# Raspberry PI + cross compile & build kernel

* https://www.raspberrypi.com/documentation/computers/linux_kernel.html
* recommend: use another micro SD card for this lab!

# topic 1. build kernel

## 先準備VM (ubuntu 20, 64bit)

* 使用virtualbox + Ubuntu 20 (64bit)
* https://releases.ubuntu.com/focal/ubuntu-20.04.6-desktop-amd64.iso
* https://ubuntu.cs.nycu.edu.tw/ubuntu-release/20.04.6/ubuntu-20.04.6-desktop-amd64.iso

* 官方文件: https://www.raspberrypi.com/documentation/computers/linux_kernel.html#cross-compiling-the-kernel

## virtualbox setting

1. Install VirtualBox Guest Additions

    sudo apt update
    sudo apt upgrade
    sudo apt install build-essential dkms linux-headers-$(uname -r)

* 重開機後, 再執行 Insert Guest Additional CD image
* 裝完之後, 再重開機

## Dependencies and Toolchain

* Install Required Dependencies and Toolchain
`sudo apt install git bc bison flex libssl-dev make libc6-dev libncurses5-dev`

* Install the 64-bit Toolchain for a 64-bit Kernel
`sudo apt install crossbuild-essential-arm64`

* Get the Kernel Sources
`git clone --depth=1 https://github.com/raspberrypi/linux`
* Omitting the --depth=1 will download the entire repository, including the full history of all branches, but this takes much longer and occupies much more storage.

## Load kernel config and build
* Load 64-bit default configs (For Raspberry Pi 3, 3+, 4, 400 and Zero 2 W, and Raspberry Pi Compute Modules 3, 3+ and 4:)

    cd linux
    KERNEL=kernel8
    make ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- bcm2711_defconfig

* Build image for all 64-bit Builds with default config
* 第一次跑大約要1小時 (看電腦的性能, 我花了77min)

make ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- Image modules dtbs

### Configuring the Kernel by menuconfig
* Preparing to make menuconfig

* cross-compiling a 64-bit kernel 

* Go to “General setup” -> Select “Local version” -> Put your local version (student id number)
* save and exit, then start to build kernel

* After building, your image locates: arch/arm64/boot/




## Install the new kernel 
* 把micro sd卡連接到虛擬機裡面, 系統會自動掛載磁區, 可以用df -h檢查before/after 的變化
  * 下圖例子, 是/media/pi/bootfs 與 /media/pi/rootfs
![image](https://hackmd.io/_uploads/rkeYeqGlMl.png)

* install the kernel modules onto the SD card

* back up old kernel

KERNEL=kernel8
sudo cp /media/pi/bootfs/$KERNEL.img   /media/pi/bootfs/$KERNEL-backup.img

* copy the kernel and Device Tree blobs onto the SD card

KERNEL=kernel8
sudo cp arch/arm64/boot/Image	/media/pi/bootfs/$KERNEL-new.img
sudo cp arch/arm64/boot/dts/broadcom/*.dtb	 /media/pi/bootfs/
sudo cp arch/arm64/boot/dts/overlays/*.dtb*	 /media/pi/bootfs/overlays/

* edit the /media/pi/bootfs/config.txt file, add the following setting

    * To restore the original kernel, remove the above setting

* remove SD card on VM

sudo umount /dev/sdX1
sudo umount /dev/sdX2

* Insert SD card to PI and boot!! check your kernel version


---
---

# topic 2. cross compile
0. 準備一個c code

#include <stdio.h>
int main()
{
	printf("hello, world\n");
        return 0;
}

1. 建立流程

nano hello.c		    // write your code
gcc hello.c -o hello.o	// compile it, the output file is hello.o
./hello.o			    // execute hello.o

2. Cross compile, then copy it to PI and execute

(@VM) aarch64-linux-gnu-gcc hello.c -o hello.arm
(@VM) sudo cp hello.arm /media/pi/rootfs/home/pi/
// 複製hello.arm到PI裡面的家目錄
(@PI) chmod  +x  hello.arm
(@PI) ./hello.arm

---
---

## topic 3. cross compile Tensorflow lite for PI
* Install docker on Ubuntu (VM) or WSL
* WSL = Windows Subsystem for Linux
* Target: Raspberry PI4 (bookworm 64bit) + python 3.11
* 編譯時, 會自動下載相關套件, 建議在網路穩定的地方進行

### download source code

git clone https://github.com/tensorflow/tensorflow.git tensorflow_src
cd tensorflow_src
git checkout v2.15.1

### 修改Makefile

* 修改: BASE_IMAGE ?= ubuntu:20.04
* 調整為: BASE_IMAGE ?= ubuntu:22.04
![image](https://hackmd.io/_uploads/HJnuXH7efx.png)



### 修改Dockerfile 

* 新增參數: `ENV DEBIAN_FRONTEND=noninteractive`
* 移除舊有cmake安裝指令, 改由apt安裝

### 開始cross compile

make -C tensorflow/lite/tools/pip_package docker-build \
TENSORFLOW_TARGET=aarch64 PYTHON_VERSION=3.11

* 編譯完產生的檔案路徑: `tensorflow/tensorflow/lite/tools/pip_package/gen/tflite_pip/python3.11/dist/tflite_runtime-2.15.1-cp311-cp311-linux_aarch64.whl`

#### troubleshooting
* If you see "Killed signal terminated program cc1plus",
it usually means WSL/Docker ran out of memory.
Try BUILD_NUM_JOBS=1 or increase WSL memory/swap.

### 複製檔案到PI
* 如果使用WSL, 可於本機端使用網路芳鄰進入WSL資料夾
`檔案總管路徑: \\wsl$\Ubuntu\root\tensorflow_src
`
* 將編譯完的*.whl複製到PI, 然後在新的虛擬環境安裝

python -m venv tflite_vmtest
source tflite_vmtest/bin/activate
pip install ./tflite_runtime…..whl
pip list