Slimline Open Firmware - SLOF

Copyright (C) 2004, 2012  IBM Corporation


Index
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
1.0 Introduction to Open Firmware
2.0 Using the source code
2.1 Build process
2.2 Overview of the source code
2.4 Extending the Forth engine
3.0 Limitations
4.0 Submitting patches
5.0 Coding style


1.0 Introduction to Slimline Open Firmware
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

The IEEE Standard 1275-1994 [1], Standard for Boot (Initialization Configura-
tion) Firmware, Core Requirements and Practices, was the first non-proprietary
open standard for boot firmware that is usable on different processors and
buses. Firmware which complies with this standard (also known as Open Firmware)
includes a processor-independent device interface that allows add-in devices
to identify itself and to supply a single boot driver that can be used,
unchanged, on any CPU.  In addition, Open Firmware includes a user interface
with powerful scripting and debugging support and a client interface that
allows an operating system and its loaders to use Open Firmware services
during the configuration and initialization process.  Open Firmware stores
information about the hardware in a tree structure called the
"device tree".  This device tree supports multiple interconnected system
buses and offers a framework for "plug and play"-type auto configuration
across different buses.  It was designed to support a variety of different
processor Instruction Set Architectures (ISAs) and buses.

The full documentation of this Standard can be found in [1].

Slimline Open Firmware (SLOF) is now an implementation of the IEEE 1275
standard that is available under a BSD-style license. Please see the file
LICENSE for details.


2.0 Using the source code
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

This version of SLOF currently supports two major platforms ("boards" in the
SLOF jargon):

- js2x : The PowerPC 970 based systems JS20, JS21 and the PowerStation
- qemu : Used as partition firmware for pseries machines running on KVM/QEMU

The following sections will give you a short introduction about how to compile
and improve the source code.
Please read the file INSTALL for details about how to install the firmware on
your target system.


2.1 Build process
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

 To build SLOF you need:
  - Recent GNU tools, configured for powerpc64-linux
    - GCC: 3.3.3 and newer are known to work
    - Binutils: use a version as new as possible
    - Subversion (for retrieving the x86 emulator)

  - set the CROSS variable
    - something like export CROSS="powerpc64-unknown-linux-gnu-"
      when using a cross compiler
    or
    - export CROSS=""
      when using a native compiler

  - For building SLOF for the PowerStation, it is necessary to
    download a x86 emulator which is used to execute the BIOS
    of VGA card; to download the x86 emulator following steps are
    required:
    - cd other-licence/x86emu/
    - ./x86emu_download.sh      # this downloads the x86 emulator sources
    - cd -

  - Now you can compile the firmware.
    - For building SLOF for JS20, JS21 or the PowerStation, type:
        make js2x
      You also might want to build the takeover executable by typing:
        make -C board-js2x takeover
    - For building SLOF as the partition firmware for KVM/QEMU, type:
        make qemu
    The resulting ROM image "boot_rom.bin" can then be found in the main
    directory.


2.2 Overview of the source code
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

The SLOF source code is structured into the following directories:

- llfw : The Low-Level Firmware - this part is platform-specific firmware
         that is responsible to boot the system from the reset vector to a
         state where it is possible to run the Open Firmware Forth engine
         (i.e. it sets up the necessary CPU registers, intializes the memory,
         does some board-specific hardware configuration, etc.)

- slof : The code for the Open Firmware environment, including the Forth
         engine (called "Paflof") and the necessary Forth source files.

- rtas : The Run-Time Abstraction Services, which can be used by the operating
         system to access certain hardware without knowing the details.
         See [2] for a description of these services.

- clients : Code that runs on top of the Open Firmware client interface.
            Currently, there are two clients:
            - net-snk : Used for network bootloading (a TFTP client)
            - takeover : A separate binary that can be used for bootstrapping
                      SLOF on a JS20/JS21 (see FlashingSLOF.pdf for details).

- drivers : Driver code for various hardware (currently only NIC drivers).

- lib : Libraries with common code.

- romfs / tools : Tools that are required for building the firmware image.

- board-* : The board directories contain all the code that is unique to the
            corresponding platform.


2.3 The Open Firmware engine
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Open Firmware (OF) is based on the programming language Forth. 
SLOF use Paflof as the Forth engine, which was originally developed by
Segher Boessenkool.  Most parts of the Forth engine are implemented in C, by
using GNU extensions of ANSI C, (e.g. assigned goto, often misnamed "computed
goto"), resulting in a very efficient yet still quite portable engine.  

The basic Forth words, so-called primitives,  are implemented with 
a set of C macros.  A set of .in and .code files are provided, which
define the semantic of the Forth primitives.  A Perl script translates 
these files into valid C code, which will be compiled into the Forth engine.
The complete Forth system composes of the basic Forth primitives and
a set of Forth words, which are compiled during the start of the Forth
system.

