AMD Rembrandt & AMD Rembrandt-R: Suspend hangs system

[Timo Aaltonen]

[Impact]

On some OEM platforms observed bad suspend occurs on lid close and power LED stays on without normal sleep behavior at that time. Needs to call GFXOFF to the right state during the suspend stage.

https://github.com/torvalds/linux/commit/ca299b4512d4b4f516732a48ce9aa19d91f4473e

Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3132
Fixes: ab4750332dbe

[Test case]

test that s2idle works after installing the update

[Regression potential]

minimal

[Ubuntu Kernel Bot (ubuntu-kernel-bot)]

This bug is awaiting verification that the linux-oem-6.5/6.5.0-1023.24 kernel in -proposed solves the problem. Please test the kernel and update this bug with the results. If the problem is solved, change the tag 'verification-needed-jammy-linux-oem-6.5' to 'verification-done-jammy-linux-oem-6.5'. If the problem still exists, change the tag 'verification-needed-jammy-linux-oem-6.5' to 'verification-failed-jammy-linux-oem-6.5'.

If verification is not done by 5 working days from today, this fix will be dropped from the source code, and this bug will be closed.

See https://wiki.ubuntu.com/Testing/EnableProposed for documentation how to enable and use -proposed. Thank you!

[Launchpad Janitor (janitor)]

This bug was fixed in the package linux-oem-6.5 - 6.5.0-1023.24

---------------
linux-oem-6.5 (6.5.0-1023.24) jammy; urgency=medium

  * jammy/linux-oem-6.5: 6.5.0-1023.24 -proposed tracker (LP: #2063580)

  * Add support for Quectel RM520N-GL modem [1eac:1007] (LP: #2063529)
    - bus: mhi: host: pci_generic: Add support for Quectel RM520N-GL modem
    - bus: mhi: host: pci_generic: Add support for Quectel RM520N-GL Lenovo
      variant

  * S2idle regression (LP: #2064595)
    - drm/amd: Evict resources during PM ops prepare() callback
    - drm/amd: Add concept of running prepare_suspend() sequence for IP blocks
    - drm/amd: Flush GFXOFF requests in prepare stage

  * Add support of TAS2781 amp of audio (LP: #2064064)
    - ALSA: hda/tas2781: Add tas2781 HDA driver
    - ALSA: hda/tas2781: Add tas2781 HDA driver
    - ALSA: hda/tas2781: handle missing EFI calibration data
    - ALSA: hda/tas2781: Add new vendor_id and subsystem_id to support ThinkPad
      ICE-1
    - ALSA: hda/realtek: tas2781: enable subwoofer volume control
    - ALSA: hda/tas2781: leave hda_component in usable state
    - ALSA: hda/tas2781: call cleanup functions only once
    - ALSA: hda/tas2781: do not use regcache
    - [Config] enable TAS2781 amplifier

  * Fix system hang while entering suspend with AMD Navi3x graphics
    (LP: #2063417)
    - drm/amdgpu: skip to program GFXDEC registers for suspend abort
    - drm/amdgpu: Reset dGPU if suspend got aborted
    - SAUCE: drm/amdgpu/mes: fix use-after-free issue

  * Add support for Quectel EM160R-GL modem [1eac:100d] (LP: #2063399)
    - bus: mhi: host: pci_generic: Add support for Quectel EM160R-GL modem

  * RTL8852BE fw security fail then lost WIFI function during suspend/resume
    cycle (LP: #2063096)
    - wifi: rtw89: download firmware with five times retry

  * Fix bluetooth connections with 3.0 device (LP: #2063067)
    - Bluetooth: hci_event: Fix handling of HCI_EV_IO_CAPA_REQUEST

  * Fix the RTL8852CE BT FW Crash based on SER false alarm (LP: #2060904)
    - wifi: rtw89: disable txptctrl IMR to avoid flase alarm
    - SAUCE: wifi: rtw89: pci: correct TX resource checking for PCI DMA channel of
      firmware command

  * Add Cirrus Logic CS35L56 amplifier support (LP: #2062135)
    - ASoC: cs35l56: Patch soft registers to defaults
    - ASoC: cs35l56: Move shared data into a common data structure
    - ASoC: cs35l56: Make cs35l56_system_reset() code more generic
    - ASoC: cs35l56: Convert utility functions to use common data structure
    - ASoC: cs35l56: Move utility functions to shared file
    - ASoC: cs35l56: Move runtime suspend/resume to shared library
    - ASoC: cs35l56: Move cs_dsp init into shared library
    - ASoC: cs35l56: Move part of cs35l56_init() to shared library
    - ASoC: cs35l56: Make common function for control port wait
    - ASoC: cs35l56: Make a common function to shutdown the DSP
    - ALSA: hda: Fix missing header dependencies
    - ALSA: hda/cs35l56: Add driver for Cirrus Logic CS35L56 amplifier
    - ALSA: hda: realtek: Re-work CS35L41 fixups to re-use for other amps
    - ALSA: hda/realtek: Add quirks for HP G11 Laptops using CS35L56
    - ALSA: hda: cs35l56: Add ACPI ...

[Timo Aaltonen (tjaalton)]

included in 6.8.5

[Alessandro Astone (aleasto)]

I can reproduce on 6.8.0-38-generic from noble-proposed

So either this is not fixed in 6.8.5, or https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2069416 which matches exactly what I see is actually a different bug

[Mario Limonciello (superm1)]

Please try 6.8.5 mainline and 6.8.12 mainline. There are binary builds created here:

https://kernel.ubuntu.com/mainline/

[Pierre Paridans (pparidans)]

Like @aleasto, I can still reproduce it on 6.8.0-38-generic from noble-proposed.
I'm wondering too if https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2069416 isn't actually a different bug.

I also tried 6.8.5 mainline, 6.8.12 mainline like recommended by @superm1 and even 6.8.3 mainline, but couldn't reproduce it on neither of those.

[Mario Limonciello (superm1)]

It sure sounds like there are some other patches in Ubuntu's kernel causing problems.

Could you check mainline v6.9 and v6.9.7? I know they backported some ASPM patches. Maybe those cause issues for some of these machines?

If those are fine then what I suggest doing is bisecting Ubuntu's kernel tree, specifically with your test being "echo mem | sudo tee /sys/power/state" to suspend the machine and making sure GFX is otherwise idle.

This will ensure you're not hitting the issue of the race condition mentioned in the description while doing the test.

Here is the Ubuntu kernel tree:

https://git.launchpad.net/~ubuntu-kernel/ubuntu/+source/linux/+git/noble

[Pierre Paridans (pparidans)]

Actually, my laptop is not running Ubuntu currently, but Debian Sid.

For what it's worth, it does use the Debian's linux-image-6.9.7-amd64 package and I don't have the issue either.

[Iain Lane (laney)]

I tried to bisect but I had a hard time. `git bisect --first-parent` is a good trick if you think the problem is introduced in Ubuntu (it excludes the upstream commits and that's really helpful: those are a nightmare because the packaging isn't there).

But still I keep on having random build failures :( :(

[Mario Limonciello (superm1)]

I do have a theory.

Maybe it's bug 2071889

[Pierre Paridans (pparidans)]

Quick update: I did try the new kernel released in noble-proposed (6.8.0-40.40) and the problem looks to be fixed!

Maybe @aleasto can confirm if it's working for him too ?

[Alessandro Astone (aleasto)]

Indeed `6.8.0-40-generic` is working fine for me!