Project

General

Profile

Actions

DeprecatedPortingToAndroid10 » History » Revision 286

« Previous | Revision 286/289 (diff) | Next »
Kurtis Hanna, 08/02/2020 12:08 AM
Updated ComicoTeam's Pie build date


Porting Replicant to Android 10

Active development has moved towards LineageOS 17 (based on Android 10).

The basic features working on Replicant 9 have been confirmed to work on Replicant 10, but not everything has been tested yet on either version. Replicant 9 source code and build instruction have been kept to do regression tracking: Porting Replicant to Android 9.

Building Replicant 10

Source code

$ repo init -u https://git.replicant.us/replicant-next/manifest.git -b replicant-10-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-10-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 10 build system can only use the prebuilt binaries it ships.

While having binary toolchains is better for reproducible builds, and that the binaries are free software, this creates a number of issues:
  • 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.

Setting the following envrionment variable allows to use your distribution tools:

$ export TEMPORARY_DISABLE_PATH_RESTRICTIONS=true

Note that setting this variable does not automatically make the build system use only system binaries: if a prebuilt binary exist, it will use it, if not, it will use your system binary.

Lots of further effort must be put into transitioning to the system binaries and/or creating a scripts that would build all the required tools.

Mako (Python) for Mesa

To avoid the following error:

16:20:13 See https://android.googlesource.com/platform/build/+/master/Changes.md#PATH_Tools for more information.
[  3% 2585/70375] build out/target/product/i9305/gen/STATIC_LIBRARIES/libmesa_nir_intermediates/nir/nir_builder_opcodes.h
FAILED: out/target/product/i9305/gen/STATIC_LIBRARIES/libmesa_nir_intermediates/nir/nir_builder_opcodes.h
/bin/bash -c "python external/mesa3d/src/compiler/nir/nir_builder_opcodes_h.py external/mesa3d/src/compiler/nir/nir_opcodes.py > out/target/product/i9305/gen/STATIC_LIBRARIES/libmesa_nir_intermediates/nir/nir_builder_opcodes.h" 
Traceback (most recent call last):
  File "external/mesa3d/src/compiler/nir/nir_builder_opcodes_h.py", line 106, in <module>
    from mako.template import Template
ImportError: No module named mako.template
16:20:20 ninja failed with: exit status 1

You need to run the following command:

$ cd prebuilts/build-tools/path/linux-x86/
$ rm python && ln -s /usr/bin/python python

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

$ 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

So make sure to kill the adb-server and run it as root:
$ 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-10-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
    

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 advantages of this solution are that:
  • The LXC guest and host shares their resources (CPU, RAM) with almost no penalty
  • Trisquel 8 is not a rolling release distribution
The disadvantage of this solution are that:
  • 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

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

libsamsung-ipc:
  • 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
libsamsung-ril:
  • 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:

Time estimation Task Comments
DONE boot a device under AOSP9 Only boots with graphics, not much more
7h DONE build it under a FSDG compliant distribution like Trisquel8 WIP for AOSP, It's difficult to do precise time estimations as it could work out of the box or require one full time month of work depending on how much issues are encountered
Builds under Trisquel8
21h DONE * port the changes from AOSP9 to LineageOS 16
* cleanup the code
* build the kernel from the Android build system
* make sure it builds with an FSDG compliant distribution
* document the build procedure

