Setting up a cross-compilation toolchain

This page is a brief summary of the cross-compilation discussion on the SamyGO forum. Main contributors to this information are Robbiesz and marcelru. The information in this page pertains to the building of a toolchain, based on the source code as stored in the file 32B650.zip, available from Samsung.

We are still in the testing phase, for now, check out the forum discussion (topic: cross-compilation [1]). For more info, errors you may encounter, support for other Samsung toolchains, please refer to the forum, or better still, participate.

Toolchains 101

A toolchain is a set of executable programs that enables you to build (assemble, compile and link) your own executable programs from source code (typically C, C++, fortran or other programming languages). To build these executables, the source code needs to be translated into machine-readable code. This is what a toolchain does. The translation proceeds in three steps (well, you may think of it this way): translation into low-level code (compilation), translation to machine-readable code (assembly), combination with other code libraries (linking). Compilation/assembly mostly proceeds in a single go. So, for a toolchain you need at least three programs: a compiler, an assembler, and a linker. In almost all cases you will also need a C-library to link with.

Unlike source code, executables are machine-specific. That means that executables built for the Intel x86 architecture (the basis for most computers in the world today) will not run on other types of processors, for example the ARM architecture, the basis for many embedded systems. Our Sammys are ARM-based machines. To build executables for a Samsung TV on an x86-based machine, you will need a toolchain that runs on the x86-architecture and builds executable code for the ARM-architecture. This procedure is called cross-compilation, and requires a cross compilation toolchain. This wiki page deals with the installation of the toolchain provided by Samsung, on an x86, Linux or Windows (Cygwin) platform.

The Samsung toolchain

The source code for the Samsung ARM toolchain is available on the net, but needs some adjustments before it can be built. We will discuss that later. The main parts are:

• GNU binutils (containing among other things an assembler and a linker)
• GCC, the GNU C/C++ compiler (and a host of other languages we will not use)
• Glibc, the GNU C library
• linux kernel sources, for operating system/processor specific details.

For the remainder of this wiki page, we will use the toolchain for the 32B650/32B550/... TV's as an example. The source code for this toolchain is stored in the file 32B650.zip [2]. Download this file and store it in the directory where you want to build your toolchain. Download the SamyGO patches that enable compilation of the toolchain here: [3]

Building the toolchain on Linux

Since most Linux distro's come standard with a full toolchain for building packages, building the cross-compilation toolchain on this platform is fairly straightforward, and proceeds in 6 stages.

Source installation

Unpack and install the source code files. This is done with the script install_sources.sh: edit paths and code versions to your liking:

#! /bin/sh
#
# install_sources.sh installs the source code for the toolchain for Samsung
# TV's, as found in the file 32B650.zip. Place this file together with the 
# patches tarball (SamyGO-toolchain_fc11_patches.tgz) in the ${BUILDROOT} 
# directory, defined below, and run this script.
#
BUILDROOT=/tmp/arm-tools         # edit according to your needs. 

cd ${BUILDROOT}
mkdir -p ${BUILDROOT}/src

# unpack the Samsung zip file

unzip 32B650.zip

# remove unnecessary stuff and move the kernel and toolchain code

rm -f libgphoto2-2.3.1.tar.zip
rm -f libusb-0.1.12.tar.gz
rm -f SDL-1.2.11.zip
mv linux.chelsea.tgz ${BUILDROOT}/src

# unpack the toolchain and remove unnecessary stuff

tar -zxvf SELP.3.2.x-Chelsea.src.tgz
rm -f SELP.3.2.x-Chelsea.src.tgz

mv ./SELP.3.2.x-Chelsea.src/Toolchain/gcc-4.2.0-4.0.9.tgz ${BUILDROOT}/src
mv ./SELP.3.2.x-Chelsea.src/Toolchain/binutils-2.17.50.tgz ${BUILDROOT}/src
mv ./SELP.3.2.x-Chelsea.src/Toolchain/glibc-2.5.90-9.0.9.tgz ${BUILDROOT}/src

rm -rf ./SELP.3.2.x-Chelsea.src

# unpack the patches and move them

tar -zxvf SamyGO-toolchain_fc11_patches.tgz
mv *.patch ${BUILDROOT}/src

Kernel headers installation

Install the kernel header files. You will need these when you want to build your own kernel modules for your TV. For that, the kernel source code's version has to match your TV's kernel version _exactly_. The installation is done with the install_kernel_src.sh script: again, edit to meet your needs.

