Building OFW for ARM: Difference between revisions

Latest revision as of 18:45, 11 August 2010

This page tells how to build Open Firmware for an ARM target system.

Host System Requirements

During the Open Firmware build process, the Forth system inside OFW is executed several times to extend itself. Those steps must run on the same CPU instruction set (in this case ARM) as the final target. So you must build OFW either on an ARM computer or by using an ARM instruction set emulator running on some other computer. [QEMU] works well for this purpose.

Compiler Setup for Native ARM Host

To compile on an ARM host, you need Linux on that host, along with the GNU toolchain (GCC), GNU make, and the Subversion version control system.

Compiler Setup for x86 Host

To compile on an x86 host, you need everything listed for the native host (Linux, GNU toolchain, GNU make, and Subversion), plus QEMU.

Note that the GNU toolchain that you need is the native x86 one, not an ARM cross-toolchain. OFW uses its own built-in ARM assembler. The native x86 toolchain is needed for the one-time step of compiling a small interface program that lets the OFW builder operate under Linux. It would be possible to eliminate that need by distributing a precompiled binary of that interface program, but in practice, most developers already have the native x86 toolchain already installed, so requiring it isn't a problem (as opposed to cross-toolchains, which can be a serious hassle to set up correctly).

To install QEMU:

Debian: sudo apt-get install qemu
Ubuntu 8.04 (Hardy): sudo apt-get install qemu
Ubuntu 9.10 (Karmic): sudo apt-get install qemu qemu-kvm-extras
Fedora: sudo yum install qemu-user

In general, the file that you need is "qemu-arm". You can use commands like (Fedora) "yum whatprovides qemu-arm" or (Debian/Ubuntu) "apt-file search qemu-arm" to work out which top-level package to install.

For Fedora, you might also need to temporarily disable SELinux security before OFW compilation:

 sudo /usr/sbin/setenforce 0

Building Open Firmware

Get the Open Firmware source:

 svn co svn://openfirmware.info/openfirmware

Build OFW:

 cd openfirmware/cpu/arm/mmp2/build
 make

The last line of the compilation output should say something like:

 --- Saving as ofw.rom

That tells you the name of the output file.

There are some other build directories like "cpu/arm/versatilepb/build". They work the same way - just type "make" in the directory. At present, those other versions are fairly rudimentary, containing core OFW plus a serial driver for interaction, but lacking drivers for most of the other hardware on those systems. Essentially, they are "quick ports to get an ok prompt". The MMP2 version is quite a bit more complete, on its way to being a fully-fledged OFW system. The MMP2 is Marvell's hardware development platform for their PXA688 chip, which is slated for use in the OLPC XO-1.75 system. We are using the MMP2 to get a jump-start on XO-1.75 development.

@@ Line 7: / Line 7: @@
 ==== Compiler Setup for Native ARM Host ====
-To compile on an ARM host, you need Linux on that host, along with the GNU toolchain (GCC and GNU make) plus the Subversion version control system.
+To compile on an ARM host, you need Linux on that host, along with the GNU toolchain (GCC), GNU make, and the Subversion version control system.
 ==== Compiler Setup for x86 Host  ====
-To compile on an x86 host, you need everything listed for the native host (Linux, GNU toolchain, and Subversion), plus QEMU.
+To compile on an x86 host, you need everything listed for the native host (Linux, GNU toolchain, GNU make, and Subversion), plus QEMU.
+Note that the GNU toolchain that you need is the native x86 one, not an ARM cross-toolchain.  OFW uses its own built-in ARM assembler. The native x86 toolchain is needed for the one-time step of compiling a small interface program that lets the OFW builder operate under Linux.  It would be possible to eliminate that need by distributing a precompiled binary of that interface program, but in practice, most developers already have the native x86 toolchain already installed, so requiring it isn't a problem (as opposed to cross-toolchains, which can be a serious hassle to set up correctly).
 To install QEMU:
 * Debian:  sudo apt-get install qemu
-* Ubuntu:  sudo apt-get install qemu
+* Ubuntu 8.04 (Hardy):  sudo apt-get install qemu
+* Ubuntu 9.10 (Karmic):  sudo apt-get install qemu qemu-kvm-extras
 * Fedora:  sudo yum install qemu-user
-For Fedora, you might also need to disable SELinux security during OFW compilation:
+In general, the file that you need is "qemu-arm".  You can use commands like (Fedora) "yum whatprovides qemu-arm" or (Debian/Ubuntu) "apt-file search qemu-arm" to work out which top-level package to install.
+For Fedora, you might also need to temporarily disable SELinux security before OFW compilation:
    sudo /usr/sbin/setenforce 0
@@ Line 36: / Line 41: @@
 The last line of the compilation output should say something like:
-   --- Saving as mmp2.rom
+   --- Saving as ofw.rom
 That tells you the name of the output file.
-There are some other build directories like "cpu/arm/versatilepb/build".  At the time of this writing, they have not been fixed up to "do the right thing" with the emulated build environment.  The Makefile in cpu/arm/mmp2/build is the current "gold standard" for the build automation.
+There are some other build directories like "cpu/arm/versatilepb/build".  They work the same way - just type "make" in the directory.  At present, those other versions are fairly rudimentary, containing core OFW plus a serial driver for interaction, but lacking drivers for most of the other hardware on those systems.  Essentially, they are "quick ports to get an ok prompt".  The MMP2 version is quite a bit more complete, on its way to being a fully-fledged OFW system.  The MMP2 is Marvell's hardware development platform for their PXA688 chip, which is slated for use in the OLPC XO-1.75 system.  We are using the MMP2 to get a jump-start on XO-1.75 development.