Status:
* Boots with adb.
* Has ultra slow graphics
14h find, remove and document proprietary software in LineageOS 16
21h find, remove and document privacy issues in LineageOS 16
7h Add support for the touch keys driver in the galaxy-s3 dts applied
7h upstream the AAT1290 flash led Linux dts for the galaxy-s3 boards Now in 5.3
7h rebrand LineageOS as Replicant
70h port and cleanup the the Galaxy SIII (i9300) modem Linux driver from 4.16 to 5.0 See the modem status for more details
Total: 147h (~1 month)
Second month of full time equivalent work:
* port libsamsung-ril and libsamsung-ipc to Android 9
* Make the modem driver and libsamsung-ipc work together
157h See the modem status for more details
Total 157h ~1 month
Third month of full time equivalent work:
Task Time estimation Comments
port the sensors libraries and other device specific libraries
Look which sensor libraries can be used
70h Already done by the unofficial LineageOS port of the Galaxy SIII (i9300), needs testing
add support for Audio with the upstream kernel driver 70h Might be way faster, depending on what Android 9 uses
See also this bugreport
add partial (no modem) support for the Galaxy SIII 4G (i9305) and factorize the code with i9300 14h * The source code on which the work was based changed from AOSP to an unofficial LineageOS port to a port of i9305 support for AOSP by Joonas to the official LineageOS so it's now supported by default
* The work to factorize the code between the i9300 and i9305 still need to be done
Total 154h ~1 month
Task Time estimation Comments
create a recovery 21h
add internal WiFi support and validate the functionality 6h
add external WiFi dongles support 20h External dongles support might be tricky
create new update the install and upgrade instructions 35h Our current install instructions don't scale as we have one copy for each device.
We also created generic instructions but they tend to be harder to follow1 than the device specific ones.
This will be made in a modular format (for instance in LaTeX) that enables to generate per device install instructions without requiring to have multiples copy of the same text.
The instructions will need to be able to be modified and compiled on an FSDG compliant distribution.
Mostly done:
* The installation instructions are now generic enough.
* Some long standing TODO were also done along the way like adding backup instructions for the EFS.
* The current instructions are still for Replicant 6.0 and will need to be updated for Replicant 9.0
Task Time estimation Comments
Estimate the amount of work to Reduce the attack surface ?
Estimate the amount of work to add in-system upgrades ?

1 The generic instructions were tested at Install parties in Paris

Devices support:

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

Galaxy Note II (N7100)
Task Time estimation Comments
port the EA8061 LCD Linux driver 35h
port the S6EVR02 LCD Linux driver 35h
port the MAX77693 flash led Linux driver 7h
android: add support for the Note II (N7100) and factorize the code with Galaxy SIII (i9300) and Galaxy SIII 4G (i9305) 14h Should be similar to the Galaxy SIII
port the sensors libraries and other device specific libraries 70h It's difficult to evaluate how much time it could take
add support for Audio with the upstream kernel driver 14h Should be similar to the Galaxy SIII
Galaxy Note 8.0 (N5100) and 8.0 WiFi (N5110)
Task Time estimation Comments
Evaluate the time required to do the port 14h TODO

Needs more work and unknown upstream Linux status

Device Time estimation Comments
Galaxy S II (i9100) Linux: devboard dts upstream? unknown status
Galaxy Note (N7000) unknown Linux upstream status
Galaxy Nexus (I9250) OMAP4, no dts upstream
Galaxy Tab 2 7.0 (P3100), 7.0 WiFi (P3110), 10.1 (P5100), 10.1 WiFi (P5110)
GTA04 >= A4 TODO: a RIL needs to be written, userspace GPS support is missing, audio scenarios, etc

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.
Device(s) Repository status Comments
i9300 CustomROMs * February 8 2020 release
i9300 Team InFusion * August 20 2019 release Issues: * Uses a Samsung kernel
* Uses too many nonfree libraries
=> Probably nothing we could reuse from its code
N7100 ComicoTeam * January 4 2020 release
I9100 XDA

Links for other attempts

CustomROMs i9300 components

Repository Tree path Dependencies Function Comments
https://github.com/CustomROMs/android_hardware_samsung
lineage-16.0 branch
hardware/samsung/macloader Loads the MAC Address of the WiFi network interface Might be useful
hardware/samsung/wifiloader Loads the wifi kernel module (like modprobe) and setup firmware filesystems permissions May be useful
hardware/samsung/audio seems to contains ril related stuff as well Look if the ril stuff is required, go for standard audio
hardware/samsung/lineagehw/hidl/livedisplay livedisplay is a feature similar to what redshift does on GNU/Linux Not sure if it works with mainline
hardware/samsung/exynos/multimedia/utils/ seem meant for audio/video decoding offload assembly obtimized color conversion and resize code check assembly code license, not sure if useful
all other directories in hardware/samsung/exynos/ nonfree firmwares, nonfree software?, smdk kernel? audio/video decoding offload Avoid using that
hardware/samsung/exynos3 nonfree firmwares?, nonfree software?, smdk kernel? some light libraries, display stuff (gralloc, etc), 2D acceleration (FIMG), camera (FIMC), 3D acceleration, etc Avoid using that for now
hardware/samsung/exynos4

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

Updated by Kurtis Hanna over 4 years ago · 286 revisions

Also available in: PDF HTML TXT