README

Old image format is based on binary files which can be basically anything,
preceded by a special header; see the definitions in include/image.h for
details; basically, the header defines the following image properties:

* Target Operating System (Provisions for OpenBSD, NetBSD, FreeBSD,
  4.4BSD, Linux, SVR4, Esix, Solaris, Irix, SCO, Dell, NCR, VxWorks,
  LynxOS, pSOS, QNX, RTEMS, INTEGRITY;
  Currently supported: Linux, NetBSD, VxWorks, QNX, RTEMS, LynxOS,
  INTEGRITY).
* Target CPU Architecture (Provisions for Alpha, ARM, AVR32, Intel x86,
  IA64, MIPS, NDS32, Nios II, PowerPC, IBM S390, SuperH, Sparc, Sparc 64 Bit;
  Currently supported: ARM, AVR32, Intel x86, MIPS, NDS32, Nios II, PowerPC).
* Compression Type (uncompressed, gzip, bzip2)
* Load Address
* Entry Point
* Image Name
* Image Timestamp

The header is marked by a special Magic Number, and both the header
and the data portions of the image are secured against corruption by
CRC32 checksums.


Linux Support:
==============

Although U-Boot should support any OS or standalone application
easily, the main focus has always been on Linux during the design of
U-Boot.

U-Boot includes many features that so far have been part of some
special "boot loader" code within the Linux kernel. Also, any
"initrd" images to be used are no longer part of one big Linux image;
instead, kernel and "initrd" are separate images. This implementation
serves several purposes:

- the same features can be used for other OS or standalone
  applications (for instance: using compressed images to reduce the
  Flash memory footprint)

- it becomes much easier to port new Linux kernel versions because
  lots of low-level, hardware dependent stuff are done by U-Boot

- the same Linux kernel image can now be used with different "initrd"
  images; of course this also means that different kernel images can
  be run with the same "initrd". This makes testing easier (you don't
  have to build a new "zImage.initrd" Linux image when you just
  change a file in your "initrd"). Also, a field-upgrade of the
  software is easier now.


Linux HOWTO:
============

Porting Linux to U-Boot based systems:
---------------------------------------

U-Boot cannot save you from doing all the necessary modifications to
configure the Linux device drivers for use with your target hardware
(no, we don't intend to provide a full virtual machine interface to
Linux :-).

But now you can ignore ALL boot loader code (in arch/powerpc/mbxboot).

Just make sure your machine specific header file (for instance
include/asm-ppc/tqm8xx.h) includes the same definition of the Board
Information structure as we define in include/asm-<arch>/u-boot.h,
and make sure that your definition of IMAP_ADDR uses the same value
as your U-Boot configuration in CONFIG_SYS_IMMR.


Configuring the Linux kernel:
-----------------------------

No specific requirements for U-Boot. Make sure you have some root
device (initial ramdisk, NFS) for your target system.


Building a Linux Image:
-----------------------

With U-Boot, "normal" build targets like "zImage" or "bzImage" are
not used. If you use recent kernel source, a new build target
"uImage" will exist which automatically builds an image usable by
U-Boot. Most older kernels also have support for a "pImage" target,
which was introduced for our predecessor project PPCBoot and uses a
100% compatible format.

Example:

	make TQM850L_config
	make oldconfig
	make dep
	make uImage

The "uImage" build target uses a special tool (in 'tools/mkimage') to
encapsulate a compressed Linux kernel image with header	 information,
CRC32 checksum etc. for use with U-Boot. This is what we are doing:

* build a standard "vmlinux" kernel image (in ELF binary format):

* convert the kernel into a raw binary image:

	${CROSS_COMPILE}-objcopy -O binary \
				 -R .note -R .comment \
				 -S vmlinux linux.bin

* compress the binary image:

	gzip -9 linux.bin

* package compressed binary image for U-Boot:

	mkimage -A ppc -O linux -T kernel -C gzip \
		-a 0 -e 0 -n "Linux Kernel Image" \
		-d linux.bin.gz uImage


The "mkimage" tool can also be used to create ramdisk images for use
with U-Boot, either separated from the Linux kernel image, or
combined into one file. "mkimage" encapsulates the images with a 64
byte header containing information about target architecture,
operating system, image type, compression method, entry points, time
stamp, CRC32 checksums, etc.

"mkimage" can be called in two ways: to verify existing images and
print the header information, or to build new images.

