Framework Desktop AMD (Strix Halo): Aquamarine/DRM crashes on connector teardown and render path under Hyprland

[johngig1]

System

• Hardware: Framework Desktop AMD Ryzen AI Max 300 Series
• GPU: AMD Strix Halo / Radeon 8050S/8060S (1002:1586)
• OS: Arch Linux / Omarchy
• Hyprland: 0.54.3
• Aquamarine: 0.10.0
• Mesa: 26.0.3
• Kernel tested:
  • linux
  • linux-lts

Symptoms

Frequent Hyprland crashes and earlier full freezes, often while using the mouse / scrolling in Chrome on X.com.

Originally the machine sometimes hard-froze and required reset. After switching to linux-lts and adding:

• amdgpu.gpu_recovery=1
• amdgpu.vm_update_mode=3
• amdgpu.gttsize=8192

the kernel-level symptoms improved, but Hyprland/Aquamarine crashes still occur.

Crash signature 1: Aquamarine connector teardown SIGSEGV

This happened even in safe mode:
Hyprland --watchdog-fd 4 --safe-mode
coredumpctl / gdb shows:

Aquamarine::SDRMConnector::disconnect()
Aquamarine::SDRMConnector::~SDRMConnector()
Aquamarine::CDRMBackend::~CDRMBackend()

and the actual segfault is:
Aquamarine::CLogger::log(this=0x0, ...)

with message:
drm: Not disconnecting connector DP-1 because it's already disconnected

Relevant backtrace excerpt:

#0 Aquamarine::CLogger::log (this=0x0, level=Aquamarine::AQ_LOG_DEBUG, str="drm: Not disconnecting connector DP-1 because it's already disconnected")
#1 Aquamarine::SDRMConnector::disconnect
#2 Aquamarine::SDRMConnector::~SDRMConnector
#10 Aquamarine::CDRMBackend::~CDRMBackend
#24 Aquamarine::CBackend::~CBackend

Crash signature 2: render-path SIGABRT

A newer crash also happens on HDMI, so this is not only DP-specific.

Backtrace:
#0 abort
#5 CHyprOpenGLImpl::begin
#6 CHyprRenderer::beginRender
#7 CHyprRenderer::renderMonitor
#8 CMonitorFrameScheduler::onFrame
#11 libaquamarine.so.9 + 0xad509
#12 drmHandleEvent
#13 libaquamarine.so.9 + 0xba1a8

Hyprland Crash Report:

---

Hyprland Crash Report

Oops

Hyprland received signal 6(ABRT)
Version: 521ece463c4a9d3d128670688a34756805a4328f
Tag: v0.54.3
Date: Fri Mar 27 18:17:50 2026
Flags:

System info:
System name: Linux
Node name: framework1
Release: 6.18.21-1-lts
Version: #1 SMP PREEMPT_DYNAMIC Thu, 02 Apr 2026 15:44:36 +0000

GPU:
c2:00.0 Display controller [0380]: Advanced Micro Devices, Inc. [AMD/ATI] Strix Halo [Radeon Graphics / Radeon 8050S Graphics / Radeon 8060S Graphics] [1002:1586] (rev c1)

os-release:
NAME="Arch Linux"
PRETTY_NAME="Arch Linux"
ID=arch
BUILD_ID=rolling
ANSI_COLOR="38;2;23;147;209"
HOME_URL="https://archlinux.org/"
DOCUMENTATION_URL="https://wiki.archlinux.org/"
SUPPORT_URL="https://bbs.archlinux.org/"
BUG_REPORT_URL="https://gitlab.archlinux.org/groups/archlinux/-/issues"
PRIVACY_POLICY_URL="https://terms.archlinux.org/docs/privacy-policy/"
LOGO=archlinux-logo

Libraries:
Hyprgraphics: built against 0.5.0, system has 0.5.0
Hyprutils: built against 0.12.0, system has 0.12.0
Hyprcursor: built against 0.1.13, system has 0.1.13
Hyprlang: built against 0.6.8, system has 0.6.8
Aquamarine: built against 0.10.0, system has 0.10.0