Example:
Forth primitive 'dup'

	dup ( a -- a a) \ Duplicate top of stack element


prim.in:
	cod(DUP)

prim.code:
	PRIM(DUP) cell x = TOS; PUSH; TOS = x; MIRP

Generated code:

static cell xt_DUP[] = { { .a = xt_DOTICK }, { .c = "\000\003DUP" },
	 { .a = &&code_DUP }, };

code_DUP: { asm("#### " "DUP"); void *w = (cfa = (++ip)->a)->a;
	 cell x = (*dp); dp++; (*dp) = x; goto *w; }

Without going into detail, it can be seen, that the data stack is
implemented in C as an array of cells, where dp is the pointer to the top of
stack. 

For the implementation of the Open Firmware, most of the code is added as
Forth code and bound to the engine.  Also the system vectors for all kinds of
exceptions will be part of the image. Additionally a secondary boot-loader
or any other client application can be bound to the code as payload, 
e.g. diagnostics and test programs.

The Open Firmware image will be put together by the build 
process, with a loader at the start of the image. This loader
is called by Low Level Firmware and loads at boot time the Open 
Firmware to it's location in memory (see 1.3 Load process). Additionally 
a secondary boot loader or any other client application can be bound
to the code as payload.

The Low Level Firmware (LLFW) is responsible for setting up the 
system in an initial state. This task includes the setup of the 
CPUs, the system memory and all the buses as well as the serial port
itself.


2.4 Extending the Forth engine
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

In the following paragraphs it will be shown how to add
new primitive words (i.e., words implemented not by building
pre-existing Forth words together, but instead implemented in
C or assembler).  With this, it is possible to adapt SLOF to
the specific needs of different hardware and architectures.


To add primitives:

   For a new primitive, following steps have to be done:

   + Definition of primitive name in <arch>.in
     - cod(ABC) defines primitive ABC

     You can also use the following in a .in file, see existing
     code for how to use these:
     - con(ABC) defines constant ABC   
     - col(ABC) defines colon definition ABC
     - dfr(ABC) defines defer definition ABC

   + Definition of the primitives effects in <arch>.code
     - PRIM(ABC) ... MIRP

       The code for the primitive body is any C-code. With
       the macros of prim.code the data and return stack of 
       the Forth engine can be appropriately manipulated.


3.0 Limitations of this package
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

 On a JS20 the memory setup is very static and therefore there are
 only very few combinations of memory DIMM placement actually work.

 Known booting configurations:

    * 4x 256 MB (filling all slots) -- only "0.5 GB" reported.
    * 2x 1 GB, slots 3/4 -- only "0.5 GB" reported.

 Known failing configurations

    * 2x 256 MB, slots 3/4
    * 2x 256 MB, slots 1/2

 On a JS20 SLOF wil always report 0.5 GB even if there is much more memory
 available.

 On a JS21 all memory configurations should work.


4.0 Submitting patches
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Patches for SLOF should be made against https://github.com/aik/SLOF,
the master branch and posted to [email protected].
The patches must be signed using "Signed-off-by" tag with a real name to
confirm that you certify the Developer Certificate of Origin  Version 1.1,
see [3] for details.


5.0 Coding style
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

New C code submitted to SLOF should follow the coding style guidelines
for the Linux kernel [4] with the following exceptions:

- in the event that you require a specific width, use a standard type
  like int32_t, uint32_t, uint64_t, etc. Don't use Linux kernel internal
  types like u32, __u32 or __le32.

New Forth code should use 4 space indentations and no tabs. Patches for
the old code should keep the existing style which usually is
3 space indentation.

New assembly code submitted to SLOF should follow the coding style
guidelines for the Linux kernel [4], i.e. indent with tabs, not with spaces.


6.0 Tips and tricks
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

- Note that the serial console support is written with serial terminals in
  mind. Thus SLOF only looks for CR (0x0d) characters as line endings at the
  command prompt. If you want to connect to the console with a program like
  "ncat" for example, you have to make sure that it indeed sends CR characters
  instead of plain LF (0x0a) characters. For ncat that means that you have to
  use "nc -C ..." to enable the CR-LF mode, or even better, switch the terminal
  to raw mode instead, for example like this:

    stty raw -echo ; nc -U /path/to/chardev-socket-from-qemu ; stty sane


Documentation
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

[1] IEEE 1275-1994 Standard, Standard for Boot (Initialization Configuration)
    Firmware: Core Requirements and Practices

[2] PAPR Standard, Power.org(TM) Standard for Power Architecture(R) Platform
    Requirements (Workstation, Server), Version 2.4, December 7, 2009

[3] Developer Certificate of Origin Version 1.1
    http://developercertificate.org/

[4] Linux kernel coding style
    https://github.com/torvalds/linux/blob/master/Documentation/process/coding-style.rst