#! /bin/sh 

# how to install kernel headers using this script:
# 
# Create the ${BUILDROOT} directory, put this script (install_kernel_src.sh) 
# in it.
# In ${BUILDROOT}, create a directory ./src and put the linux tarball in it.

# go up one directory and run this script. 

TARGET=arm-SamyGO-linux-gnueabi
BUILDROOT=/tmp/arm-tools
PREFIX=/usr/local                            # change to your needs
SYSROOT=${BUILDROOT}/sysroot

export ARCH=arm
export CROSS_COMPILE=${TARGET}-
export PATH=$PATH:${PREFIX}/bin
 
cd ${BUILDROOT}/src

# remove previous installation if it is there:
rm -rf linux 
rm -rf linux-r011 
rm -rf RFS 
rm -rf ssdtv_platform

# unpack, move up one dir and link:

tar -zxvf linux.chelsea.tgz              # from 32B650.zip
mv ./linux/* .
rm -rf linux/

ln -s ./linux-r011 linux

cd linux

# copy header files to sysroot:
mkdir -p ${SYSROOT}/usr/include
cp -a ${BUILDROOT}/src/linux/include/linux ${SYSROOT}/usr/include/linux
cp -a ${BUILDROOT}/src/linux/include/asm-arm ${SYSROOT}/usr/include/asm
cp -a ${BUILDROOT}/src/linux/include/asm-generic ${SYSROOT}/usr/include/asm-generic

# remove stale link to ssdtv headers:
rm -rf ${SYSROOT}/usr/include/asm/arch-ssdtv

# ... and copy these as well:
cp -a ${BUILDROOT}/src/ssdtv_platform/include/asm-arm/arch-ssdtv ${SYSROOT}/usr/include/asm

Everything is now installed (but not yet unpacked). The remainder of the installation consists of the actual building of the tools: First the binutils are built: as, ld and a bunch of others. For cross-compilation, these tools do not need any resources for the target platform (ARM, in this case).

Building of binutils

Apart from a small bug in one of the source files and a configuration change to be made to part of the code (patches are included) the build is straightforward, done with mkbinutils.sh:

#! /bin/sh

# how to build binutils using this script:
# 
# Create the ${BUILDROOT} directory, put this script (mkbinutils.sh) in it.
# In ${BUILDROOT}, create a directory ./src and put the binutils tarball and 
# the patch in it.

# go up one directory and run this script. 

TARGET=arm-SamyGO-linux-gnueabi
BUILDROOT=/tmp/arm-tools                     # change to your needs
PREFIX=/usr/local                            # change to your liking 
SYSROOT=${BUILDROOT}/sysroot
UTILS=binutils-2.17.50

export ARCH=arm
export CROSS_COMPILE=${TARGET}-
export PATH=$PATH:${PREFIX}/bin
 
cd ${BUILDROOT}/src

tar -zxvf binutils-2.17.50.tgz               # from 32B650.zip
patch -p0 < ./binutils-2.17.50_fc11.patch 

mkdir -p BUILD/${UTILS}
cd BUILD/${UTILS}

../../${UTILS}/configure --prefix=${PREFIX} --target=${TARGET} \
    --with-sysroot=${SYSROOT} 
make

make install

mkdir -p $PREFIX/$TARGET/include
cp ${BUILDROOT}/src/${UTILS}/include/libiberty.h $PREFIX/$TARGET/include

When all goes well, you will now have the following executables in /usr/local/bin:

arm-SamyGO-linux-gnueabi-addr2line
arm-SamyGO-linux-gnueabi-ar
arm-SamyGO-linux-gnueabi-as
arm-SamyGO-linux-gnueabi-c++filt
arm-SamyGO-linux-gnueabi-gprof
arm-SamyGO-linux-gnueabi-ld
arm-SamyGO-linux-gnueabi-nm
arm-SamyGO-linux-gnueabi-objcopy
arm-SamyGO-linux-gnueabi-objdump
arm-SamyGO-linux-gnueabi-ranlib
arm-SamyGO-linux-gnueabi-readelf
arm-SamyGO-linux-gnueabi-size
arm-SamyGO-linux-gnueabi-strings
arm-SamyGO-linux-gnueabi-strip

GCC installation, stage 1

The C-compiler (gcc) has to be built in two stages: First, it is built and linked against the existing C-library (glibc). This version of the compiler is used to build glibc for the target platform, and then gcc is built again, now with the just compiled C-library.

