Porting Replicant to Android 11¶

• Table of contents
• Porting Replicant to Android 11
  • Building Replicant 11
    • Source code
    • Build dependencies
      • For Trisquel 8
    • Fixing the build environment
      • Allow system binaries for building
      • Java heap space
      • Reduce parallel jobs to avoid killed processes
    • Launching the build
    • Install the images
      • From scratch
      • Update previous installation
    • Get adb
    • Boot progress
    • Getting the latest changes
    • Known working distributions
    • Build VM
    • Known issues
      • repo not working
  • Cleanups to be done
  • Upstreaming status
  • Graphics status
  • Modem status
    • Modem status TODO
  • TODO
    • Devices support:
      • Easy, because it's similar enough to the Galaxy SIII (I9300)
      • Needs more work and unknown upstream Linux status
  • Documentation
    • Replicant 6.0 changes
    • Other rebases
  • Other attempts
    • Links for other attempts
    • CustomROMs i9300 components
  • Known error messages that are safe to ignore
  • Links

Active development has moved towards AOSP 11.

The basic features working on Replicant 10 have been confirmed to work on Replicant 11, but not everything has been tested yet on either version. Replicant 10 source code and build instruction have been kept to do regression tracking: Porting Replicant to Android 10.

Building Replicant 11¶

Source code¶

$ repo init -u https://git.replicant.us/replicant-next/manifest.git -b replicant-11-dev
$ repo sync

Alternatively a shallow copy of the source tree can be fetched in order to save on disk space:

$ repo init -u https://git.replicant.us/replicant-next/manifest.git -b replicant-11-dev --depth=1
$ repo sync -c

To unshallow a specific module:

$ cd path/to/module
$ git fetch --unshallow <remote>

Build dependencies¶

For Trisquel 8¶

sudo apt-get install bc bison build-essential bsdmainutils ccache curl flex g++-multilib gcc-multilib gettext git gnupg gperf imagemagick lib32ncurses5-dev lib32readline-dev lib32z1-dev liblz4-tool libncurses5-dev libsdl1.2-dev libssl-dev libwxgtk3.0-dev libxml2 libxml2-utils lzop python-mako pngcrush rsync schedtool squashfs-tools xsltproc zip zlib1g-dev
sudo apt-get install gcc-5-arm-linux-gnueabi

Fixing the build environment¶

Allow system binaries for building¶

By default, the Android 11 build system can only use the prebuilt binaries it ships.

• We need to be able to rebuild the binaries, and so far no one did it yet.
• Binaries are way harder to trust than source code and not everyone trust Google.

As GNU/Linux distribution's tools can be rebuilt and are easier to trust, we are using that for now.

Java heap space¶

The Java heap size is automatically set according to the available system memory. On machines with 8 GB or less RAM, it is set to a value which is too low, and will result in the following error during the build:

Exception in thread "main" java.lang.OutOfMemoryError: Java heap space

The heap size can be increased with an envirnoment variable:

$ export _JAVA_OPTIONS="-Xmx3g" 

Reduce parallel jobs to avoid killed processes¶

Increasing the Java heap space is not enough to get a successful build on machines with 8 GB or less RAM. It is also necessary to reduce the number of parallel jobs, to avoid processes from being killed due to lack of memory. This typically happens during the build of frameworks/base components.

For greater speed, you may let your build run with the defaults, wait for it to fail due to killed processes, and then relunch the build with:

$ make -j1

By default, Ninja, the underlaying build system for Android, used when you run make bacon, computes the number of parallel jobs according to the number of CPUs on your machine (typically #CPUs + 2 parallel jobs).

Launching the build¶

$ cd kernel/replicant/linux && ARCH=arm CROSS_COMPILE=arm-none-eabi- make replicant_defconfig && ARCH=arm CROSS_COMPILE=arm-none-eabi- make -j9 && cat arch/arm/boot/zImage arch/arm/boot/dts/exynos4412-i9305.dtb > zImage-dtb
$ source build/envsetup.sh
$ lunch lineage_i9305-eng
$ make

Install the images¶

From scratch¶

$ cd out/target/product/i9305
$ sudo heimdall flash --BOOT boot.img --USERDATA userdata.img --SYSTEM system.img 

Update previous installation¶

adb remount
adb sync

Get adb¶

As the device IDs are the ones given by the Linux kernel, they are not in the adb udev rules, so for now it requires to run adb as root:

$ sudo adb shell
* daemon not running; starting now at tcp:5037
* daemon started successfully
i9305:/ #                                 
$ sudo adb kill-server
$ adb shell
* daemon not running; starting now at tcp:5037
* daemon started successfully
error: no devices/emulators found

$ adb kill-server
$ sudo adb shell
* daemon not running; starting now at tcp:5037
* daemon started successfully
i9305:/ # 

Boot progress¶

You can also follow the boot progress with adb:

adb logcat
adb logcat -b main

Note that the device can go into suspend at any time, so adb might be interrupted. That looks like that:
First you get a shell

$ sudo adb shell
i9305:/ #

Then the connection is interrupted:

$ adb shell
i9305:/ # [randomdev@fullyfreelaptop ]$                                                                                                     

The effect with adb logcat is similar.

Getting the latest changes¶

• The repositories are being constantly modified with git push --force as we are trying things out, and don't want to make the commits history look too dirty, so be sure to backup your local changes.
• Sometimes the manifest repository is also modified with git push --force. In that case the following commands will loose all the work you did locally but will make the repository consistent with upstream repositories again:
  

  $ rm -rf .repo/manifests .repo/manifests.git .repo/manifest.xml
  $ repo init -u https://git.replicant.us/replicant-next/manifest.git -b replicant-11-dev
  $ repo sync --force-sync
  

• The following command might also be necessary to make the state consistent with upstream repositories again, when the manifest history wasn't rewritten, but it will also loose all the work you did locally:
  

  $ repo sync --force-sync
  

Known working distributions¶

Build VM¶

If you use Parabola, you may be interested in running Trisquel 8 in LXC.

