate-all

**Suggested fix:** revert `yellow_carp_dmcub.bin` to the 0x0400004C build from
20260810, or rebase to a DMCUB release >= 0x0400004C.

Comment 1 Peter Robinson 2026-09-14 09:57:14 UTC
Mario: is this a known issue? Seemingly caused by upstream commit ed94b3559

Comment 2 Kushagra Kumar 2026-09-14 12:15:39 UTC
Confirming identical regression on a second OEM and APU model (MSI Thin A15 with Ryzen 5 7535HS / Radeon 660M):

Hardware:
- Model: MSI Thin A15 B7UCX
- BIOS: E16RKAMS.10E
- CPU: AMD Ryzen 5 7535HS with Radeon Graphics
- iGPU: Rembrandt [Radeon 660M] [1002:1681] rev 0b, subsystem [1462:1437]
- dGPU: NVIDIA GeForce RTX 2050 Mobile [10de:25ad] rev a1, subsystem [1462:1437]
- Kernel: 7.2.5-200.fc44.x86_64

Symptoms:
Immediately upon updating to amd-gpu-firmware-20260910-1.fc44 and rebooting, the internal eDP display remains completely black with no Plymouth splash and no GDM output.

dmesg shows the exact same PSP rejection and DMCUB load failure:
amdgpu 0000:05:00.0: [drm] Loading DMUB firmware via PSP: version=0x0400004A
amdgpu 0000:05:00.0: [drm] *ERROR* Error getting DMUB FW meta info: 4
amdgpu 0000:05:00.0: failed to load ucode DMCUB(0x3F) 
amdgpu 0000:05:00.0: psp gfx command LOAD_IP_FW(0x6) failed and response status is (0xFFFF0008)
amdgpu 0000:05:00.0: [drm] *ERROR* Error queueing DMUB command: status=2

Workaround verified:
Copying yellow_carp_dmcub.bin.xz (version 0x0400004C, SHA256: 3f871acbdb93e5c9aac03805abd597c59bd220388f3e37dc9ccc2cf21c884b6f) from amd-gpu-firmware-20260810-1.fc44 back into /usr/lib/firmware/amdgpu/ and rebuilding initramfs with dracut completely restores display functionality on kernel 7.2.5-200.fc44.x86_64.

Comment 3 Mario Limonciello 2026-09-14 13:56:08 UTC
Sorry folks of the issue.  But thanks for gathering all the details and bisecting down to the specific commit that caused it.  I've sent up the revert:

https://gitlab.com/kernel-firmware/linux-firmware/-/merge_requests/1241

We'll investigate internally what went wrong with this one.

Comment 4 Peter Robinson 2026-09-14 14:04:49 UTC
We'll land this in Fedora with the next tagged upstream release, in the meantime it's likely easiest for the users hitting this to downgrade.

Comment 5 Peter Robinson 2026-09-14 18:06:37 UTC
*** Bug 2533250 has been marked as a duplicate of this bug. ***

Comment 6 Peter Robinson 2026-09-15 09:47:20 UTC
*** Bug 2533632 has been marked as a duplicate of this bug. ***

Comment 7 Peter Robinson 2026-09-15 11:00:08 UTC
*** Bug 2533325 has been marked as a duplicate of this bug. ***

Comment 8 Peter Robinson 2026-09-15 14:26:51 UTC
*** Bug 2533809 has been marked as a duplicate of this bug. ***

Comment 9 Christopher Klooz 2026-09-15 14:30:36 UTC
*** Bug 2533809 has been marked as a duplicate of this bug. ***

Comment 10 gui97cham 2026-09-16 14:45:24 UTC
(In reply to Peter Robinson from comment #4)
> We'll land this in Fedora with the next tagged upstream release, in the
> meantime it's likely easiest for the users hitting this to downgrade.

Sorry for the advocacy, however may I ask you to reconsider this decision and push a (dummy/revert) upgrade out of schedule ? Diagnosing then downgrading a broken system is far from easy unless you already know how to read logs and manually manage firmware. This is not a regular minor inconvenience to the average user, and with automatic upgrades the more you wait the more users will upgrade to the broken firmware.

Comment 11 Peter Robinson 2026-09-16 14:50:34 UTC
We are working with various vendors to ensure all the fixes are in place, and with upstream, ultimately I don't want to make the problem worse, and we have 1000s of firmwares in this package that affect different users in different ways.

Comment 12 Fedora Update System 2026-09-16 17:16:25 UTC
FEDORA-2026-91cb8c4497 (linux-firmware-20260916-1.fc44) has been submitted as an update to Fedora 44.
https://bodhi.fedoraproject.org/updates/FEDORA-2026-91cb8c4497

Comment 13 Fedora Update System 2026-09-16 17:16:28 UTC
FEDORA-2026-e671ef50bd (linux-firmware-20260916-1.fc45) has been submitted as an update to Fedora 45.
https://bodhi.fedoraproject.org/updates/FEDORA-2026-e671ef50bd

Comment 14 Fedora Update System 2026-09-16 17:16:30 UTC
FEDORA-2026-6568e9177a (linux-firmware-20260916-1.fc43) has been submitted as an update to Fedora 43.
https://bodhi.fedoraproject.org/updates/FEDORA-2026-6568e9177a

Comment 15 Fedora Update System 2026-09-17 01:34:42 UTC
FEDORA-2026-e671ef50bd has been pushed to the Fedora 45 testing repository.
Soon you'll be able to install the update with the following command:
`sudo dnf upgrade --enablerepo=updates-testing --refresh --advisory=FEDORA-2026-e671ef50bd`
You can provide feedback for this update here: https://bodhi.fedoraproject.org/updates/FEDORA-2026-e671ef50bd

See also https://fedoraproject.org/wiki/QA:Updates_Testing for more information on how to test updates.

