Crash/black screen: after Aquamarine update with AMD iGPU (Ryzen AI 9 370HX)

[marvinkurock]

Hey, this is a followup discussion for this issue, as it still exists.
#8308

I have a laptop with dual gpu setup and a Ryzen AI 9 370hx. I have the nvidia gpu disabled so the only card visible to linux is iGPU from amd though.

When I try to start Hyperland via tty I get a crash with the error that looks resolution related. It can't figure out what resolution to use (I think). See attached crash report.

When I set a supported resolution via the config ([email protected]) I don't get a crash, gut I do get a black screen.
However, if I connect an external display I do get a hyprland session. I did this and attach the regular logs and the trace enabled logs.
I tried a few resolutions but all give me a black screen on the internal display.

If I do "AQ_NO_ATOMIC=1" I can a image on the internal display aswell. But I need to disable hardware cursor and it seems to get very slow overtime. But this way it works.

Hope this helps to get a lead on the issue.
Thank you very much in advance

hyprland.log
hyprlandCrashReport2239.txt
hyprland_trace.log
systeminfo.txt

[ottersome]

Try the following monitor specification:

monitor=eDP-1,[email protected],auto-down,1.25

You might have to change the eDP-1 part.

I am still having the problem I reported in #8308 btw. Still need to use AQ_NO_ATOMIC=1 to avoid the crash at startup.

[vaxerski]

can you get a trace log?

[marvinkurock]

I did attach a tracelog with HYPRLAND_TRACE=1 AQ_TRACE=1 as hyprland_trace.log, where I started the session with an external monitor and forced resolution which blacks out. Or in what combination you need a trace log?

[vaxerski]

oops, my bad, didnt notice

[vaxerski]

[AQ] atomic drm request: failed to commit: Cannot allocate memory

mm, this looks like a driver issue to me tbh :/

[marvinkurock]

mh, that sucks... we are using the mesa drivers for the "new" AMD Ryzen AI 9 370HX. I do have the black screen issue/atomic comit error with hyprland and sway. The others work fine for me:

• x11
• gnome (wayland)
• cosmic epoch
  not sure how to fix it. Would love to use hyprland though

[vaxerski]

they might be falling back to the legacy api, which we don't support in aq
(in the case of xorg, I'm pretty sure it doesn't even support AMS so it uses legacy anyways)

[m-bartlett]

Based on the hardware specs and using kernel g14 I'm guessing this is a ASUS Zephyrus? I recently got a G16 with the AMD CPU and had the exact same problem when trying to use the AMD iGPU, namely this sort of error in the logs before crash:

[LOG] [AQ] atomic drm request: failed to commit: Cannot allocate memory, flags: ATOMIC_ALLOW_MODESET ATOMIC_TEST_ONLY 
[ERR] Monitor eDP-1 has NO FALLBACK MODES, and an INVALID one was requested: [email protected]

Also was observing that KDE Plasma and Cosmic work, and sway does not.

I just tried @ottersome's suggestion and it looks like env = AQ_NO_ATOMIC,1 fixed it (along with forcing the AMD DRM device first). The docs say this is "HEAVILY NOT RECOMMENDED" so is there a risk to stability long term with this?

[m-bartlett]

Whoops yeah I meant it's working now without AQ_NO_ATOMIC.

Ok couldn´t let it go. Its working now. The nvidia gpu didnt go to sleep coz of waybar and the backlight module. removed it and now gpu does to sleep as expected.

Last issue I have, sleep does not work well with hyprland. When it does to sleep or sometimes when I exit hyprland to gdm the display goes balck and the fans speed up. Other than that, works like a charm... finally

I'm also experiencing this. May warrant a separate discussion thread.

[ottersome]

+1 Confirm. Works on 6.15.5-arch1-1

[marvinkurock]

I found out for me its working after suspend now. The only thing is the internal display is not getting a image. It flickeres once, then black. When I, blindly, open a terminal an type reboot it reboots. I connected an external monitor and got my desktop back.
I once saw it set a resolution not supported to the internal display: 1920x1200@60
Not sure why, cause all of my config sets the native or preferred. Other than that its quite good now

[m-bartlett]

For anyone else getting a blank screen after suspend or dpms off I noticed that changing my monitor line to have bitdepth, 10, vrr, 1, cm, hdr seems to have resolved it and I can toggle dpms and shut the laptop lid with the screen restoring when turned back on.
i.e. I made this change:

- monitor=eDP-1, [email protected], 0x0, 1.25
+ monitor=eDP-1, [email protected], 0x0, 1.25, bitdepth, 10, vrr, 1, cm, hdr

Haven't tested yet which exact setting is the fix, but this seems to be working so far.

[marvinkurock]

great find. works for me as well

[m-bartlett]

Anyone else having to use AQ_NO_ATOMIC=1 again after recent updates? I tried downgrading back to hyprland 0.49.0 and the previous aquamarine, and different Linux kernels, all to no success. My last guess is maybe it's the mesa updates. The errors in the debug logs appear to be the same as before

My OS info

6.15.7-arch1-1

Hyprland 0.50.0 built from branch  at commit c4a4c341568944bd4fb9cd503558b2de602c0213 dirty (version: bump to 0.50.0).
Date: Wed Jul 16 03:33:42 2025
Tag: v0.50.0, commits: 1
built against:
 aquamarine 0.9.2
 hyprlang 0.6.3
 hyprutils 0.8.1
 hyprcursor 0.1.12
 hyprgraphics 0.1.5


no flags were set

[m-bartlett]

Actually it was the aquamarine update, I didn't notice it defaults to installing to /usr/local so hyprland wasn't actually using the downgraded version with my earlier tests.