Backtrace:
# | Hyprland(_Z12getBacktracev+0x61) [0x55c987ab0121]
getBacktrace()
??:?
#1 | Hyprland(_ZN13CrashReporter18createAndSaveCrashEi+0xcd6) [0x55c987a13026]
CrashReporter::createAndSaveCrash(int)
??:?
#2 | Hyprland(+0x284bde) [0x55c987962bde]
std::__format::_Formatting_scanner<std::__format::_Sink_iter, char>::_M_format_arg(unsigned long)
??:?
#3 | /usr/lib/libc.so.6(+0x3e2d0) [0x7f712fb2f2d0]
??
??:0
#4 | /usr/lib/libc.so.6(+0x98a2c) [0x7f712fb89a2c]
??
??:0
#5 | /usr/lib/libc.so.6(gsignal+0x20) [0x7f712fb2f1a0]
??
??:0
#6 | Hyprland(_ZN15CHyprOpenGLImpl5beginEN9Hyprutils6Memory14CSharedPointerI8CMonitorEERKNS0_4Math7CRegionEP12CFramebufferSt8optionalIS6_E+0x918) [0x55c987e511f8]
CHyprOpenGLImpl::begin(Hyprutils::Memory::CSharedPointer, Hyprutils::Math::CRegion const&, CFramebuffer*, std::optionalHyprutils::Math::CRegion)
??:?
#7 | Hyprland(_ZN13CHyprRenderer11beginRenderEN9Hyprutils6Memory14CSharedPointerI8CMonitorEERNS0_4Math7CRegionE11eRenderModeNS2_I9IHLBufferEEP12CFramebufferb+0xbb2) [0x55c987e91692]
CHyprRenderer::beginRender(Hyprutils::Memory::CSharedPointer, Hyprutils::Math::CRegion&, eRenderMode, Hyprutils::Memory::CSharedPointer, CFramebuffer*, bool)
??:?
#8 | Hyprland(_ZN13CHyprRenderer13renderMonitorEN9Hyprutils6Memory14CSharedPointerI8CMonitorEEb+0xc0f) [0x55c987e9252f]
CHyprRenderer::renderMonitor(Hyprutils::Memory::CSharedPointer, bool)
??:?
#9 | Hyprland(_ZN22CMonitorFrameScheduler7onFrameEv+0x270) [0x55c987ae3cf0]
CMonitorFrameScheduler::onFrame()
??:?
#1 | /usr/lib/libhyprutils.so.11(_ZN9Hyprutils6Signal15CSignalListener12emitInternalEPv+0x2d) [0x7f713094436d]
??
??:0
#11 | /usr/lib/libhyprutils.so.11(_ZN9Hyprutils6Signal11CSignalBase12emitInternalEPv+0x2b3) [0x7f71309446e3]
??
??:0
#12 | /usr/lib/libaquamarine.so.9(+0xad509) [0x7f7130a49509]
??
??:0
#13 | /usr/lib/libdrm.so.2(drmHandleEvent+0x130) [0x7f71305820e0]
??
??:0
#14 | /usr/lib/libaquamarine.so.9(+0xba1a8) [0x7f7130a561a8]
??
??:0
#15 | Hyprland(+0x50639a) [0x55c987be439a]
Cursor::CShapeOverrideController::unsetOverride(Cursor::eCursorShapeOverrideGroup)
??:?
#16 | /usr/lib/libwayland-server.so.0(wl_event_loop_dispatch+0x1d2) [0x7f713080b182]
??
??:0
#17 | /usr/lib/libwayland-server.so.0(wl_display_run+0x37) [0x7f713080d297]
??
??:0
#18 | Hyprland(_ZN17CEventLoopManager9enterLoopEv+0x2c1) [0x55c987be6ef1]
CEventLoopManager::enterLoop()
??:?
#19 | Hyprland(main+0x1476) [0x55c9878e4ef6]
main
??:?
#2 | /usr/lib/libc.so.6(+0x276c1) [0x7f712fb186c1]
??
??:0
#21 | /usr/lib/libc.so.6(__libc_start_main+0x89) [0x7f712fb187f9]
??
??:0
#22 | Hyprland(_start+0x25) [0x55c98794d5f5]
_start
??:?