In the first form (with "-l" option) mkimage lists the information
contained in the header of an existing U-Boot image; this includes
checksum verification:

	tools/mkimage -l image
	  -l ==> list image header information

The second form (with "-d" option) is used to build a U-Boot image
from a "data file" which is used as image payload:

	tools/mkimage -A arch -O os -T type -C comp -a addr -e ep \
		      -n name -d data_file image
	  -A ==> set architecture to 'arch'
	  -O ==> set operating system to 'os'
	  -T ==> set image type to 'type'
	  -C ==> set compression type 'comp'
	  -a ==> set load address to 'addr' (hex)
	  -e ==> set entry point to 'ep' (hex)
	  -n ==> set image name to 'name'
	  -d ==> use image data from 'datafile'

Right now, all Linux kernels for PowerPC systems use the same load
address (0x00000000), but the entry point address depends on the
kernel version:

- 2.2.x kernels have the entry point at 0x0000000C,
- 2.3.x and later kernels have the entry point at 0x00000000.

So a typical call to build a U-Boot image would read:

	-> tools/mkimage -n '2.4.4 kernel for TQM850L' \
	> -A ppc -O linux -T kernel -C gzip -a 0 -e 0 \
	> -d /opt/elsk/ppc_8xx/usr/src/linux-2.4.4/arch/powerpc/coffboot/vmlinux.gz \
	> examples/uImage.TQM850L
	Image Name:   2.4.4 kernel for TQM850L
	Created:      Wed Jul 19 02:34:59 2000
	Image Type:   PowerPC Linux Kernel Image (gzip compressed)
	Data Size:    335725 Bytes = 327.86 kB = 0.32 MB
	Load Address: 0x00000000
	Entry Point:  0x00000000

To verify the contents of the image (or check for corruption):

	-> tools/mkimage -l examples/uImage.TQM850L
	Image Name:   2.4.4 kernel for TQM850L
	Created:      Wed Jul 19 02:34:59 2000
	Image Type:   PowerPC Linux Kernel Image (gzip compressed)
	Data Size:    335725 Bytes = 327.86 kB = 0.32 MB
	Load Address: 0x00000000
	Entry Point:  0x00000000

NOTE: for embedded systems where boot time is critical you can trade
speed for memory and install an UNCOMPRESSED image instead: this
needs more space in Flash, but boots much faster since it does not
need to be uncompressed:

	-> gunzip /opt/elsk/ppc_8xx/usr/src/linux-2.4.4/arch/powerpc/coffboot/vmlinux.gz
	-> tools/mkimage -n '2.4.4 kernel for TQM850L' \
	> -A ppc -O linux -T kernel -C none -a 0 -e 0 \
	> -d /opt/elsk/ppc_8xx/usr/src/linux-2.4.4/arch/powerpc/coffboot/vmlinux \
	> examples/uImage.TQM850L-uncompressed
	Image Name:   2.4.4 kernel for TQM850L
	Created:      Wed Jul 19 02:34:59 2000
	Image Type:   PowerPC Linux Kernel Image (uncompressed)
	Data Size:    792160 Bytes = 773.59 kB = 0.76 MB
	Load Address: 0x00000000
	Entry Point:  0x00000000


Similar you can build U-Boot images from a 'ramdisk.image.gz' file
when your kernel is intended to use an initial ramdisk:

	-> tools/mkimage -n 'Simple Ramdisk Image' \
	> -A ppc -O linux -T ramdisk -C gzip \
	> -d /LinuxPPC/images/SIMPLE-ramdisk.image.gz examples/simple-initrd
	Image Name:   Simple Ramdisk Image
	Created:      Wed Jan 12 14:01:50 2000
	Image Type:   PowerPC Linux RAMDisk Image (gzip compressed)
	Data Size:    566530 Bytes = 553.25 kB = 0.54 MB
	Load Address: 0x00000000
	Entry Point:  0x00000000


Installing a Linux Image:
-------------------------

To downloading a U-Boot image over the serial (console) interface,
you must convert the image to S-Record format:

	objcopy -I binary -O srec examples/image examples/image.srec

The 'objcopy' does not understand the information in the U-Boot
image header, so the resulting S-Record file will be relative to
address 0x00000000. To load it to a given address, you need to
specify the target address as 'offset' parameter with the 'loads'
command.

