1
User’s Manual for the
Boundary Devices
R
Neonꢀ board
December 28, 2005
December 28, 2005
Revision 2.8
CONTENTS
3
Contents
2
5
5
5
5
6
6
8
9
9
26
December 28, 2005
Revision 2.8
CONTENTS
4
34
December 28, 2005
Revision 2.8
5
2 Intended Audience
This document aims to provide the information needed to integrate the
R
Neonꢀ board into your application. As such, it addresses both hardware
and software integration.
3 Overview of features
R
The following are highlights of the Neonꢀ board.
• Available with Windows Ce or Linux Operating Systems
• 400 MHz Intel PXA-255 CPU
• 32 or 64MB SDRAM
• 8 or 32MB Intel StrataFlash (tm) EEPROM
• Active Matrix LCD Support,
• Including Full-Motion Video
• STN Passive LCD Display Support
• 4 or 5-Wire Resistive Touch-Screen Support
• 44KHz Stereo 16-Bit Audio Output, for Headphones or Speakers
• 44KHz Monaural Audio Input (microphone)
• 1 RS-232 or TTL Serial Port
• 1 USB 1.1 Slave Port
• 1 USB 1.1 Master Port
• Built-in 10/100 Ethernet Controller,
• Built-in Interface for Magnetic Stripe Readers and Printers
• MMC Slot for Expanded Storage
• General Purpose I/O for Device Control
• Built-in Switching Power Supply for 5V DC Input
• JTAG Interface
• Customized Versions Available
4 Hardware feature
4.1 Layout
R
options for use in your application. Note that some of these may not be
populated on an evaluation or production board.
December 28, 2005
Revision 2.8
4.2 Mounting
6
Figure 1: Neon board
4.2 Mounting
R
R
The Neonꢀ board measures 2.75” by 6.75”, slightly larger than the Hitachiꢀ
6.2” display, to allow for easy mounting.
There are four mounting holes 1/4” from each edge in each of the four
corners, and the holes are 1/8” in diameter.
4.3 Connector reference
R
The following is a list of all connector part numbers used on the Neonꢀ
platform for use in identifying mating parts for your application. Note that
Boundary Deviceswill periodically switch vendors for these parts, but will
notify you of any changes that require a new mating part.
December 28, 2005
Revision 2.8
4.4 Electrical characteristics
8
Description
USB Master
USB Slave
I2C
Manufacturer Part
FCI
87520-0010B
KS-001-BNW
68897-001
SINGATRON
FCI
Stereo Audio
Backlight inverter Molex
Halo
Singatron
HFJ11-2450E
2SJ-43723N13
53048-0210
MMC/SD
TFT Display
Touch Screen
Serial Port
JTAG
AVX
14 5638 009 511 862
Molex
FCI
Molex
52207-0590
68897-001
53048-0810
4.4 Electrical characteristics
December 28, 2005
Revision 2.8
9
5 Software features
R
As provided by Boundary Devices, the Neonꢀ board supports either Win-
R
dows CE 5ꢀ or Linux.
the use of either operating system.
5.1 Das U-Boot
Windows CE that supports a wide variety of options for loading your Op-
erating System and application.
Boundary Devices ships U-Boot both as a binary image and as source
code in the form of a patch that adds support for either Neon or BD-2003
devices.
The binary image may be burned directly to sector zero of the on-board
flash.
for cross-compilation. The following section will detail the requirements and
steps for building.
5.1.1 Requirements for building under Linux
nents will generally be available on a GNU/Linux system.
The three pieces which may not commonly be installed are the bzip2
Boundary Devices typically uses GCC-2.95.3 to create U-Boot images,
since that matches what we use to build the Linux image to run on the
nice alternative.
There are two primary requirements for building under Windows.
operating system. Since the Cygwin installer allows components to be se-
lected individually, the following list shows the requirements for building a
R
incomplete, but these should be the only required items which differ from
the Cygwin installation defaults.
December 28, 2005
Revision 2.8
5.1 Das U-Boot
10
Base/diffutils
Devel/binutils
Devel/gcc
Devel/make
Devel/patchutils
Utils/bzip2
Web/wget
The second requirement for building is the X-Scale cross-compiler itself.
