GUI acceleration support

[PabloPL]

We can use:

• g2d (there are two drivers for it, one in media directory but old. New is in drm directory, but it don't have support for our soc and also would require changes in libdrm)
• powervr sgx540

[xc-racer99]

Looking into G2D, while there are is a libdrm component, this doesn't appear to be used by anything (checked weston, x11 arm-soc, kwin) so I've abandoned my preliminary attempts at adding support for the v3.0 g2d used in the s5pv210.

For the future, I've a few notes for the libdrm side of things at https://gist.github.com/xc-racer99/6c5466d820ae69090189adf249b75d99 and have created the branch at https://github.com/xc-racer99/linux/tree/g2d-wip Notably, it doesn't support the dma insertion that the supported v4.1 does (libdrm creates a list of commands for the g2d, kernel driver writes them to a dma address; writes dma address to a g2d register; starts g2d which works through the command list). We would need to copy all of the commands to their registers; then run the g2d and rinse and repeat for each section of the command list received from libdrm.

[xc-racer99]

With regards to powervr, something else to try would be Galaxy Nexus/Tuna Android blobs. They can be found at xc-racer99/proprietary_vendor_samsung@ac9e2d3 and need a matching kernel driver patch like xc-racer99/android_kernel_samsung_aries@0471749#diff-9a0ae2f209e5e27649c5b796c42e3c5e

The only thing not working with them on s5pc110 on Android is some colour conversion used by hardware decoding. They may or may not properly expose the configs.

Failing that, we could try the DDK 1.9 blobs for OMAP4 and see if they work.

Additionally, we could try recompiling the hwcomposer module so that it uses /dev/fb{x} as opposed to /dev/graphics/fb{x}. Or we could simply make it a dummy hwcomposer that always returns the same gralloc usage flags which it pretty much does right now anyways :)

[PabloPL]

BTW: I wonder if that android logger module is working (didn't saw devices created in /dev, wasn't able to use logcat).

[xc-racer99]

Probably the Android logger isn't used as we were using a 7.1 based system image for libhybris, it should be using pstore instead (is this enabled in the defconfig?).

I backported the Android logger driver so I could try booting KitKat which is the android version on my phone as it is too old for the pstore version.

I should try setting up libhybris again (I accidentally corrupted my SD card and had to reset it).

[PabloPL]

Ok, so we can revert that commit and make pstore work (defconfig + dts change needed).
Libhybris setup is easy - some time ago when i looked at this, it required to rebuild android device image two times (from source code) - without and with some patches applied. Now it's just a matter of compiling libhybris from source code.

[xc-racer99]

Libhybris setup is easy - some time ago when i looked at this, it required to rebuild android device image two times (from source code) - without and with some patches applied. Now it's just a matter of compiling libhybris from source code.

Ok, I got libhybris setup again and succesfully ran pvrsrvinit (although that doesn't require libhybris, as you already know). test_egl ends with a segfault; test_hwcomposer. More on this below...

Additionally, we could try recompiling the hwcomposer module so that it uses /dev/fb{x} as opposed to /dev/graphics/fb{x}. Or we could simply make it a dummy hwcomposer that always returns the same gralloc usage flags which it pretty much does right now anyways :)

We are definitely going to have to make some modifications to the hwcomposer to support a mainline kernel. For one, it tries to open the FIMC which we haven't setup, it has some special FB initializing that we can probably skip, and for vsync it uses a custom IOCTl or else a ueventd handler - neither of which we probably want to use.

I'm wondering if there's a way to do a render-only driver using the supposed pixmap support if we could find a driver (http://processors.wiki.ti.com/index.php/RN_4_04_00_04?keyMatch=SGX540&tisearch=Search-EN briefly describes that the pixmaps are supported)

[PabloPL]

Fixed powervr logging a bit (was hard to do this...), can be enabled with CONFIG_PVR_NEED_PVR_DPF.

Edit 1:
Fixed also debug build (with all debug/trace options enabled).

[PabloPL]

Added workaround so we can use debug driver with release binaries. Maybe all debug enabled will help to trace why we can't rung egl programs.

[PabloPL]

@xc-racer99 Are You able to run libhybris binaries under android (or we could write some sample code with android-ndk - maybe could use some code from https://github.com/limadriver-ng/lima/tree/master/limare/tests)?
I wonder what could be log, when running under stock kernel with following patches applied
f5a3e48
c4c80bc
31ba0b9

I'm thinking also about making some wrap (like it was done for lima) so we can dump content of ioctl.

[PabloPL]

Currently eglGetConfig in test_egl_configs returns with error EGL_NOT_INITIALIZED

[xc-racer99]

@xc-racer99 Are You able to run libhybris binaries under android (or we could write some sample code with android-ndk - maybe could use some code from https://github.com/limadriver-ng/lima/tree/master/limare/tests)?
I wonder what could be log, when running under stock kernel with following patches applied

As in running the Android binaries under Android or the libhybris specific ones? I doubt I could run anything via libhybris on Android; would need to recompile it at any rate. Are you wanting to try running the test_egl, etc binaries under Android?

Sure, I can cherry-pick those commits and run them from an Android kernel and get you the logs.