Example: install the image to address 0x40100000 (which on the
TQM8xxL is in the first Flash bank):

	=> erase 40100000 401FFFFF

	.......... done
	Erased 8 sectors

	=> loads 40100000
	## Ready for S-Record download ...
	~>examples/image.srec
	1 2 3 4 5 6 7 8 9 10 11 12 13 ...
	...
	15989 15990 15991 15992
	[file transfer complete]
	[connected]
	## Start Addr = 0x00000000


You can check the success of the download using the 'iminfo' command;
this includes a checksum verification so you can be sure no data
corruption happened:

	=> imi 40100000

	## Checking Image at 40100000 ...
	   Image Name:	 2.2.13 for initrd on TQM850L
	   Image Type:	 PowerPC Linux Kernel Image (gzip compressed)
	   Data Size:	 335725 Bytes = 327 kB = 0 MB
	   Load Address: 00000000
	   Entry Point:	 0000000c
	   Verifying Checksum ... OK


Boot Linux:
-----------

The "bootm" command is used to boot an application that is stored in
memory (RAM or Flash). In case of a Linux kernel image, the contents
of the "bootargs" environment variable is passed to the kernel as
parameters. You can check and modify this variable using the
"printenv" and "setenv" commands:


	=> printenv bootargs
	bootargs=root=/dev/ram

	=> setenv bootargs root=/dev/nfs rw nfsroot=10.0.0.2:/LinuxPPC nfsaddrs=10.0.0.99:10.0.0.2

	=> printenv bootargs
	bootargs=root=/dev/nfs rw nfsroot=10.0.0.2:/LinuxPPC nfsaddrs=10.0.0.99:10.0.0.2

	=> bootm 40020000
	## Booting Linux kernel at 40020000 ...
	   Image Name:	 2.2.13 for NFS on TQM850L
	   Image Type:	 PowerPC Linux Kernel Image (gzip compressed)
	   Data Size:	 381681 Bytes = 372 kB = 0 MB
	   Load Address: 00000000
	   Entry Point:	 0000000c
	   Verifying Checksum ... OK
	   Uncompressing Kernel Image ... OK
	Linux version 2.2.13 (wd@denx.local.net) (gcc version 2.95.2 19991024 (release)) #1 Wed Jul 19 02:35:17 MEST 2000
	Boot arguments: root=/dev/nfs rw nfsroot=10.0.0.2:/LinuxPPC nfsaddrs=10.0.0.99:10.0.0.2
	time_init: decrementer frequency = 187500000/60
	Calibrating delay loop... 49.77 BogoMIPS
	Memory: 15208k available (700k kernel code, 444k data, 32k init) [c0000000,c1000000]
	...

If you want to boot a Linux kernel with initial RAM disk, you pass
the memory addresses of both the kernel and the initrd image (PPBCOOT
format!) to the "bootm" command:

	=> imi 40100000 40200000

	## Checking Image at 40100000 ...
	   Image Name:	 2.2.13 for initrd on TQM850L
	   Image Type:	 PowerPC Linux Kernel Image (gzip compressed)
	   Data Size:	 335725 Bytes = 327 kB = 0 MB
	   Load Address: 00000000
	   Entry Point:	 0000000c
	   Verifying Checksum ... OK

	## Checking Image at 40200000 ...
	   Image Name:	 Simple Ramdisk Image
	   Image Type:	 PowerPC Linux RAMDisk Image (gzip compressed)
	   Data Size:	 566530 Bytes = 553 kB = 0 MB
	   Load Address: 00000000
	   Entry Point:	 00000000
	   Verifying Checksum ... OK

	=> bootm 40100000 40200000
	## Booting Linux kernel at 40100000 ...
	   Image Name:	 2.2.13 for initrd on TQM850L
	   Image Type:	 PowerPC Linux Kernel Image (gzip compressed)
	   Data Size:	 335725 Bytes = 327 kB = 0 MB
	   Load Address: 00000000
	   Entry Point:	 0000000c
	   Verifying Checksum ... OK
	   Uncompressing Kernel Image ... OK
	## Loading RAMDisk Image at 40200000 ...
	   Image Name:	 Simple Ramdisk Image
	   Image Type:	 PowerPC Linux RAMDisk Image (gzip compressed)
	   Data Size:	 566530 Bytes = 553 kB = 0 MB
	   Load Address: 00000000
	   Entry Point:	 00000000
	   Verifying Checksum ... OK
	   Loading Ramdisk ... OK
	Linux version 2.2.13 (wd@denx.local.net) (gcc version 2.95.2 19991024 (release)) #1 Wed Jul 19 02:32:08 MEST 2000
	Boot arguments: root=/dev/ram
	time_init: decrementer frequency = 187500000/60
	Calibrating delay loop... 49.77 BogoMIPS
	...
	RAMDISK: Compressed image found at block 0
	VFS: Mounted root (ext2 filesystem).

