On a TUXEDO Stellaris 16 Gen7 AMD (Ryzen 9 9955HX3D, Raphael iGPU, PMFW
98.26.0) suspend to idle fails on nearly every attempt with firmware-
amd-graphics 20260810-1: the iGPU SMU no longer responds on resume
(amdgpu "SMU: No response msg_reg: b", resume of IP block <smu> failed
-62), the display stays dead and only a hard reset helps. Reproducible
with pm_test=devices, i.e. without platform sleep.

Replacing only the Raphael blobs (dcn_3_1_5_dmcub, psp_13_0_8_*,
vcn_3_1_2, sdma_5_2_6, gc_10_3_6_*) with the versions from linux-
firmware 20240318 fixes it on the same kernel (7.2.7.1); the system
then reaches S0i3. Kernels 7.1.8, 7.1.10 and 7.2.7 are affected alike.

Upstream report with logs and firmware versions:
https://gitlab.freedesktop.org/drm/amd/-/issues/5962

This makes suspend unusable on affected Ryzen 9000HX laptops. Please
consider holding back the affected files or adding a NEWS/known-issue
entry until upstream has a fix.

Control: tag -1 moreinfo
[...]

Upstream versions 20260910 and 20260916 reverted changes to some of
these blobs.  Please test whether Debian version 20260916-1 fixes this
issue.

Ben.

Hi Ben,

thanks for the pointer. 20260916-1 fixes it.

Tested on the same machine and kernel (7.2.7.1), package installed from
sid, no files replaced:
- pm_test=devices: "amdgpu 0000:07:00.0: SMU is resumed successfully!"
- real s2idle, 30 s via rtcwake: clean resume, amd_pmc reports "Last
S0i3 Status: Success",
  42 ms to S0i3, 22.7 s in S0i3, 0.54 s resume. No ALIB/SMU timeouts.

Loaded firmware versions (amdgpu_firmware_info), in case it helps to
narrow down the culprit:

                 20260810-1 (broken)   20260916-1 (works)   linux-
firmware 20240318 (works)
  DMCUB          0x05003300            0x05003500           0x05001c00
  PSP ASD        0x21000115            0x21000118           0x210000c7
  TA HDCP / DTM  0x17000052/1200001e   0x17000053/1200001f
0x1700003c/12000016
  VCN            0x04121015            0x0412100f           0x0311e004
  GC RLC         0x20                  0x1f                 0x1f
  GC MEC / PFP   0x1c / 0x12           0x1c / 0x12          0x14 / 0x0e
  SMU (BIOS)     98.26.0               98.26.0              98.26.0

So MEC/PFP are not involved; the regression in 20260810 is in DMCUB,
the PSP TA, VCN or RLC. Given the
"Revert updates to VCN firmware in 20260810" entry (#1148284), VCN
looks like the likely one. I have not
bisected the single blob.

From my side this bug can be closed as fixed in 20260916-1.

Regards,
Daniel

Am Samstag, dem 10.10.2026 um 18:21 +0200 schrieb Ben Hutchings: