17:35 <Putti> We think the graphics corruption on i9305 / exynos4412 might be related to another issue we noticed started happening after we updated kernel: the s-boot "Samsung Galaxy" logo gets purple/pink colored before the display gets initialized by kernel.

17:42 <Putti> so some change in kernel that is less than 1 year old might be causing this

19:06 <Putti> disabling IOMMU gets almost totally rid of the graphics corruption with the sboot logo, before it was corrupted for maybe 2 seconds but now it flashes quickly corrupt/purple just some tens of milliseconds before the 4 tux logos show up from the kernel meaning the display driver is initialized.

21:04 <Viciouss> Putti: I gave up on Android for now, focusing on the kernel driver part as it is already a lot to digest there in the time I can put into it. I had hwc.drm.use_overlay_planes=1 active in my configuration and it worked, I noticed some pixels with wrong color on the background but only if no app was open

21:05 <Viciouss> I've never had any issues with the Samsung logo

21:06 <Putti> Viciouss, which device did you use that worked?

21:07 <Viciouss> I used card0 for both hwc and gralloc

21:07 <Putti> Viciouss, also unless you use the lima render node for gralloc then it appears prime is not used

21:07 <Putti> aha, so that's it

21:08 <Putti> the prime stuff was not tested

21:08 <Putti> s/not tested/not used in your case/

21:10 <Putti> I was reading through some commits from icenowy and the prime buffer imports don't seem to work on Allwinner SoC either

21:20 <Putti> actually, now I'm not totally sure if that would not work. Will re-test your setup.

21:36 <freekurt> i'm in over my head with this discussion, but i saw this issue in the glodroid issue tracker about drm_hwcomposer, gbm_gralloc, KMS driver, and card0. https://github.com/GloDroid/glodroid_manifest/issues/31 not sure if it helps, but if it seems at all relevant to Replicant work, maybe it is worth adding a question or comment in that thread about this too since i think they are also working with Mali400.

21:39 <Putti> TBH, I don't know if we should use card or render node for gralloc, maybe the reason why HW plane composition is not done is something else, and card node would be the right one.

21:40 <Putti> everything works fine with card0 except the hw plane compositions don't seem to happen