	bash#

Boot Linux and pass a flat device tree:
-----------

First, U-Boot must be compiled with the appropriate defines. See the section
titled "Linux Kernel Interface" above for a more in depth explanation. The
following is an example of how to start a kernel and pass an updated
flat device tree:

=> print oftaddr
oftaddr=0x300000
=> print oft
oft=oftrees/mpc8540ads.dtb
=> tftp $oftaddr $oft
Speed: 1000, full duplex
Using TSEC0 device
TFTP from server 192.168.1.1; our IP address is 192.168.1.101
Filename 'oftrees/mpc8540ads.dtb'.
Load address: 0x300000
Loading: #
done
Bytes transferred = 4106 (100a hex)
=> tftp $loadaddr $bootfile
Speed: 1000, full duplex
Using TSEC0 device
TFTP from server 192.168.1.1; our IP address is 192.168.1.2
Filename 'uImage'.
Load address: 0x200000
Loading:############
done
Bytes transferred = 1029407 (fb51f hex)
=> print loadaddr
loadaddr=200000
=> print oftaddr
oftaddr=0x300000
=> bootm $loadaddr - $oftaddr
## Booting image at 00200000 ...
   Image Name:	 Linux-2.6.17-dirty
   Image Type:	 PowerPC Linux Kernel Image (gzip compressed)
   Data Size:	 1029343 Bytes = 1005.2 kB
   Load Address: 00000000
   Entry Point:	 00000000
   Verifying Checksum ... OK
   Uncompressing Kernel Image ... OK
Booting using flat device tree at 0x300000
Using MPC85xx ADS machine description
Memory CAM mapping: CAM0=256Mb, CAM1=256Mb, CAM2=0Mb residual: 0Mb
[snip]


More About U-Boot Image Types:
------------------------------

U-Boot supports the following image types:

   "Standalone Programs" are directly runnable in the environment
	provided by U-Boot; it is expected that (if they behave
	well) you can continue to work in U-Boot after return from
	the Standalone Program.
   "OS Kernel Images" are usually images of some Embedded OS which
	will take over control completely. Usually these programs
	will install their own set of exception handlers, device
	drivers, set up the MMU, etc. - this means, that you cannot
	expect to re-enter U-Boot except by resetting the CPU.
   "RAMDisk Images" are more or less just data blocks, and their
	parameters (address, size) are passed to an OS kernel that is
	being started.
   "Multi-File Images" contain several images, typically an OS
	(Linux) kernel image and one or more data images like
	RAMDisks. This construct is useful for instance when you want
	to boot over the network using BOOTP etc., where the boot
	server provides just a single image file, but you want to get
	for instance an OS kernel and a RAMDisk image.

	"Multi-File Images" start with a list of image sizes, each
	image size (in bytes) specified by an "uint32_t" in network
	byte order. This list is terminated by an "(uint32_t)0".
	Immediately after the terminating 0 follow the images, one by
	one, all aligned on "uint32_t" boundaries (size rounded up to
	a multiple of 4 bytes).

   "Firmware Images" are binary images containing firmware (like
	U-Boot or FPGA images) which usually will be programmed to
	flash memory.

   "Script files" are command sequences that will be executed by
	U-Boot's command interpreter; this feature is especially
	useful when you configure U-Boot to use a real shell (hush)
	as command interpreter.

Booting the Linux zImage:
-------------------------

On some platforms, it's possible to boot Linux zImage. This is done
using the "bootz" command. The syntax of "bootz" command is the same
as the syntax of "bootm" command.


Standalone HOWTO:
=================

One of the features of U-Boot is that you can dynamically load and
run "standalone" applications, which can use some resources of
U-Boot like console I/O functions or interrupt services.

Two simple examples are included with the sources:

"Hello World" Demo:
-------------------

'examples/hello_world.c' contains a small "Hello World" Demo
application; it is automatically compiled when you build U-Boot.
It's configured to run at address 0x00040004, so you can play with it
like that:

	=> loads
	## Ready for S-Record download ...
	~>examples/hello_world.srec
	1 2 3 4 5 6 7 8 9 10 11 ...
	[file transfer complete]
	[connected]
	## Start Addr = 0x00040004

	=> go 40004 Hello World! This is a test.
	## Starting application at 0x00040004 ...
	Hello World
	argc = 7
	argv[0] = "40004"
	argv[1] = "Hello"
	argv[2] = "World!"
	argv[3] = "This"
	argv[4] = "is"
	argv[5] = "a"
	argv[6] = "test."
	argv[7] = "<NULL>"
	Hit any key to exit ...

