gpu hotplug doesn't work

[athul-krishna-kr]

Device and Software Info:

• Sway Version: sway version 1.9
• Device: Asus Zephyrus G14 GA402RJ
• BIOS: 319
• OS: Arch Linux
• Kernel: 6.8.9-arch1-1
• DGPU: AMD Radeon RX 6700S(amdgpu)

dgpu_add:

• echo 1 | sudo tee /sys/bus/pci/rescan

dgpu_remove:

• echo "0000:03:00.0" | sudo tee /sys/bus/pci/devices/0000:03:00.0/driver/unbind && echo 1 | sudo tee /sys/bus/pci/devices/0000:03:00.0/remove

udev_trigger:

• sudo udevadm trigger --verbose --type=devices --action=remove --subsystem-match=drm --property-match="MINOR=0"

Bug report:

lsof_before shows output of sudo lsof /dev/dri/card* before udev_trigger. After udev_trigger there seems to be one thread(?) using file /dev/dri/card0(see lsof_after).

After dgpu_remove and dgpu_add, dgpu comes backup with different card number(card1) tries to initialize drm backend and fails to initialize egl context. Again removing and adding dgpu, it comes back with different card number(card3) and again fails to initialize egl context with error;

• [ERROR] [wlr] [EGL] command: eglQueryDeviceStringEXT, error: EGL_BAD_PARAMETER (0x300c), message: "eglQueryDeviceStringEXT"
• [ERROR] [wlr] [EGL] command: eglQueryDeviceStringEXT, error: EGL_BAD_PARAMETER (0x300c), message: "eglQueryDeviceStringEXT"
• amdgpu_device_initialize: amdgpu_get_auth (2) failed (-1)
• amdgpu: amdgpu_device_initialize failed.

both times.

Closing sway session and removing and adding dgpu from tty does reset the dgpu card number to 0 or1.
Starting a new sway session with dgpu removed, and then adding dgpu does work. It does show up in sudo lsof /dev/dri/card*.

Log files:

• lsof_before
• lsof_after
• sway_log
• Link: https://gist.github.com/athul-krishna-kr/a6da70cc3a3b1196a3f6e0370387f737

[bl4ckb0ne]

That's a wlroots bug.

It's weird that eglQueryDeviceStringEXT fails with EGL_BAD_PARAMETER, from the spec 1

On failure, NULL is returned.  An EGL_BAD_DEVICE_EXT error is
generated if <device> is not a valid EGLDeviceEXT.  An
EGL_BAD_PARAMETER error is generated if <name> is not one of the
values described above.

Would you be able to get a stacktrace?

[athul-krishna-kr]

How should I get stacktrace? stacktrace of what?

[bl4ckb0ne]

of the value being given to eglQueryDeviceStringEXT

[athul-krishna-kr]

Please give me instructions to get stack trace.

[bl4ckb0ne]

See the build instruction in the repo to build in debug, then start sway in a gdb session from another computer in ssh.

[athul-krishna-kr]

I should add debug flags to only wlroots?