As of this writing, Boundary Devices currently uses the GCC-3.4.3 package
5.1.3 General build steps
Quick start:
bzcat u-boot-1.1.2.tar.bz2
|
tar -xvf
-
gunzip u-boot-2005-10-21.patch.gz
patch -p0 <u-boot-2005-10-21.patch
cd u-boot-1.1.2
CROSS_COMPILE=arm-elf- make neon_config
-------- U-Boot Boundary Devices Specific Configuration Script --------
Choose display type (DA640X240 DA320X240 DA800X480 DA640X480 DA240X320 DA1024X768) []: DA1024X768
answer
Choose hardware type (NEONB NEON BD2003) [NEON]:
answer
Choose software type (WINCE LINUX) []: WINCE
answer
Include minidebug (y n) []:
answer
y
CPU speed (100 200 300 400) []: 400
answer
Configuration successful.
make
Explanation.
R
support for the Neonꢀ and BD-2003 devices.
R
The last two lines configure for the Neonꢀ board itself, and finally, build
a U-Boot binary. The prompts allow you to select the compile-time defaults
for the display, operating system, and CPU speed. Including minidebug
in your U-Boot image allows you to access the debugger while developing
U-Boot scripts.
When complete, you’ll find a file named u-boot.bin in your u-boot-1.1.2
directory.
5.1.4 Tailoring U-Boot for your application
The Boundary Devices patches (uboot neon bd2003.diff) make a variety of
decisions about the boot process which may not match with the needs of
December 28, 2005
Revision 2.8
5.1 Das U-Boot
your application.
11
In general, the file u-boot-1.1.2/include/configs/neon.h defines these
choices.
In particular, the distributed copy currently expects a Windows BMP
file named bdlogo.bmp to be present on the MMC card and writes it to the
display, then loads an operating system image from a file named nk.nb0 to
RAM address 0xa0030000 and executes it.
Both of these are defined by the lines which resemble this:
#define CONFIG_BOOTCOMMAND
"mmcinit; " \
"fatload mmc 0 a0000000 init.scr ; " \
"autoscr a0000000 ; "
loader with support for USB and network boot, including BOOTP/DHCP,
details.
December 28, 2005
Revision 2.8
5.1 Das U-Boot
12
5.1.5 U-Boot Memory layout
32K segment used for page tables.
Unused RAM
0xA4000000
Page Tables
0xA3FF8000
0xA3FF7FFF
Unused High
Tail of 32MB
Das U-Boot image
Heap and Stack
Frame Buffer
Unused Low
0xA2000000
0xA1FFFFFF
boundary
the 32MB boundary
0xA1F00000+
0xA1F00000
0xA1EFFFFF
The heap and stack are allocated in space
1
preceding the U-Boot image.
0xA1EFFFFF-
0xA1EFFFFF-
Frame Buffer for BD-2003
Unused Low RAM
0xA1EFFFFF--
0xA1EFFFFF--
0xA0000000
December 28, 2005
Revision 2.8
5.1 Das U-Boot
13
5.1.6 U-Boot Init Script
the Hush parser and the autoscript command. You should notice when first
compiling the package that the Boundary Devices sample uses this to defer
most board initialization to the MMC card. It does this by setting the
CONFIG BOOTCOMMAND environment variable as follows.
#define CONFIG_BOOTCOMMAND
"mmcinit; fatload mmc
0
a0000000 init.scr
;
autoscr a0000000
"
In English, this instructs U-Boot to initialize the MMC/SD card driver,
load a file named init.scr from the card to address A0000000 (the start of
RAM), and execute the script from that memory address. This little bit of
scripting effectively passes all responsibility of what to do at boot time to
the MMC card.
The sample script is defined in u-boot-1.1.2/board/neon/init.script and
performs the following steps.
1. Loads and displays a logo. The script looks for an image file named
logo.bmp on the MMC/SD card. If found, it displays the logo on the
LCD panel. We recommend that you place a splash image of a size
matching your display on the MMC card. Note that the bitmap must
be an 8-bit color bitmap.