	## Application terminated, rc = 0x0

Another example, which demonstrates how to register a CPM interrupt
handler with the U-Boot code, can be found in 'examples/timer.c'.
Here, a CPM timer is set up to generate an interrupt every second.
The interrupt service routine is trivial, just printing a '.'
character, but this is just a demo program. The application can be
controlled by the following keys:

	? - print current values og the CPM Timer registers
	b - enable interrupts and start timer
	e - stop timer and disable interrupts
	q - quit application

	=> loads
	## Ready for S-Record download ...
	~>examples/timer.srec
	1 2 3 4 5 6 7 8 9 10 11 ...
	[file transfer complete]
	[connected]
	## Start Addr = 0x00040004

	=> go 40004
	## Starting application at 0x00040004 ...
	TIMERS=0xfff00980
	Using timer 1
	  tgcr @ 0xfff00980, tmr @ 0xfff00990, trr @ 0xfff00994, tcr @ 0xfff00998, tcn @ 0xfff0099c, ter @ 0xfff009b0

Hit 'b':
	[q, b, e, ?] Set interval 1000000 us
	Enabling timer
Hit '?':
	[q, b, e, ?] ........
	tgcr=0x1, tmr=0xff1c, trr=0x3d09, tcr=0x0, tcn=0xef6, ter=0x0
Hit '?':
	[q, b, e, ?] .
	tgcr=0x1, tmr=0xff1c, trr=0x3d09, tcr=0x0, tcn=0x2ad4, ter=0x0
Hit '?':
	[q, b, e, ?] .
	tgcr=0x1, tmr=0xff1c, trr=0x3d09, tcr=0x0, tcn=0x1efc, ter=0x0
Hit '?':
	[q, b, e, ?] .
	tgcr=0x1, tmr=0xff1c, trr=0x3d09, tcr=0x0, tcn=0x169d, ter=0x0
Hit 'e':
	[q, b, e, ?] ...Stopping timer
Hit 'q':
	[q, b, e, ?] ## Application terminated, rc = 0x0


Minicom warning:
================

Over time, many people have reported problems when trying to use the
"minicom" terminal emulation program for serial download. I (wd)
consider minicom to be broken, and recommend not to use it. Under
Unix, I recommend to use C-Kermit for general purpose use (and
especially for kermit binary protocol download ("loadb" command), and
use "cu" for S-Record download ("loads" command).

Nevertheless, if you absolutely want to use it try adding this
configuration to your "File transfer protocols" section:

	   Name	   Program			Name U/D FullScr IO-Red. Multi
	X  kermit  /usr/bin/kermit -i -l %l -s	 Y    U	   Y	   N	  N
	Y  kermit  /usr/bin/kermit -i -l %l -r	 N    D	   Y	   N	  N


NetBSD Notes:
=============

Starting at version 0.9.2, U-Boot supports NetBSD both as host
(build U-Boot) and target system (boots NetBSD/mpc8xx).

Building requires a cross environment; it is known to work on
NetBSD/i386 with the cross-powerpc-netbsd-1.3 package (you will also
need gmake since the Makefiles are not compatible with BSD make).
Note that the cross-powerpc package does not install include files;
attempting to build U-Boot will fail because <machine/ansi.h> is
missing.  This file has to be installed and patched manually:

	# cd /usr/pkg/cross/powerpc-netbsd/include
	# mkdir powerpc
	# ln -s powerpc machine
	# cp /usr/src/sys/arch/powerpc/include/ansi.h powerpc/ansi.h
	# ${EDIT} powerpc/ansi.h	## must remove __va_list, _BSD_VA_LIST

Native builds *don't* work due to incompatibilities between native
and U-Boot include files.

Booting assumes that (the first part of) the image booted is a
stage-2 loader which in turn loads and then invokes the kernel
proper. Loader sources will eventually appear in the NetBSD source
tree (probably in sys/arc/mpc8xx/stand/u-boot_stage2/); in the
meantime, see ftp://ftp.denx.de/pub/u-boot/ppcboot_stage2.tar.gz


Implementation Internals:
=========================

The following is not intended to be a complete description of every
implementation detail. However, it should help to understand the
inner workings of U-Boot and make it easier to port it to custom
hardware.