To do that first debootstrap a Trisquel 8 rootfs.

Parabola's debootstrap does support Trisquel 8 and its manual has an example on how to do that:

$ man debootstrap
[...]
# debootstrap flidas flidas-root http://archive.trisquel.info/trisquel

Then you can use virt-manager to setup the LXC instance.

• The LXC guest and host shares their resources (CPU, RAM) with almost no penalty
• Trisquel 8 is not a rolling release distribution

• you need to configure the Trisquel 8 LXC instance (vimrc, sshd_config, etc)
• It's more complicated to setup
• The Android build system outputs a warning message about not being able to use namespaces which may become mandatory in newer Android versions

Known issues¶

repo not working¶

After some repo sync, repo can stop working on Trisquel 8:

$ repo init [...]
Traceback (most recent call last):
  File "[...]/replicant-11/.repo/repo/main.py", line 56, in <module>
    from subcmds.version import Version
  File "[...]/replicant-11/.repo/repo/subcmds/__init__.py", line 38, in <module>
    ['%s' % name])
  File "[...]/replicant-11/.repo/repo/subcmds/upload.py", line 27, in <module>
    from hooks import RepoHook
  File "[...]/replicant-11/.repo/repo/hooks.py", line 472
    file=sys.stderr)
        ^
SyntaxError: invalid syntax

This is because /usr/bin/python is python2 instead of python3.

In that case you can use python3 $(which repo) instead of repo) like that:

$ python3 $(which repo) init [...]

Or you can install repo in /usr/local/bin and modify the #!/usr/bin/env python inside it with the following commands:

sudo install -m 755 .repo/repo/repo -t /usr/local/bin
sudo sed 's&^#!/usr/bin/env python$&#!/usr/bin/env python3&' /usr/local/bin/repo -i

Cleanups to be done¶

• Make adb work as user by using the right USB IDs, and make userspace do the USB setup.
• Make the kernel not use hardcoded CMDLINE_FORCE Not possible unless the bootloader is changed or Linux is very heavily patched.
• Make the kernel not use hardcoded partitions if possible (though we use system as root)
• Make a clean Gatekeeper HAL module implementation instead of using the same hack than goldfish
• Look at init.rc documentation to see if init.rc can be overriden clearly with the override statement to see if it's possible to keep the serial console patch for -eng

Upstreaming status¶

• The stock bootloader is incompatible with Linux, see BootloadersIncompatibleWithLinux for more information. So we maintain patches to enable the Galaxy SIII, and Galaxy Note II to boot with the stock bootloader. In the long run we need to look into using u-boot in the kernel partition as using u-boot instead of the stock bootloader currently require nonfree and non-redistributable software (BL1).
• For the patches that are not merged yet, see the issues of the redmine upstreaming sub-project

Graphics status¶

Progress of the graphics related tasks is tracked at GraphicsReplicant10.

Modem status¶

• Fully tested under GNU/Linux only
• Can initialize completely the modem and receive messages (see the #1954 bug report for the logs)
• Needs more cleanup but there is now a better abstraction

• Ported to Replicant 9 using a wrapper for the API >=12 in libsamsung-ril source code that needs to be removed
• Tested under Replicant 9 without up to date libsamsung-ipc (no modem init)
• Tested and validated under Replicant 6 (doesn't break telephony)

Modem status TODO¶

• Implement the missing part to shut down the modem, close the interfaces and such, in order to need to reboot after each test.
• Continue to clean up the libsamsung-ipc and libsamsung-ril patches, test the patches in Replicant 6 when applicable, and merge them in the upstream repositories.
• Convert the firmware loading driver to the upstream API and then adapt libsamsung-ipc for that. This should also benefit other devices like the Galaxy SIII 4G, and the Galaxy Note II 4G which probably don't need much more to get their modem supported by upstream Linux.
• Cleanup and convert the rest of the drivers to look like the ones for the Nokia N900 and adapt the userspace in libsamsung-ipc, and merge libsamsung-ipc support for that once the kernel API is stable.
• Test the code under Replicant 10 when the graphics status will enable to have decent enough speed to do some testing through human interaction.

TODO¶

First month of full time equivalent work:

¹ The generic instructions were tested at Install parties in Paris

Devices support:¶

Easy, because it's similar enough to the Galaxy SIII (I9300)¶

Needs more work and unknown upstream Linux status¶

Documentation¶

Replicant 6.0 changes¶

See Replican6Changes.

Other rebases¶

See the Samsung-ipc page.

Other attempts¶

• It might be interesting to contact the people doing ports once we have something working well enough.
• It is also interesting to look at other attempts to understand if a given device is powerful enough to run Android 9 and what configuration was used to achieve it.

Links for other attempts¶

• https://www.xda-developers.com/android-pie-android-9-port-custom-roms/

CustomROMs i9300 components¶

Known error messages that are safe to ignore¶

• TestHarnessModeService: Failed to start Test Harness Mode; no implementation of PersistentDataBlockManagerInternal was bound
• JniUtils: Could not load native library jni_latinimegoogle

Links¶

• Android build requirements - hardware and software
• Why LineageOS Developers are building Android Go-optimized custom ROMs
• Android Go recommended default values for propreties for optimization
• Team InFusion i9300 optimized system.prop
• Hack to fix high CPU usage caused by logd
• Use low-end video codecs
• Optimize ActivityManager cached apps
• Use 1Gb Dalvik config