09:28 austriancoder: If someone can review a small spirv change that would be great: !42043
10:27 jani: airlied: sima: I added a manual drm-tip fixup patch to drm-next merge. it appears to bring back the unused "retry:" label in drivers/gpu/drm/amd/amdgpu/amdgpu_hmm.c
11:14 pq: How do I rmmod vkms when it's claimed to be in use?
11:36 jadahl: pq: did you try terminating the process that uses it? lsof should help
11:36 jadahl: iirc that allowed me to unload it when I tried, which on the other hand was a long time ago
12:06 pq: jadahl, I was kinda hoping I didn't have to quit my main desktop :-)
12:07 pq: but now I've rebooted, and configured VKMS into its own seat.
12:08 jadahl: pq: thats surprising; I would have assumed that if you disabled the vkms monitor there would be no reason for the compositor to keep that fd open
12:09 pq: hmm, that I didn't try. How would it keep track of its existence though?
12:10 jadahl: what mutter does is keep a "holder" of it open, and on the next hotplug or reconfiguration, open it on-demand
12:10 pq: I see.
12:10 jadahl: we also have this udev rule so you can have vkms enabled for testing purposes without screwing up your main session: ENV{ID_PATH}=="platform-vkms", TAG+="mutter-device-ignore"
12:12 pq: I just added /etc/udev/rules.d/72-vkms.rules: SUBSYSTEM=="drm", KERNELS=="vkms", ENV{ID_SEAT}="seat-vkms"
12:12 jadahl: sounds like a reasonable solution too
12:13 pq: I'm not sure platform-vkms would match it anymore, though? VKMS has moved to the faux bus.
12:13 jadahl: there is also a ENV{ID_PATH}=="platform-vkms", TAG+="mutter-device-ignore" rule
12:13 jadahl: eh
12:14 jadahl: ENV{DEVPATH}=="/devices/faux/vkms/drm/card[0-9]*", TAG+="mutter-device-ignore"
12:14 pq: I'm on Kwin :-)
12:14 jadahl: yea I know :)
12:15 pq: why are there DE-specific rules?
12:15 jadahl: but TAG+=... to ENV{ID_SEAT}=... for that DEVPATH would work for you too
12:15 jadahl: these are installed by mutter
12:15 jadahl: udev is used for attaching various compositor specific tags then used to apply policy
12:17 pq: I see a slight complication with ID_SEAT, uaccess is not installed since I'm not logged in on that seat.
12:18 pq: maybe that's not a problem...
12:18 zamundaaa[m]: pq: you can set KWIN_DRM_DEVICES to a colon-separated list of paths to KMS nodes to avoid vkms being used
12:18 pq: avoiding on kwin is just half the story, I want it to be picked up by Weston (test suite)
12:19 jadahl: easiest would be to keep the device on teh default seat and set that KWIN_DRM_DEVICES to something then perhaps
12:19 zamundaaa[m]: jadahl: do you have other reasons than vkms to close the KMS node so aggressively? KWin only does it when there's no outputs connected to it anymore, and lazily opening it would be some more effort
12:20 jadahl: zamundaaa[m]: to allow dGPU to enter deeper sleep
12:20 zamundaaa[m]: pq: right, I assumed whatever else you're using vkms for would have similar configuration. The env var is how I run KWin tests on vkms from time to time
12:21 pq: I'm kinda fond of the seat concept.
12:22 pq: KMS nodes might get swapped on reboot
13:23 MrCooper: jadahl: just keeping a DRM FD open doesn't prevent the GPU from sleeping, at least not with sane drivers :)
13:54 jadahl: MrCooper: maybe it was not sleep but memory tied to the drm client
13:56 pq: or maybe it was for nvidia
13:58 jadahl: could be