08:39 <bentiss> pinchartl: FDO_DISTRIBUTION_EXEC executes in a chroot of the target container, so you can not really get out of that chroot without monkey patching cbuild IMO

08:43 <bentiss> you could make it 3 jobs: 1. build your kernel build container, 2. with that container, build your kernel, export artifacts, 3. integrate your kernel into either the kenrel build container or on a different libcamera container (see FDO_BASE_IMAGE)

10:53 <pinchartl> bentiss: interesting, I hadn't thought about that multi-step setup

10:53 <pinchartl> thanks

17:48 <DavidHeidelberg> lumag: since I have infinite-loop on -517 with a660 + 6.6 kernel; I thought I'll give your patchset a shot and found out that "soc: qcom: pmic: Fix resource leaks in a device_for_each_child_node() loop" ( 5692aeea5bcb9331e956628c3bc8fc9afcc9765d ) is not marked to be backported. I didn't CCed you on it, but your patches apply cleanly with that patch added on 6.6

17:49 <DavidHeidelberg> for me the mentioned patch is just needed to be applied to get your patchset clean, thou I think backporting because it resolves resource leak ofc (not only to make my life easier :P )

18:12 <lumag> DavidHeidelberg, yes, sounds like a good idea

18:13 <lumag> Yes, the transparent bridges patchset was creted exactly to fix these -517 loops.

18:13 <lumag> Then the https://patchwork.freedesktop.org/series/120375/ and https://patchwork.freedesktop.org/series/126046/ should fix the last issue there.

18:16 <DavidHeidelberg> btw. should be freedreno, but sometimes I click on freedreno :D

18:17 <DavidHeidelberg> lumag: what you mean last issue?

18:18 <DavidHeidelberg> s/freedreno/freedesktop/ aaagh!

18:19 <lumag> DavidHeidelberg, there is a possiblity of EPROBE_DEFER loop if e.g. QMP or NB7 is not availabe when drm/msm tries to bind/init. To fix that we need to move get_next_bridge to the probe() time. But it was not really possible until the last patch of Kuogee's patchset.

18:39 <pinchartl> is it acceptable to generate a 500MB rootfs for a DUT as an artifact ? it would not contain the test binaries, which would be compiled as another job, with a much smaller artifact, and would only be generated infrequently, when the rootfs has to change