Building ARM toolchain - part 1: libs and binutils

Searching the internet for information on how to build arm toolchain from scratch I realize that it is very hard to find  information about this matter (and recent one even harder). I will try to fill this lack of information and try to build toolchain. My main goal is to use a component based on the GNU public license, and using them in as the newest version as it is possible.

What is toolchain ? (according to wikipedia):

In software, a toolchain is the set of programming tools that are used to create a product (typically another computer program or system of programs). The tools may be used in a chain, so that the output of each tool becomes the input for the next, but the term is used widely to refer to any set of linked development tools.

Requirements:
- Cross compiler, I create one using corsstool-ng and describe this process in previous post. I will use arm-unknown-linux-gnueabi as entry point compiler. - $TARGET is defined as my destination directory

export TARGET=/home/pietrushnic/sandbox/toolchain

Procedure:

1. Kernel header files:
1. clone linux git repository

   git clone git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git

2. install headers $TARGET is home of our toolchain

   make ARCH=arm INSTALL\_HDR\_PATH=\$TARGET/usr headers\_install

2. Libraries (gmp, mpfr, mpc)
   1. GMP (GNU Multiple Precision Arithmetic Library) - changeset used: 14765:0acae62fa162
      • As I said before I use "latest greatest" ;P version, and for gmp we can reach it using:

        hg clone http://gmplib.org:8000/gmp

      • create configuration files

        ./.bootstrap

      • configure

        ./configure --prefix=$TARGET/arm-x-tools --enable-shared=n \
        --enable-static

      • compile

        make
        make check
        make install

   2. MPFR (GNU Multiple Precision Floating-Point Reliably) - version: r8103
      • get latest version

        svn checkout svn://scm.gforge.inria.fr/svn/mpfr/trunk mpfr

      • create configuration file

        autoreconf -i

      • configure

        ./configure --prefix=$TARGET/arm-x-tools --enable-thread-safe \
        --with-gmp=$TARGET/arm-x-tools --disable-shared --enable-static

      • compile

        make
        make install

   3. MPC (Multiple Precision Complex)- version: r1146
      • checkout svn

        svn checkout svn://scm.gforge.inria.fr/svnroot/mpc/trunk mpc

      • create configuration file

        autoreconf -i

      • configure

        ./configure --prefix=$TARGET/arm-x-tools \
        --with-gmp=$TARGET/arm-x-tools --with-mpfr=$TARGET/arm-x-tools \
        --disable-shared --enable-static

      • compile

        make
        make install

3. Binutils - collection of a GNU binary tools:
   • checkout version from anonymous cvs

     cvs -z 9 -d :pserver:anoncvs@sourceware.org:/cvs/src login

     create directory for checkout

     mkdir binutils

     checkout sources

     cvs -z 9 -d :pserver:anoncvs@sourceware.org:/cvs/src co binutils

   • configure

     LDFLAGS="-Wl,-rpath -Wl,$TARGET/arm-x-tools/lib" ./configure \
     --build=x86_64-pc-linux-gnu --host=x86_64-pc-linux-gnu \
     --target=arm-unknown-linux-gnueabi --prefix=$TARGET/arm-x-tools \
     --disable-nls --disable-multilib --disable-werror --with-float=soft \
     --with-gmp=$TARGET/arm-x-tools --with-mpfr=$TARGET/arm-x-tools \
     --with-mpc=$TARGET/arm-x-tools --with-sysroot=$TARGET
     

   • compile

     make configure-host
     make
     make install

   • to check if everything was made correctly

     ldd $TARGET/arm-x-tools/bin/arm-unknown-linux-gnueabi-ldd

     it should show that it use library compiled previously by us:

     libz.so.1 => /home/pietrushnic/sandbox/toolchain/arm-x-tools/lib/libz.so.1 \
     (0x00007f0086cc5000)

This set gives us a solid base to build the compiler. However, it will be in the next section.

Quick build of arm-unknown-linux-gnueabi with crosstool-ng

You might be surprised at how much you have to make to correctly build arm-unknown-linux-gnueabi config based toolchain with crosstool-ng. As you can see examples of many open source projects, the man's work is a rare resource. The result of this economic fact is that the attempt to build configuration arm-unknown-linux-gnueabi is not a simple task and during an operation you can come across many problems. Although I am not afraid of problems and effectively try to fight them and of course sharing the results of my work. My build system parameters:
Debian GNU/Linux wheezy/sid 3.2.0-2-amd64

1. Clone crosstools-ng (you need mercurial)

   hg clone http://crosstool-ng.org/hg/crosstool-ng

2. Create temporary directory and run

   ct-ng arm-unknown-linux-eabi

   choose latest kernel (for me it was 3.2.9)
3. We probably want to change directory to which all stuff will be build (default is $HOME/x-tools):

4. ct-ng menuconfig

   And go to:

   Paths and misc options  ---> (${HOME}/x-tools/${CT_TARGET}) Prefix directory

   Change it according to your needs. Exit end save configuration.
5. Build (number depend on how many command we want to run simultaneously):

   ct-ng build.4

   It can take a lot of time. On my machine with 5k BogoMips it takes over 1h.