Comment 16 Peter Robinson 2026-09-17 13:53:34 UTC
*** Bug 2536023 has been marked as a duplicate of this bug. ***

Comment 17 Fedora Update System 2026-09-17 18:14:36 UTC
FEDORA-2026-91cb8c4497 has been pushed to the Fedora 44 testing repository.
Soon you'll be able to install the update with the following command:
`sudo dnf upgrade --enablerepo=updates-testing --refresh --advisory=FEDORA-2026-91cb8c4497`
You can provide feedback for this update here: https://bodhi.fedoraproject.org/updates/FEDORA-2026-91cb8c4497

See also https://fedoraproject.org/wiki/QA:Updates_Testing for more information on how to test updates.

Comment 18 Fedora Update System 2026-09-17 18:38:54 UTC
FEDORA-2026-6568e9177a has been pushed to the Fedora 43 testing repository.
Soon you'll be able to install the update with the following command:
`sudo dnf upgrade --enablerepo=updates-testing --refresh --advisory=FEDORA-2026-6568e9177a`
You can provide feedback for this update here: https://bodhi.fedoraproject.org/updates/FEDORA-2026-6568e9177a

See also https://fedoraproject.org/wiki/QA:Updates_Testing for more information on how to test updates.

Comment 19 Fedora Update System 2026-09-18 00:18:07 UTC
FEDORA-2026-e671ef50bd (linux-firmware-20260916-1.fc45) has been pushed to the Fedora 45 stable repository.
If problem still persists, please make note of it in this bug report.

Comment 20 Fedora Update System 2026-09-18 01:16:27 UTC
FEDORA-2026-91cb8c4497 (linux-firmware-20260916-1.fc44) has been pushed to the Fedora 44 stable repository.
If problem still persists, please make note of it in this bug report.

Comment 21 Vangreen 2026-09-18 05:37:51 UTC
Problem still persists for Ryzen 5 7535hs [Radeon 660M] on Leneovo ThinkBook

Comment 22 Peter Robinson 2026-09-18 06:54:47 UTC
(In reply to Vangreen from comment #21)
> Problem still persists for Ryzen 5 7535hs [Radeon 660M] on Leneovo ThinkBook

Did you regenerate the initrd?

Comment 23 malcmerlyn 2026-09-18 09:43:54 UTC
Hello

I was not very patient and I switched to Ubuntu 26.04.1.

Is this bug fixed on Fedora KDE Plasma Edition 44 ?

Regards
YE

Comment 24 Fedora Update System 2026-09-19 00:58:18 UTC
FEDORA-2026-6568e9177a (linux-firmware-20260916-1.fc43) has been pushed to the Fedora 43 stable repository.
If problem still persists, please make note of it in this bug report.

Comment 25 piotr.kalemba 2026-09-19 05:13:12 UTC
Confirming the fix on a fourth OEM/model: Lenovo ThinkPad E16 Gen 2 (Ryzen 7 7735HS, Radeon 680M, 1002:1681 rev 0a), Fedora 43 KDE, kernel 7.2.5-100.fc43. With linux-firmware-20260916-1.fc43 the DMUB loads ("DMUB hardware initialized: version=0x0400004C"), zero DMCUB errors, and both external displays over USB-C DP Alt Mode are detected again. The shipped yellow_carp_dmcub.bin is byte-identical (sha256 4bcb91d58ba727684845b02a9cdd32f37cb5a9f71bd4b93d4e678094465cda47) to the one from 20260810-1.

Re comment #21: on my system the fix only takes effect after the initramfs is regenerated (sudo dracut -f --regenerate-all) and a reboot - the package upgrade alone left the old blob in the initrd of the running kernel.

Re comment #23: yes, 20260916-1 reached stable for F43/F44/F45 (comments #19, #20, #24).

Comment 26 Lucas 2026-09-19 15:47:43 UTC
I experienced the same DMCUB/DMUB issue on a system with a Radeon 680M (YELLOW_CARP / Rembrandt).

My system was running:

Fedora 44
Kernel: 7.2.5-200.fc44.x86_64
GPU: Radeon 680M (1002:1681)
amd-gpu-firmware-20260916-1.fc44

Even though the installed firmware package was already the fixed 20260916-1 version, dmesg still showed:

amdgpu 0000:04:00.0: [drm] Loading DMUB firmware via PSP: version=0x0400004A
amdgpu 0000:04:00.0: [drm] *ERROR* Error getting DMUB FW meta info: 4
amdgpu 0000:04:00.0: failed to load ucode DMCUB(0x3F)
amdgpu 0000:04:00.0: [drm] Wait for DMUB auto-load failed: 3
amdgpu 0000:04:00.0: [drm] *ERROR* dc_dmub_srv_log_diagnostic_data: DMCUB error - collecting diagnostic data

The installed firmware file in /usr/lib/firmware/amdgpu/yellow_carp_dmcub.bin.xz was already 0x0400004C, but the initramfs still contained an older copy of yellow_carp_dmcub.bin.xz.

I regenerated all initramfs images with:

dracut --regenerate-all --force

After rebooting, the system loaded the correct DMCUB firmware (0x0400004C) and the DMCUB/HDMI errors were gone. HDMI output then worked normally.

So, in my case, the problem persisted because the initramfs was still containing the older 0x0400004A firmware even after the fixed linux-firmware package had been installed.


Note You need to log in before you can comment on or make changes to this bug.