Log tail:
DEBUG from aquamarine ]: udev: new udev change event for card1
DEBUG from aquamarine ]: udev: DRM device card1 changed
DEBUG from aquamarine ]: udev: DRM device card1 change event unrecognized
DEBUG from aquamarine ]: drm: Got a hotplug event for /dev/dri/card1
DEBUG from aquamarine ]: drm: Scanning connectors for /dev/dri/card1
DEBUG from aquamarine ]: drm: Scanning connector id 100
DEBUG from aquamarine ]: drm: Connector id 100 already initialized
DEBUG from aquamarine ]: drm: Connector 100 connection state: 2
DEBUG from aquamarine ]: drm: Scanning connector id 110
DEBUG from aquamarine ]: drm: Connector id 110 already initialized
DEBUG from aquamarine ]: drm: connector HDMI-A-1 crtc is incapable of vrr: props.vrr_capable -> 115, crtc->props.vrr_enabled -> 24
DEBUG from aquamarine ]: drm: Explicit sync supported
DEBUG from aquamarine ]: drm: connector HDMI-A-1 crtc supports CTM
DEBUG from aquamarine ]: drm: connector HDMI-A-1 crtc supports HDR (7)
DEBUG from aquamarine ]: drm: connector HDMI-A-1 crtc supports Colorspace (114)
DEBUG from aquamarine ]: drm: Connector 110 connection state: 1
DEBUG from aquamarine ]: drm: Scanning connector id 117
DEBUG from aquamarine ]: drm: Connector id 117 already initialized
DEBUG from aquamarine ]: drm: Connector 117 connection state: 2
DEBUG from aquamarine ]: drm: Scanning connector id 123
DEBUG from aquamarine ]: drm: Connector id 123 already initialized
DEBUG from aquamarine ]: drm: Connector 123 connection state: 2
DEBUG from aquamarine ]: drm: Scanning connector id 129
DEBUG from aquamarine ]: drm: Connector id 129 already initialized
DEBUG from aquamarine ]: drm: Connector 129 connection state: 2
DEBUG from aquamarine ]: drm: Scanning connector id 135
DEBUG from aquamarine ]: drm: Connector id 135 already initialized
DEBUG from aquamarine ]: drm: Connector 135 connection state: 2
DEBUG from aquamarine ]: drm: Scanning connector id 141
DEBUG from aquamarine ]: drm: Connector id 141 already initialized
DEBUG from aquamarine ]: drm: Connector 141 connection state: 2
DEBUG from aquamarine ]: drm: Scanning connector id 147
DEBUG from aquamarine ]: drm: Connector id 147 already initialized
DEBUG from aquamarine ]: drm: Connector 147 connection state: 2
DEBUG from aquamarine ]: drm: Scanning connector id 153
DEBUG from aquamarine ]: drm: Connector id 153 already initialized
DEBUG from aquamarine ]: drm: Connector 153 connection state: 2
DEBUG from aquamarine ]: drm: Rechecking CRTCs
DEBUG from aquamarine ]: drm: connector DP-1, has crtc 86, will be rechecked
DEBUG from aquamarine ]: drm: Skipping connector HDMI-A-1, has crtc 90 and is connected
DEBUG from aquamarine ]: drm: connector DP-2, has crtc -1, will be rechecked
DEBUG from aquamarine ]: drm: connector DP-3, has crtc -1, will be rechecked
DEBUG from aquamarine ]: drm: connector DP-4, has crtc 86, will be rechecked
DEBUG from aquamarine ]: drm: connector DP-5, has crtc -1, will be rechecked
DEBUG from aquamarine ]: drm: connector DP-6, has crtc -1, will be rechecked
DEBUG from aquamarine ]: drm: connector DP-7, has crtc -1, will be rechecked
DEBUG from aquamarine ]: drm: connector DP-8, has crtc -1, will be rechecked
DEBUG from aquamarine ]: drm: slot 0 crtc 86 unassigned
DEBUG from aquamarine ]: drm: slot 1 crtc 90 taken by HDMI-A-1, skipping
DEBUG from aquamarine ]: drm: slot 2 crtc 94 unassigned
DEBUG from aquamarine ]: drm: slot 3 crtc 98 unassigned
DEBUG from aquamarine ]: drm: Connector DP-1 is not connected, removing old crtc 86
DEBUG from aquamarine ]: drm: Connector DP-2 is not connected
DEBUG from aquamarine ]: drm: Connector DP-3 is not connected
DEBUG from aquamarine ]: drm: Connector DP-4 is not connected, removing old crtc 86
DEBUG from aquamarine ]: drm: Connector DP-5 is not connected
DEBUG from aquamarine ]: drm: Connector DP-6 is not connected
DEBUG from aquamarine ]: drm: Connector DP-7 is not connected
DEBUG from aquamarine ]: drm: Connector DP-8 is not connected
ERR from aquamarine ]: [EGL] Command eglDupNativeFenceFDANDROID errored out with EGL_BAD_PARAMETER (0x12300): eglDupNativeFenceFDANDROID

[johngig1]

More info...

Kernel now shows AMDGPU job timeout and reset path before userspace crashes:

• drm_sched_job_timedout [gpu_sched]
• dcn35_smu_send_msg_with_param warnings
• dc_commit_state_no_check / dm_suspend in call trace
• amdgpu: VPE queue reset failed
• amdgpu: MODE2 reset
• [drm] device wedged, but recovered through reset

After recovery, Hyprland later segfaults in libaquamarine.so.0.10.0.
This suggests a kernel AMD display/DCN/VPE/SMU issue may be the primary trigger, with Aquamarine/Hyprland crashing secondarily.