2. Loads and runs Windows CE. Next, the script attempts to load
NK.nb0 from the MMC/SD card and run it.
As mentioned earlier, the initialization has been mostly deferred to the
MMC/SD card, so the compiled script (init.scr) must be placed on the
during the U-Boot build process.
The following list is a recap the expected content of the MMC/SD card
when using the Boundary Devices initialization script.
Filename Description
init.scr
logo.bmp 8-bit color splash image
NK.nb0 Windows CE image
Compiled initialization script
December 28, 2005
Revision 2.8
5.2 Windows CE
14
5.2 Windows CE
R
As mentioned earlier, the Neonꢀ board ships with a runnable Windows CE
necessary to tailor the operating system for a given application.
The following sections describe the process of producing an image match-
R
ing the one shipped with the Neonꢀ board.
5.2.1 Prerequisites and components
R
Most of the tools needed to create a bootable Windows CE 5ꢀ application
R
for the Neonꢀ board are provided by Microsoft. The following is a complete
list of components and where they may be obtained.
R
Windows CE 5ꢀ
Embedded Visual C++ 4.0
R
Neonꢀ Board Support Package
5.2.2 BSP Installation
The Neon BSP is made available as a Windows installer file on the Boundary Devices
website. This file defines a single BSP for the BD2003 and SM501-supporting
variants. Installation consists of running the .msi file.
c:\> .\bsp20050413.msi
of the Windows CE BSP.
As a reference tool for the content of the BSP, you should consider using
MSI2XML to view the content.
December 28, 2005
Revision 2.8
5.2 Windows CE
15
5.2.3 Building the demo
After installation of the BSP, this project may be copied to a new direc-
tory within the WINCE500 PBWorkspaces directory and built using Plat-
form Builder.
C:\WINCE500\PBWorkspaces>md bdWeb
C:\WINCE500\PBWorkspaces>cd bdWeb
=> ‘bdWeb.pbxml’
Resolving boundarydevices.com... 66.113.228.134
Connecting to boundarydevices.com[66.113.228.134]:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 45,478 [text/plain]
100%[============================================================================>] 45,478
17:37:40 (58.90 KB/s) ‘bdWeb.pbxml’ saved [45478/45478]
58.90K/s
-
C:\WINCE500\PBWorkspaces\bdWeb>.\bdWeb.pbxml
C:\WINCE500\PBWorkspaces\bdWeb>
After this is done, you should be able to build the sample WinCE
platform through the Build OS|Sysgen and Build OS|Build and Sysgen
Current BSP menu options.
December 28, 2005
Revision 2.8
5.3 Linux Support
16
5.3 Linux Support
The Linux Environment for Boundary Devices boards consists of four pri-
mary pieces, a toolchain, the kernel and device drivers, a user-space build
capabilities of the hardware.
5.3.1 Crosstool Linux Toolchain
Before the kernel and applications can be built, it is first necessary to have
a cross-compiler toolchain.
The following examples show how we at Boundary Devices set up our
First, you’ll need to download and unpack crosstool;
$ tar zxvf crosstool-0.37.tar.gz
a starting point with one of the demo build scripts. We’re currently using
demo-arm-xscale.sh with the following settings (GCC 3.4.3 with Glibc
version 2.3.5):
TARBALLS_DIR=/armArchives
RESULT_TOP=/opt/crosstool
eval ‘cat arm-xscale.dat gcc-3.4.3-glibc-2.3.5.dat‘ sh all.sh --notest
We also build the compiler to use software floating point in user space,
rather than hardware floating point (which traps to the kernel). To do this,
modify arm-xscale.dat and add the --with-soft-float and --without-fp
flags as shown below.
GCC_EXTRA_CONFIG="--with-cpu=xscale --enable-cxx-flags=-mcpu=xscale --with-float=soft"
GLIBC_EXTRA_CONFIG="--without-fp"
Also, we typically change the TARGET to read as follows:
TARGET=arm-linux
because arm-linux-gcc is just too long!
Having completed these edits, you can execute the script as follows:
sh demo-arm-xscale.sh
December 28, 2005
Revision 2.8
5.3 Linux Support
17
you wait.
