Android (AOSP) booting with kernel 5.10 - display rotation issue

[roberto-sartori-gl]

Hi, I copy here a message I left on a commit, so it does not get lost:

Hi Jami,
I've a question on this. I'm working on getting Android somehow working with a somehow mainline kernel. Currently, Android (AOSP) is booting with kernel 5.10: https://github.com/robertosartori/mainline-manifest
..
I've an issue with the display rotation. I get these warnings while the device is booting:
https://pastebin.com/HsGWu3pr
When the Android hwcomposer kicks in, apparently it isn't able to correctly get the display rotation and the image is upside down on Android (https://android.googlesource.com/platform/external/drm_hwcomposer/+/b7cd811394e2afe1c62a2feb15e16e3948dc3307/compositor/DrmDisplayCompositor.cpp#405).

Jami was able to replicate the stacktrace on postmarketOS, so we can continue the discussion here! :)

[Atemu]

For reference, the commit with comments 4678078#commitcomment-49573872

[JamiKettunen]

@robertosartori I looked into the issue and it appears the orientation property change might not be parsed by userspace because the panel already is partially added. What is strange is that all other panel drivers on mainline implementing customizable orientation also seem to do it the same way, so they should be hitting the same issue; I can attempt reaching the author (from Intel) who added those WARN_ON()'s in the source to begin with to come to a conclusion on what should be done here.

This appears to not concern Linux userspace too much however as the video output is still rotated properly. If on Android you can use /etc/modprobe.d style configs (and assuming drm/msm + panel-samsung-s6e3fa5 are modules), you could try configuring something like this to load the panel driver before msm DRM driver is probed:

softdep msm pre: panel-samsung-s6e3fa5

If this doesn't help, you could just try simply not rotating the panel in the OP5 DTS and instead ask around some Android (mainline/bringup/dev) channels about potentially doing this in DRM HWComposer directly somehow (as I believe this might be the intended approach).

Also if you don't mind could you take a picture (or even screenshot) of the device in Android's about screen showing the kernel version? Would be nice to see some of the progress without clearing disk space to build all of AOSP :)

[roberto-sartori-gl]

Thanks for the analysis.

I looked into the issue and it appears the orientation property change might not be parsed by userspace because the panel already is partially added. What is strange is that all other panel drivers on mainline implementing customizable orientation also seem to do it the same way, so they should be hitting the same issue;

Yes, this was the same conclusion I got. From the commits I saw, it was known that a lot of panel drivers are doing the same, but as you said it is not an issue on Linux apparently.

If this doesn't help, you could just try simply not rotating the panel in the OP5 DTS and instead ask around some Android (mainline/bringup/dev) channels about potentially doing this in DRM HWComposer directly somehow (as I believe this might be the intended approach).

Also if you don't mind could you take a picture (or even screenshot) of the device in Android's about screen showing the kernel version? Would be nice to see some of the progress without clearing disk space to build all of AOSP :)

Merging this two points: https://www.youtube.com/watch?v=CyfeY3d1vQ4&t=2s
I did this video a couple of weeks ago. As you can see, I'm able to get the display mostly rotated from userspace, but there is some issues with Android navigation and status bars. Note that I did not use a lot of hours for this, as my idea was to check if the issue was caused by some underline conditions (e.g. the driver). But as you said, it could be easier to just find a way to do that from Android, so this is the way I'll go from now on.

I can share more updated photos/screenshots of course: https://imgur.com/a/ohU44S4

This appears to not concern Linux userspace too much however as the video output is still rotated properly. If on Android you can use /etc/modprobe.d style configs (and assuming drm/msm + panel-samsung-s6e3fa5 are modules), you could try configuring something like this to load the panel driver before msm DRM driver is probed

I quickly tried to use the panel-samsung-s6e3fa5 as module with no success and I'm not even sure about the order the modules were loaded, but as I said earlier, I did not use a lot of time for this. I'll go back to this issue and get back to you hopefully soon :)

[roberto-sartori-gl]

Sorry for the delay.

https://patchwork.kernel.org/project/linux-mediatek/patch/[email protected]/
This appears to remove the warning on my setup, however the screen is still upside down on Android.
I have also ported my changes over to the 5.11 kernel tree. I now think that this is an issue with the Android HWComposer, so I'll look into that.

I'll keep you posted.

[JamiKettunen]

Cool, didn't know about this patch :) This drm_hwcomposer MR could perhaps be of interest as well: https://gitlab.freedesktop.org/drm-hwcomposer/drm-hwcomposer/-/merge_requests/151
(btw you can reply under comments as well)

[JamiKettunen]

@robertosartori Here's something cool for the OnePlus 5 (cheeseburger): I've written an RMI4 F1A driver which should enable the capacitive touch buttons on Android (it may need a key layout file?)
Feel free to grab the currently top 3 commits from v5.11 tree and test it out! :)

[roberto-sartori-gl]

Hi Jami, thanks, I'll try this soon and let you know :)

[JamiKettunen]

Note that this doesn't enable the OP5's home button however; that is glued in with the fingerprint sensor driver which appears to consist of some blob and a fpc1245 trustzone app, so it didn't work on downstream kernel either without all of the Android services up and running. Enabling just the home button functionality could be doable but will need a bunch of debugging on downstream.

[roberto-sartori-gl]

Yes, that is expected. I'm actually working also on a 4.14 kernel just to update from the 4.4 from Oneplus, using a CAF kernel, and the fingerprint stuff is a mess of blobs :) (https://github.com/robertosartori/kernel_4.14_oneplus_msm8998)
I tested your patch on 4.14 and they works great for the capacitive buttons! :) I'll let you know when I try on mainline, but I'm using the same exact driver on 4.14 so it should work without issues.