Replicant

h1. GTA04 Kernel

This page documents the various tried to get a working kernel for the GTA04 with ICS userspace.

Notes:

* The GTA04 is an AM37x OMAP board like the BeagleBoard XM. The other BeagleBoard models use another OMAP3 SoC.

* "AM37x category":http://processors.wiki.ti.com/index.php/Category:AM37x on TI wiki

Resources:

* Neil Brown's "gta04 kernel tree":http://git.neil.brown.name/git?p=gta04.git includes 3.2, 3.3, 3.4 and 3.5 kernels

* AOSP's "common kernel tree":https://android.googlesource.com/kernel/common/ includes 3.0 3.3 and 3.4 kernels

h2. Rowboat kernel

The TI porting guide for AM37x devices advices to use the rowboat tree with the rowboat-ics.xml manifest.

The kernel in that tree is branch "rowboat-ics-kernel-2.6.37":https://gitorious.org/rowboat/kernel/commits/rowboat-ics-kernel-2.6.37 from the rowboat kernel.

Porting the GTA04 board and drivers to this kernel seems possible, however this is a very old kernel revision that shouldn't be compatible with the ICS userspace (even though it apparently works with rowboat ICS userspace).

One of the biggest issues is that the adb interface was completely changed and the one in that kernel will not work with Replicant 4.0 userspace.

h2. Omapzoom kernel

Omapzoom (where TI pushes code for Android) has an "omap3 kernel":http://omapzoom.org/?p=kernel/omap.git;a=commit;h=refs/heads/p-android-omap3-3.0 that should match ICS userspace.

However, when running on the GTA04, serious power management issues caused characters to be dropped on serial after not touching the phone for a couple of seconds as well as IRQ interrupts being dropped, which caused the touchscreen to not work properly. Suspend/resume and earlysuspend works perfectly though. Maybe this kernel wasn't designed for AM37x OMAP3 devices but another kind of OMAP3 SoCs: there is no support for Android Beagleboard.

h2. AOSP's 3.4 common kernel merged with Neil Brown's 3.4 gta04 kernel

Merging AOSP's 3.4 common kernel on top of Neil Brown's 3.4 gta04 kernel and adding the Android options to the defconfig resulted in a non-working kernel where input events didn't seem to be reported correctly to userspace (even with an USB mouse, click events seemed dropped).