When complete, you should find a whole slew of programs in your
/opt/crosstool/gcc-3.4.3-glibc-2.3.5/arm-xscale-linux-gnu/bin/ di-
rectory:
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
-rwxr-xr-x
1
2
2
2
1
1
2
2
2
1
1
1
2
2
1
1
2
1
1
1
2
1
username cvsd 1900724 Jul 18 20:48 arm-linux-addr2line
username cvsd 1960214 Jul 18 20:48 arm-linux-ar
username cvsd 3339533 Jul 18 20:48 arm-linux-as
username cvsd 331791 Jul 18 21:35 arm-linux-c++
username cvsd 1855723 Jul 18 20:48 arm-linux-c++filt
username cvsd 331290 Jul 18 21:35 arm-linux-cpp
username cvsd 331791 Jul 18 21:35 arm-linux-g++
username cvsd 330887 Jul 18 21:35 arm-linux-gcc
username cvsd 330887 Jul 18 21:35 arm-linux-gcc-3.4.3
username cvsd
16265 Jul 18 21:35 arm-linux-gccbug
username cvsd 102084 Jul 18 21:35 arm-linux-gcov
username cvsd 2373278 Jul 18 20:48 arm-linux-gprof
username cvsd 2622683 Jul 18 20:48 arm-linux-ld
username cvsd 1937609 Jul 18 20:48 arm-linux-nm
username cvsd 2454999 Jul 18 20:48 arm-linux-objcopy
username cvsd 2595563 Jul 18 20:48 arm-linux-objdump
username cvsd 1960209 Jul 18 20:48 arm-linux-ranlib
username cvsd 429743 Jul 18 20:48 arm-linux-readelf
username cvsd 1806673 Jul 18 20:48 arm-linux-size
username cvsd 1780595 Jul 18 20:48 arm-linux-strings
username cvsd 2454994 Jul 18 20:48 arm-linux-strip
username cvsd
14395 Jul 18 21:47 fix-embedded-paths
The instructions above can be followed to create a toolchain suitable for
cross-compiling Arm-Linux programs on a host machine. The needs for
building the boot loader are a bit different, though. In particular, the ’glibc’
reference above refers very specifically to userspace ”C” and ”C++” libraries
that defer much of their I/O to the Linux kernel itself through the use of
system calls.
we need to build a Cross-compiler with a different set of switches. Thank-
The next couple of steps will do just that.
First of all, create a file named
crosstool-0.37/contrib/newlib/arm-elf-newlib-1.12.0.dat
and paste the following content into it.
TARGET=arm-elf
TARGET_CFLAGS="-O2"
BINUTILS_DIR=binutils-2.14
NEWLIB_DIR=newlib-1.12.0
21 hr, 15 minutes on a 1GHz Athlon w/512MB of RAM
December 28, 2005
Revision 2.8
5.3 Linux Support
18
GCC_DIR=gcc-3.4.3
GCC_EXTRA_CONFIG=
Then, create a shell script named
following content.
crosstool-0.37/contrib/newlib/arm-elf.sh
with the
#!/bin/sh
set -ex
TARBALLS_DIR=/armArchives
RESULT_TOP=/opt/crosstool
export TARBALLS_DIR RESULT_TOP
GCC_LANGUAGES="c,c++"
export GCC_LANGUAGES
# You should do the mkdir before running this,
# and chown /opt/crosstool to yourself so you
# don’t need to run as root.
mkdir -p $RESULT_TOP
# Build the toolchain.
# Takes a couple hours and a couple gigabytes.
eval ‘cat arm-elf-newlib-1.12.0.dat‘ sh all-newlib.sh --notest
echo Done.
Next, edit the contrib/newlib/getandpatch-newlib.sh file and re-
place the line that says:
with the following
Then, run the script like so.
$ time sh arm-elf.sh
them.