The stage 1 build of gcc renders a barebones gcc, just useful for compiling glibc, as described in the next step: Edit install_gcc_stage1.sh according to your needs and execute. The source code will be patched to prevent building of two runtime objects that cannot be linked yet (because we don't have a working C-library for ARM, yet)

#! /bin/sh

# how to install gcc stage 1 using this script:
# 
# Create the ${BUILDROOT} directory, put this script (install_gcc_stage1.sh) 
# in it.
# In ${BUILDROOT}, create a directory ./src and put the gcc tarball in it.

# go up one directory and run this script. 

TARGET=arm-SamyGO-linux-gnueabi
BUILDROOT=/tmp/arm-tools
PREFIX=/usr/local                            # change to your liking
SYSROOT=${BUILDROOT}/sysroot
GCC=gcc-4.2.0-4.0.9

export ARCH=arm
export CROSS_COMPILE=${TARGET}-
export PATH=$PATH:${PREFIX}/bin
 
cd ${BUILDROOT}/src

# unpack

tar -zxvf ${GCC}.tgz      # from 32B650.zip

patch -p0 < ./gcc-4.2.0-4.0.9_t-linux-eabi.patch

mkdir -p BUILD/${GCC}-stage1 

cd BUILD/${GCC}-stage1
  ../../${GCC}/configure --prefix=${PREFIX} --target=${TARGET} \
      --enable-languages=c --disable-shared --disable-threads \
      --disable-libmudflap --disable-libssp --disable-nls \
      --disable-libgomp --with-cpu=arm1136jf-s --with-fpu=vfp

  make all-gcc
  make install-gcc

Glibc installation

Build glibc for ARM: Edit the paths in mkglibc.sh (if needed) and execute:

#! /bin/sh

# how to install glibc using this script:
# 
# Create the ${BUILDROOT} directory, put this script (mkglibc.sh) 
# in it.
# In ${BUILDROOT}, create a directory ./src and put the glibc tarball in it.

# go up one directory and run this script. 

TARGET=arm-SamyGO-linux-gnueabi
BUILDROOT=/tmp/arm-tools                     # change to meet your needs
PREFIX=/usr/local                            # change to your liking
SYSROOT=${BUILDROOT}/sysroot
GLIBC_VER=glibc-2.5.90-9.0.9

export ARCH=arm
export CROSS_COMPILE=${TARGET}-
export PATH=$PATH:${PREFIX}/bin
 
cd ${BUILDROOT}/src

# unpack

tar -zxvf ${GLIBC_VER}.tgz              # from 32B650.zip

patch -p0 < ./${GLIBC_VER}_fc11.patch

rm -rf ${BUILDROOT}/src/glibc-build-localedef
mv ${BUILDROOT}/src/${GLIBC_VER}/glibc-build-localedef ./

mkdir -p ${BUILDROOT}/src/BUILD/${GLIBC_VER}
cd BUILD/${GLIBC_VER}

# create some cached defaults: 
echo "libc_cv_forced_unwind=yes" > config.cache
echo "libc_cv_c_cleanup=yes" >> config.cache

export GCC_PATH=${PREFIX}/bin

BUILD_CC=gcc CC=${GCC_PATH}/${CROSS_COMPILE}gcc ../../${GLIBC_VER}/configure \
    --target=${TARGET} --host=${TARGET} --prefix=${PREFIX}/${TARGET} \
    --with-__thread \
    --cache-file=config.cache --with-headers=${SYSROOT}/usr/include \
    --with-glibc=${BUILDROOT}/src/${GLIBC_VER} \
    --disable-profile --enable-add-ons
make 
make install

GCC installation, stage 2

Finally, gcc needs to be rebuilt and linked against the just built glibc for ARM. Now C++-support will be included as well.

Edit install_gcc_stage2.sh and execute. In this script, links to the kernel header files are created in the include/ directory of the toolchain. That's not very elegant. Normally, the C-compiler should try and find the kernel headers in e.g., /usr/src/linux/include, but that's where the native kernel sources for your build platform reside. Your build platform is not necessarily the same as your target platform (most likely not). The method presented here works, but needs some editing in the future.

#! /bin/sh

# how to install gcc stage 2 using this script:
# 
# Create the ${BUILDROOT} directory, put this script (install_gcc_stage2.sh) 
# in it.
# In ${BUILDROOT}, create a directory ./src and put the gcc tarball in it.

