22:24 <digetx_> cyndis: I fixed the performance issue, it was due to a missing RPM autosuspend delay set for 2d/3d and I have RPM enabled for them in grate kernel, it takes 30ms to change voltage on resume

22:26 <digetx_> the EPERM error indeed happens due to the timed out job, which starts to happen after host1x RPM suspending that fails with "host1x_intr_stop cannot stop syncpt intr id=12"

22:27 <digetx_> somehow the syncpt wait list is not empty on suspend, while it should be

22:32 <digetx_> I wish we could remove and restructure all the legacy code first, it's not good

23:27 <digetx_> we/me will need to write up the description of the ideal drm/host1x driver and make a plan how to implement it; I wanted to start a thread on the ML, but perhaps it will be easier to do with a doc, like we did with the uapi

23:33 <digetx_> pushed https://github.com/grate-driver/linux with the working autosuspend for 2d/3d

23:34 <digetx_> you will need to disable CONFIG_DRM_TEGRA=n and enable CONFIG_DRM_TEGRA_ORIG=y

23:37 <digetx_> it starts to fail after display power down, i.e. when all host1x children devices are rpm-suspended and host1x fails to suspend properly

23:38 <digetx_> lxde was turning off display after a minute of inactivity, hence that's why it was starting to fail after awhile

23:39 <digetx_> I haven't checked yet what's wrong with the syncpt wait list, please take a look if you will have time

23:45 <digetx_> and why host1x driver uses IRQF_SHARED, terrible, will be so much work to clean it up, omg :)