@vaxerski I did a git bisect and found that DRM errors return on hyprwm/aquamarine@8b22613

[ottersome]

Any one noticed the behavior reappearing after the latest linux update ? I am on 6.15.5-arch1-1-g14.

[ailelix]

On 6.15.8-arch1-1-g14, AQ_NO_ATOMIC=1, AQ_DRM_DEVICES and cursor.no_haerdware_cursor=1 is still required to run hyprland

[m-bartlett]

I'm able to run without AQ_NO_ATOMIC with aquamarine downgraded to 0.9.1, on kernel 6.15.8-arch1-1

[ottersome]

Can confirm. Downgraded to hyprland 0.49.0-3 and aquamarine 0.9.1-1 and it works now. Thanks @m-bartlett

[mizma]

Hi, I'm on AMD 6800U platform and experiencing similar issue with updating:

• hyprutils (0.8.1-1 -> 0.8.2-1)
• aquamarine (0.9.1-1 -> 0.9.2-1)
• hyprlang (0.6.3-2 -> 0.6.4-1)
• hyprcursor (0.1.12-4 -> 0.1.13-1)
• hyprgraphics (0.1.4-2 -> 0.1.5-2)
• hyprland (0.49.0-3 -> 0.50.1-1)

on linux 6.15.8-arch1-1.

• hyprland.log
• hyprlandCrashReport2608.txt
• coredump.log
• dmesg.log

I couldn't see if this is the same one as the issue discussed here since the ERR lines in crash report does look slightly different. Similar crash happens if I stay on the working version of hyprland and update linux kernel to 6.15.9-arch1-1, so I'm stuck not being to upgrade the kernel either. Is this the same with others here?

Sorry if this comment is out of place I'm new to hyprland discussions.

Thanks

[m-bartlett]

While we wait for a proper patch for this, I've just been building aquamarine manually with the offending commit reverted to be able to keep using recent versions of hyprland without AQ_NO_ATOMIC:

version="${1:-}"
git clone --recursive https://github.com/hyprwm/aquamarine ${version:+-b $version}
cd aquamarine
git revert --no-commit 8b22613953961cbea21a620d0559c7ea27dc7e15
cmake --no-warn-unused-cli \
    -DCMAKE_BUILD_TYPE:STRING=Release \
    -DCMAKE_INSTALL_PREFIX:PATH=/usr \
    -S . -B ./build
cmake --build ./build --config Release --target all -j$(nproc) \
&& { sudo rm -v /usr/lib/libaquamarine.so* ; sudo cmake --install ./build ; }

[m-bartlett]

Doing a bit more research into Aquamarine and the offending commit, I was able to narrow down the issue further to this add() call in Aquamarine::CDRMAtomicRequest::addConnector:

https://github.com/hyprwm/aquamarine/blob/11406cc23afeaaef1ee3ccffb9614e67e2458e29/src/backend/drm/impl/Atomic.cpp#L151-L152

For some reason, AQ isn't detecting that the display on this model is 10-bit and instead defaults to 8-bit color, which is apparently the source of the issue since if I replace getMaxBPC(data.mainFB->buffer->dmabuf().format) with a hard-coded 10 or comment out this add() call entirely, I can run hyprland without any issues or needing any extra environment variables.

Haven't solved what's going wrong in the pixel format detection logic, but I'm continuing to investigate, e.g. https://github.com/hyprwm/aquamarine/blob/11406cc23afeaaef1ee3ccffb9614e67e2458e29/src/allocator/GBM.cpp#L17

[szaboboldizsarmark]

Can confirm, I have hardcoded it to 10 as well and works perfectly! I'm on the same Zephyrus laptop.

I'm really looking towards a permanent fix ❤️

[mizma]

hmm, my 6800U platform (GPD WIN MAX 2) doesn't seem to boot even with the above (reverting commit 8b22613953...) nor AQ_NO_ATOMIC=1. Maybe I should try recreating issue again and report separately...

[enessmr]

Ooh different CPU and prob very new 👀👀👀

Likely kernel driver related tho. The kernel is trying to keep up uptodate. Maybe will be fixed normally in Kernel 6.17 or Mesa 26/27

[mizma]

so there are some driver related implementation that is needed by aquamarine that is still in the works in amd drivers? I couldn't really figure out where to search for amd drm driver issues so I'm not really informed on whats going on over in the drivers development.

[enessmr]

I also have no idea. Probably... DRM isnt installed? (Though it should come with mesa) But try emulating it

[ailelix]

Any update? It's been mesa 25.2 and linux 6.17 and the issue still exists

[ailelix]

Sorry to be off-topic: niri starts to work after installing mesa-git 26.0.0, hyprland remains the same