[4r33x]

Also happens to me on every monitor powerstate change (for example, changing HDR\free sync settings or just unplug\plug-in HDMI)
Env:
aquamarine 0.12.1-1
hyprland 0.55.4-1
7.0.12-arch1-1
AMD Strix halo 395+

[mbeadmin]

Additional data point: same crash on AMD Radeon 780M (RDNA3, Hawk Point, 1002:1900).

Hardware: Acer laptop, Ryzen 7 8845HS, Radeon 780M iGPU (primary renderer) + NVIDIA RTX 4070 Mobile (unused, PRIME offload only)
OS: Arch Linux, kernel 7.1.4-arch1-1
Hyprland: 0.56.0-2 | Aquamarine: 0.13.0-2 | Mesa: 26.1.4 (downgraded from 26.1.5, no change)

Symptoms: Hyprland aborts with SIGABRT every 30-40 minutes during normal desktop use. No specific trigger identified — happens while typing, while idle, while switching windows. Session is unrecoverable; Hyprland restarts into safe mode. Occasionally hyprlock crashes first (in libgallium), taking the compositor with it, but the compositor also crashes independently.

Backtrace (identical to the one in this issue):

#6  assertImpl<...>
#7  Render::GL::CHyprOpenGLImpl::begin(...)
#8  Render::GL::CHyprGLRenderer::beginRenderInternal(...)
#9  Render::IHyprRenderer::beginRender(...)
#10 Render::IHyprRenderer::renderMonitor(...)
#11 Monitor::CMonitorFrameScheduler::onFrame()
#15 libaquamarine.so.12
#16 drmHandleEvent

Things tried, none helped:

• Mesa downgrade 26.1.5 to 26.1.4
• Disabling the hypridle DPMS-off listener
• AQ_NO_MODIFIERS=1 — did not stop the crashes, and broke the external HDMI output entirely plus caused noticeable input lag

Setup: dual monitor, internal eDP-2 (2560x1600@180, scale 1.6) and external HDMI-A-1 (3840x2160@60, scale 1.5) via USB-C dock behind a KVM switch. Crashes occur regardless of whether the external monitor is connected.

Full logs attached.

versions.txt
coredump-backtrace.txt
hyprlandCrashReport13963.txt
systeminfo.txt

[mbeadmin]

Follow-up / correction to my earlier report here — my case turned out NOT to be
an aquamarine bug, so please disregard the log files I attached.

The stack trace was identical to the one in this issue (assertImpl ->
CHyprOpenGLImpl::begin -> beginRenderInternal -> renderMonitor -> onFrame),
but that assertion was a symptom, not the cause. The kernel log showed what
actually happened, a few hundred milliseconds before every single crash:

​ amdgpu 0000:65:00.0: ring gfx_0.0.0 timeout, signaled seq=479736, emitted seq=479738 amdgpu 0000:65:00.0: Process hyprlock pid 80977 thread hyprlock:cs0 pid 80979 amdgpu 0000:65:00.0: MES failed to respond to msg=RESET amdgpu 0000:65:00.0: Ring gfx_0.0.0 reset failed amdgpu 0000:65:00.0: GPU reset begin!. Source: 1 amdgpu 0000:65:00.0: MODE2 reset amdgpu 0000:65:00.0: GPU reset(2) succeeded! ​

The GPU hung and the driver recovered it with a MODE2 reset, which invalidates
the EGL/GL context underneath the compositor. Hyprland then asserts on the next
frame. The named process varied (sometimes hyprlock, sometimes Hyprland) —
it is simply whoever holds the hung queue at that moment.

Hardware: Radeon 780M (gfx1103, Hawk Point / Ryzen 7 8845HS), kernel 7.1.4-arch1-1,
mesa 26.1.4, Hyprland 0.56.0. Note this is different hardware from the Strix Halo
report above, so the match was coincidental.

Root cause was GFXOFF. Disabling it via a kernel parameter fixed it completely:

​ amdgpu.ppfeaturemask=0xfff73fff # default 0xfff7bfff minus bit 0x8000 ​

Several days of uptime since, zero ring timeouts, zero compositor crashes —
where previously it crashed multiple times per day, typically while idle or
locked. Things that did NOT help, in case it saves someone time: mesa downgrade,
disabling the hypridle DPMS listener, re-enabling it, reducing hyprlock's
shadow_passes, AQ_NO_MODIFIERS=1.

If the attached logs are of no use to you, feel free to delete them.