Initial Stack, Global Data:
---------------------------

The implementation of U-Boot is complicated by the fact that U-Boot
starts running out of ROM (flash memory), usually without access to
system RAM (because the memory controller is not initialized yet).
This means that we don't have writable Data or BSS segments, and BSS
is not initialized as zero. To be able to get a C environment working
at all, we have to allocate at least a minimal stack. Implementation
options for this are defined and restricted by the CPU used: Some CPU
models provide on-chip memory (like the IMMR area on MPC8xx and
MPC826x processors), on others (parts of) the data cache can be
locked as (mis-) used as memory, etc.

	Chris Hallinan posted a good summary of these issues to the
	U-Boot mailing list:

	Subject: RE: [U-Boot-Users] RE: More On Memory Bank x (nothingness)?
	From: "Chris Hallinan" <clh@net1plus.com>
	Date: Mon, 10 Feb 2003 16:43:46 -0500 (22:43 MET)
	...

	Correct me if I'm wrong, folks, but the way I understand it
	is this: Using DCACHE as initial RAM for Stack, etc, does not
	require any physical RAM backing up the cache. The cleverness
	is that the cache is being used as a temporary supply of
	necessary storage before the SDRAM controller is setup. It's
	beyond the scope of this list to explain the details, but you
	can see how this works by studying the cache architecture and
	operation in the architecture and processor-specific manuals.

	OCM is On Chip Memory, which I believe the 405GP has 4K. It
	is another option for the system designer to use as an
	initial stack/RAM area prior to SDRAM being available. Either
	option should work for you. Using CS 4 should be fine if your
	board designers haven't used it for something that would
	cause you grief during the initial boot! It is frequently not
	used.

	CONFIG_SYS_INIT_RAM_ADDR should be somewhere that won't interfere
	with your processor/board/system design. The default value
	you will find in any recent u-boot distribution in
	walnut.h should work for you. I'd set it to a value larger
	than your SDRAM module. If you have a 64MB SDRAM module, set
	it above 400_0000. Just make sure your board has no resources
	that are supposed to respond to that address! That code in
	start.S has been around a while and should work as is when
	you get the config right.

	-Chris Hallinan
	DS4.COM, Inc.

It is essential to remember this, since it has some impact on the C
code for the initialization procedures:

* Initialized global data (data segment) is read-only. Do not attempt
  to write it.

* Do not use any uninitialized global data (or implicitely initialized
  as zero data - BSS segment) at all - this is undefined, initiali-
  zation is performed later (when relocating to RAM).

* Stack space is very limited. Avoid big data buffers or things like
  that.

Having only the stack as writable memory limits means we cannot use
normal global data to share information beween the code. But it
turned out that the implementation of U-Boot can be greatly
simplified by making a global data structure (gd_t) available to all
functions. We could pass a pointer to this data as argument to _all_
functions, but this would bloat the code. Instead we use a feature of
the GCC compiler (Global Register Variables) to share the data: we
place a pointer (gd) to the global data into a register which we
reserve for this purpose.

When choosing a register for such a purpose we are restricted by the
relevant  (E)ABI  specifications for the current architecture, and by
GCC's implementation.

For PowerPC, the following registers have specific use:
	R1:	stack pointer
	R2:	reserved for system use
	R3-R4:	parameter passing and return values
	R5-R10: parameter passing
	R13:	small data area pointer
	R30:	GOT pointer
	R31:	frame pointer

	(U-Boot also uses R12 as internal GOT pointer. r12
	is a volatile register so r12 needs to be reset when
	going back and forth between asm and C)

    ==> U-Boot will use R2 to hold a pointer to the global data

    Note: on PPC, we could use a static initializer (since the
    address of the global data structure is known at compile time),
    but it turned out that reserving a register results in somewhat
    smaller code - although the code savings are not that big (on
    average for all boards 752 bytes for the whole U-Boot image,
    624 text + 127 data).

On Blackfin, the normal C ABI (except for P3) is followed as documented here:
	http://docs.blackfin.uclinux.org/doku.php?id=application_binary_interface

    ==> U-Boot will use P3 to hold a pointer to the global data

On ARM, the following registers are used:

	R0:	function argument word/integer result
	R1-R3:	function argument word
	R9:	GOT pointer
	R10:	stack limit (used only if stack checking if enabled)
	R11:	argument (frame) pointer
	R12:	temporary workspace
	R13:	stack pointer
	R14:	link register
	R15:	program counter

    ==> U-Boot will use R8 to hold a pointer to the global data

On Nios II, the ABI is documented here:
	http://www.altera.com/literature/hb/nios2/n2cpu_nii51016.pdf

    ==> U-Boot will use gp to hold a pointer to the global data

    Note: on Nios II, we give "-G0" option to gcc and don't use gp
    to access small data sections, so gp is free.