# go up one directory and run this script. 

TARGET=arm-SamyGO-linux-gnueabi
BUILDROOT=/tmp/arm-tools
PREFIX=/usr/local                            # change to your liking
SYSROOT=${BUILDROOT}/sysroot
GCC=gcc-4.2.0-4.0.9 

export ARCH=arm
export CROSS_COMPILE=${TARGET}-
export PATH=$PATH:${PREFIX}/bin
 
cd ${BUILDROOT}/src

# remove links to kernel headers if they exist:
rm -f  ${PREFIX}/${TARGET}/include/linux
rm -f  ${PREFIX}/${TARGET}/include/asm
rm -f  ${PREFIX}/${TARGET}/include/asm-generic

# create symlinks to kernel headers:

ln -s ${SYSROOT}/usr/include/linux ${PREFIX}/${TARGET}/include/linux
ln -s ${SYSROOT}/usr/include/asm ${PREFIX}/${TARGET}/include/asm
ln -s ${SYSROOT}/usr/include/asm-generic ${PREFIX}/${TARGET}/include/asm-generic

# make builddir: 

mkdir -p BUILD/${GCC}-stage2

cd BUILD/${GCC}-stage2

CC=gcc ../../${GCC}/configure --target=${TARGET} --prefix=${PREFIX} \
   --enable-languages=c,c++ --enable-threads --enable-shared \
   --disable-nls --enable- __cxa_atexit --enable-long-long \
   --enable-c99 --with-cpu=arm1136jf-s --with-fpu=vfp

make
make install

When you look in /usr/local/bin now, you should see the cross-compilers for C and C++ for arm-SamyGO-linux-gnueabi.

Building the toolchain on Windows (Cygwin)

The toolchain build on Windows/Cygwin proceeds in the same way as the build on Linux. If you have the choice between Windows and Linux, go for Linux. Shell script execution is orders of magnitude faster on Linux, and the installation procedure is carried out through _many_ shell scripts. Don't say that you haven't been warned. To be able to build the toolchain you need to have Cygwin installed, see the Cygwin website [4] for details on installation. At the time of writing, the Cygwin stable release version was 1.5.25-15, and that is the version used for the build here, on Windows XP Home Edition. Cygwin creates a Unix-style environment on the Windows platform, and comes with pretty much the same development environment as GNU/Linux. You will need this GNU/Linux-style environment to be able to build your toolchain. The native Windows environment (visual studio and the likes) will NOT work.

Cygwin installation notes

When installing Cygwin, select "Unix-style linefeeds" (the default) when prompted. Otherwise you will find yourself editing ALL installation scripts, not just the handful presented here. Apart from the default (minimal) installation, which you shouldn't touch, you should have at least the following packages installed:

(Untick the "Hide obsolete packages" box in the "Select Packages" installer window to view everything, the packages in the following list may have been selected already for default installation)

• autoconf
• automake
• bash
• binutils
• bison
• byacc
• bzip2
• diffutils
• flex
• gawk
• gcc
• gcc-core
• gcc-g++
• gcc4
• gcc4-core
• gcc4-g++
• gettext
• gzip
• libgcc1
• libiconv
• m4
• make
• patch
• patchutils
• texinfo
• unzip
• zip

It is not clear whether you actually need all of these, but most packages take little disk space and little time to install. Running into errors because of lacking packages will take more time.

With Cygwin properly installed, the same 6 steps as described in the Linux installation section need to taken. Download the SamyGO patches and scripts [5] and edit these.

Installation

In the following, only differences with respect to the Linux installation will be described.

1: Edit and execute install_sources.sh: By default, cygwin is installed in C:/cygwin. All paths used in Cygwin are relative to this installation path. Edit the line BUILDROOT in all installation scripts to make it point to the full path of your build (be sure to use the forward slashes!!).

...
BUILDROOT=c:/cygwin/home/username/tmp/arm-tools         # edit according to your needs. 
...

When using notepad.exe or another DOS-style editor, revert the script, after saving, to unix-style linefeeds with the command:

dos2unix ./install_sources.sh

Do this for all scripts you edit, BEFORE execution. Failing to do so may render errors like:

$'\r': command not found

This error type occurs when shell scripts with DOS rather than Unix-style linefeeds are being processed.

2: Install the kernel sources with the just edited ./install_kernel_src.sh

3: Edit ./mkbinutils.sh. Add a line to make the environment variable CC to point to gcc-4:

...
export ARCH=arm
export CROSS_COMPILE=${TARGET}-
export PATH=$PATH:${PREFIX}/bin
export CC=/usr/bin/gcc-4
...

and build the binutils with ./mkbinutils.sh.

To be continued soon ...