Problems that you can encounter:

1. gcj - latest changeset 2916:6f758ed4c0b9 have trouble finding gcj binary, which it show using following message:

   [ERROR]  Missing: 'x86_64-unknown-linux-gnu-gcj' or 
   'x86_64-unknown-linux-gnu-gcj' or 'gcj' : either needed!

   To workaround this install gcj and link binary like this:

   sudo ln -s /usr/bin/gcj-4.6 /usr/bin/gcj

2. duma - mentioned changeset also has problem with url to D.U.M.A library, apply below changes to workaround problems:

   --- a/scripts/build/debug/200-duma.sh   Mon Mar 12 21:19:26 2012 +0100
   +++ b/scripts/build/debug/200-duma.sh   Wed Mar 14 20:02:22 2012 +0100
   @@ -4,7 +4,7 @@
        # Downloading an non-existing file from sourceforge will give you an
        # HTML file containing an error message, instead of returning a 404.
        # Sigh...
   -    CT_GetFile "duma_${CT_DUMA_VERSION}" .tar.gz http://kent.dl.sourceforge.net/sourceforge/duma/
   +    CT_GetFile "duma_${CT_DUMA_VERSION}" .tar.gz http://downloads.sourceforge.net/project/duma/duma/2.5.15
        # Downloading from sourceforge may leave garbage, cleanup
        CT_DoExecLog ALL rm -f "${CT_TARBALLS_DIR}/showfiles.php"*
    }
   

3. mawk - if mawk return syntax error like this:

   mawk: scripts/gen-sorted.awk: line 19: regular expression compile failed (bad 
   class -- [], [^] or [)

   It could be fixed in two ways. First is to change line 19 in /path/to/tmp/dir/.build/src/glibc-2.9/scripts/gen-sorted.awk Is:

   sub(/\/[^/]+$/, "", subdir);

   Should be:

   sub(/\/[^\/]+$/, "", subdir);

   Or simply by installing gawk, reconfigure and recompile crosstools-ng.

This was my first post related to linux embedded enviroment. Hope it will be more. Enjoy!

Splitting my personal repository

After writing a few posts in which my workspace scripts could be useful, I realized that I should split blog works from the scripts that can be useful to others. From now materials related to the blog I will track in a separate repository and I'll add them as submodule into my environment. This implies that if we intend to deliver something to a broader audience this should include only those things which the recipient is expected and which we say that we provide. Any additional items should be removed. We get a lot of unwanted stuff today, at every opportunity. From an economic point of view, it is known that very often margin value law is used here i.e. by hypermarkets, I will touch on this subject on another occasion, related to economics.

Debugging coreboot in qemu environment - part 1

First of all I use testing version of Debian - wheezy. Clone coreboot repository:

git clone http://review.coreboot.org/p/coreboot

Conifigure:

cd coreboot; make menuconfig

Set:

Payload  ---> Add a payload ---> FILO
Payload  ---> FILO version ---> HEAD
Debugging  ---> Check PIRQ table consistency
Debugging  ---> Output verbose malloc debug messages
Debugging  ---> Output verbose ACPI debug messages
Debugging  ---> Enable debug messages for option ROM execution
Debugging  ---> Built-in low-level shell
Debugging  ---> Trace function calls

Try to build:

make

If everything builds correctly you can process. Sometimes there is need to use cross compiler. To build one:

cd util/crossgcc

./buildgcc

To explore coreboot code effectively I suggest to create tags and cscope database for coreboot. In my personal workspace I've got process that I go through before I start work (if you use my workspace configuration which is available @github you can follow below steps directly, if not adjust to your enviroment):

1. run vim ;)
2. :cd /path/to/code
3. s<Tab> (fuzzyfinder -> bookmark dir)
4. si (fuzzyfinder -> change dir)
5. sr (run ctags to generate tags and cscope to build symbol database - ctags -R;cscope -R -q -b -v)

After steps above we can start work with code. Run vim in coreboot dirctory. Type:
:e src/cpu/x86/16bit/reset16.inc
Put cursor over protected_start and press Ctrl-]. If everything goes ok you should jump to build/mainboard/emulation/qemu-x86/bootblock.s line 537.


In second article we dive into first phase of coreboot execution in emulated environment.  

"Human Action" Ludwig von Mises

"Human Action" Ludwig von Mises's work by some of called the pope of free market. IMHO is one of the best books on economics I read in my life. It is not easy, however, giving it time with certainly pays off. Allows to understand the basic laws which ruled the modern world of finance. Mises in a sensational way talk about the myths explaining what is inflation, why it has impact on our lives, what is and how works the supply of credit, and who depend most on the production of money without coverage (fiat money). I can say that until I read this book, I was lost in the dark. Thanks to it I received a weapon to overthrow ridiculous  world order concepts, absurd messages communicated in the media. I recommend: mises.org store - Human Action 

Nice small additions for my blog

Blog has changed slightly. I've added two new sections: social media and afk. In "social media" I will add links to my profiles on various social networking sites. You'll find my most recent development achievements (@github) and twitts on topics of interest to me (@twitter). The second section, "afk" (away from keyboard) is intended for the information of life beyond the keyboard, which in my case is not too extensive. I will publish here information about the books we read, watched movies and many other interesting things, which can be passionate away from the keyboard.