On NDS32, the following registers are used:

	R0-R1:	argument/return
	R2-R5:	argument
	R15:	temporary register for assembler
	R16:	trampoline register
	R28:	frame pointer (FP)
	R29:	global pointer (GP)
	R30:	link register (LP)
	R31:	stack pointer (SP)
	PC:	program counter (PC)

    ==> U-Boot will use R10 to hold a pointer to the global data

NOTE: DECLARE_GLOBAL_DATA_PTR must be used with file-global scope,
or current versions of GCC may "optimize" the code too much.

Memory Management:
------------------

U-Boot runs in system state and uses physical addresses, i.e. the
MMU is not used either for address mapping nor for memory protection.

The available memory is mapped to fixed addresses using the memory
controller. In this process, a contiguous block is formed for each
memory type (Flash, SDRAM, SRAM), even when it consists of several
physical memory banks.

U-Boot is installed in the first 128 kB of the first Flash bank (on
TQM8xxL modules this is the range 0x40000000 ... 0x4001FFFF). After
booting and sizing and initializing DRAM, the code relocates itself
to the upper end of DRAM. Immediately below the U-Boot code some
memory is reserved for use by malloc() [see CONFIG_SYS_MALLOC_LEN
configuration setting]. Below that, a structure with global Board
Info data is placed, followed by the stack (growing downward).

Additionally, some exception handler code is copied to the low 8 kB
of DRAM (0x00000000 ... 0x00001FFF).

So a typical memory configuration with 16 MB of DRAM could look like
this:

	0x0000 0000	Exception Vector code
	      :
	0x0000 1FFF
	0x0000 2000	Free for Application Use
	      :
	      :

	      :
	      :
	0x00FB FF20	Monitor Stack (Growing downward)
	0x00FB FFAC	Board Info Data and permanent copy of global data
	0x00FC 0000	Malloc Arena
	      :
	0x00FD FFFF
	0x00FE 0000	RAM Copy of Monitor Code
	...		eventually: LCD or video framebuffer
	...		eventually: pRAM (Protected RAM - unchanged by reset)
	0x00FF FFFF	[End of RAM]


System Initialization:
----------------------

In the reset configuration, U-Boot starts at the reset entry point
(on most PowerPC systems at address 0x00000100). Because of the reset
configuration for CS0# this is a mirror of the onboard Flash memory.
To be able to re-map memory U-Boot then jumps to its link address.
To be able to implement the initialization code in C, a (small!)
initial stack is set up in the internal Dual Ported RAM (in case CPUs
which provide such a feature like MPC8xx or MPC8260), or in a locked
part of the data cache. After that, U-Boot initializes the CPU core,
the caches and the SIU.

Next, all (potentially) available memory banks are mapped using a
preliminary mapping. For example, we put them on 512 MB boundaries
(multiples of 0x20000000: SDRAM on 0x00000000 and 0x20000000, Flash
on 0x40000000 and 0x60000000, SRAM on 0x80000000). Then UPM A is
programmed for SDRAM access. Using the temporary configuration, a
simple memory test is run that determines the size of the SDRAM
banks.

When there is more than one SDRAM bank, and the banks are of
different size, the largest is mapped first. For equal size, the first
bank (CS2#) is mapped first. The first mapping is always for address
0x00000000, with any additional banks following immediately to create
contiguous memory starting from 0.

Then, the monitor installs itself at the upper end of the SDRAM area
and allocates memory for use by malloc() and for the global Board
Info data; also, the exception vector code is copied to the low RAM
pages, and the final stack is set up.

Only after this relocation will you have a "normal" C environment;
until that you are restricted in several ways, mostly because you are
running from ROM, and because the code will have to be relocated to a
new address in RAM.


U-Boot Porting Guide:
----------------------

[Based on messages by Jerry Van Baren in the U-Boot-Users mailing
list, October 2002]


int main(int argc, char *argv[])
{
	sighandler_t no_more_time;

	signal(SIGALRM, no_more_time);
	alarm(PROJECT_DEADLINE - toSec (3 * WEEK));

	if (available_money > available_manpower) {
		Pay consultant to port U-Boot;
		return 0;
	}

	Download latest U-Boot source;

	Subscribe to u-boot mailing list;

	if (clueless)
		email("Hi, I am new to U-Boot, how do I get started?");

	while (learning) {
		Read the README file in the top level directory;
		Read http://www.denx.de/twiki/bin/view/DULG/Manual;
		Read applicable doc/*.README;
		Read the source, Luke;
		/* find . -name "*.[chS]" | xargs grep -i <keyword> */
	}

	if (available_money > toLocalCurrency ($2500))
		Buy a BDI3000;
	else
		Add a lot of aggravation and time;

	if (a similar board exists) {	/* hopefully... */
		cp -a board/<similar> board/<myboard>
		cp include/configs/<similar>.h include/configs/<myboard>.h
	} else {
		Create your own board support subdirectory;
		Create your own board include/configs/<myboard>.h file;
	}
	Edit new board/<myboard> files
	Edit new include/configs/<myboard>.h

	while (!accepted) {
		while (!running) {
			do {
				Add / modify source code;
			} until (compiles);
			Debug;
			if (clueless)
				email("Hi, I am having problems...");
		}
		Send patch file to the U-Boot email list;
		if (reasonable critiques)
			Incorporate improvements from email list code review;
		else
			Defend code as written;
	}

	return 0;
}

void no_more_time (int sig)
{
      hire_a_guru();
}


Coding Standards:
-----------------

All contributions to U-Boot should conform to the Linux kernel
coding style; see the file "Documentation/CodingStyle" and the script
"scripts/Lindent" in your Linux kernel source directory.

Source files originating from a different project (for example the
MTD subsystem) are generally exempt from these guidelines and are not
reformated to ease subsequent migration to newer versions of those
sources.

Please note that U-Boot is implemented in C (and to some small parts in
Assembler); no C++ is used, so please do not use C++ style comments (//)
in your code.

Please also stick to the following formatting rules:
- remove any trailing white space
- use TAB characters for indentation and vertical alignment, not spaces
- make sure NOT to use DOS '\r\n' line feeds
- do not add more than 2 consecutive empty lines to source files
- do not add trailing empty lines to source files

Submissions which do not conform to the standards may be returned
with a request to reformat the changes.


Submitting Patches:
-------------------

Since the number of patches for U-Boot is growing, we need to
establish some rules. Submissions which do not conform to these rules
may be rejected, even when they contain important and valuable stuff.

Please see http://www.denx.de/wiki/U-Boot/Patches for details.

Patches shall be sent to the u-boot mailing list <u-boot@lists.denx.de>;
see http://lists.denx.de/mailman/listinfo/u-boot

When you send a patch, please include the following information with
it:

* For bug fixes: a description of the bug and how your patch fixes
  this bug. Please try to include a way of demonstrating that the
  patch actually fixes something.

* For new features: a description of the feature and your
  implementation.

* A CHANGELOG entry as plaintext (separate from the patch)

* For major contributions, your entry to the CREDITS file

* When you add support for a new board, don't forget to add this
  board to the MAINTAINERS file, too.

* If your patch adds new configuration options, don't forget to
  document these in the README file.

* The patch itself. If you are using git (which is *strongly*
  recommended) you can easily generate the patch using the
  "git format-patch". If you then use "git send-email" to send it to
  the U-Boot mailing list, you will avoid most of the common problems
  with some other mail clients.

  If you cannot use git, use "diff -purN OLD NEW". If your version of
  diff does not support these options, then get the latest version of
  GNU diff.

  The current directory when running this command shall be the parent
  directory of the U-Boot source tree (i. e. please make sure that
  your patch includes sufficient directory information for the
  affected files).

  We prefer patches as plain text. MIME attachments are discouraged,
  and compressed attachments must not be used.

* If one logical set of modifications affects or creates several
  files, all these changes shall be submitted in a SINGLE patch file.

* Changesets that contain different, unrelated modifications shall be
  submitted as SEPARATE patches, one patch per changeset.


Notes:

* Before sending the patch, run the MAKEALL script on your patched
  source tree and make sure that no errors or warnings are reported
  for any of the boards.

* Keep your modifications to the necessary minimum: A patch
  containing several unrelated changes or arbitrary reformats will be
  returned with a request to re-formatting / split it.

* If you modify existing code, make sure that your new code does not
  add to the memory footprint of the code ;-) Small is beautiful!
  When adding new features, these should compile conditionally only
  (using #ifdef), and the resulting code with the new feature
  disabled must not need more memory than the old code without your
  modification.

* Remember that there is a size limit of 100 kB per message on the
  u-boot mailing list. Bigger patches will be moderated. If they are
  reasonable and not too big, they will be acknowledged. But patches
  bigger than the size limit should be avoided.