December 28, 2005
Revision 2.8
5.3 Linux Support
19
5.3.4 Kernel 2.4.19
Intel PXA support for ARM-Linux
Boundary Devices support
5.3.5 Kernel 2.6
bzcat linux-2.6.11.11.tar.bz2 tar xvf
|
-
cd linux-2.6.11.11
bzcat ../boundary-2.6.11.11-2005-11-25.patch.bz2
cp arch/arm/configs/neon_config ./.config
|
patch -p1
yes ""
make ARCH=arm CROSS_COMPILE=arm-linux- uImage
|
make ARCH=arm CROSS_COMPILE=arm-linux- oldconfig
Notes:
Five Wire touch screen support requires setting
Sound|OSS|Multimedia Capabilities Port drivers|UCB 1400|Five wire
(or edit .config and set CONFIG_UCB1400_TS_FIVE_WIRE=y)
December 28, 2005
Revision 2.8
5.3 Linux Support
20
5.3.6 Userland build tool
As mentioned before, we at Boundary Devices use a variant of an older
for various libraries. This allows inter-library dependencies to be expressed,
and also allows the canonical source locations to be used during a build.
This should really be better documented, but the short and simple build
instructions are as follows.
$ tar zxvf userland_20051126.tar.gz
$ cd userland
$ make menuconfig
-- at a minimum, you’ll need to set an archive path to
a writable directory, and validate your kernel and toolchain
paths.
$ make cramfs
Note that this takes a while (over an hour on a typical machine), but
will result in a cramfs image being created in the userland/ directory.
Also note that installation of the [[tinylogin]] program requires privileges
to [[setuid root]]. Because of this, the makefile rules/tinylogin.mak uses the
[[sudo]] program. If you don’t have sudo installed, this process will fail.
If you do, you may see a password prompt very near the end of the build
process (while installing tinylogin into the root filesystem). To avoid this,
you can either set your [[sudo]] timeout to something large and perform a
sudo operation before kicking off the build, or do as I do and set it negative
The choice of cramfs is for illustration (and because it requires that
tails about the choices available and decisions you need to make regarding
deployment.
More specifically, the userland build tool is designed to allow repro-
ducible builds of entire userland filesystems and device nodes for embedded
Linux distributions.
The general flow of the make is as follows:
1. Configure the system through the kconf tool. This step produces a
file named .config in the userland directory.
You should save this file for future reference when you have a set of
choices that meet your needs. By saving it off to say good.config,
you can copy it back to .config and reproduce the build later.
2. Get the source code for each component. Since downloading all of
December 28, 2005
Revision 2.8
5.3 Linux Support
21
the components may take a while, it is often useful to perform this
step by itself after configuration.
The get makefile target can be used to perform this step.
Note that the original web locations are generally used for each library
supported by the userland build. This is generally a good thing, but
also means that things sometimes move.
We try to keep a set of archives on the Boundary Devices website for
use when the original sources are unavailable.
3. Build libraries under build/ the system through the kconf tool.
As mentioned earlier, the build tool allows you to express inter-library
dependencies in their makefile packets.
The packets for each component are stored in userland/rules and
consist of both a configuration piece *.in and build instructions *.make.
The install target can be used to simply build the components with-
out making a root filesystem.
4. Install libraries into install/. This mingling of various libraries is
done to allow simplified include file and library references for depen-
dent packets.
5. Build a root filesystem under root/. This step gathers all of
the executable portions (applications and shared libraries) for each
component into a root filesystem image. Scripts are also commonly
installed, as are any supporting configuration files (under root/etc).
The rootfs target can be used to create the root filesystem without
creating a flattened image.
6. Build a device table. This step uses the kernel configuration file to
create devices.txt, suitable for use with genext2fs, mkcramfs, or
mkfs.jffs2.
The devices target can be used to create the device table without
performing any other build steps.
7. Flatten the root filesystem into any of cramfs, initrd, or JFFS2
images for placement in flash or SD card.
December 28, 2005
Revision 2.8
5.3 Linux Support
22
5.3.7 Userland libraries and applications
The following libraries and applications are included in the userland build.
Name
Description
Link
bdScript
busybox
cramfs tools
libcurl
e2fsprogs
flash
freetype
jpeg
Javascript
ID3 Tag Library MP3 ID tag library
MP3 Library
libpng
libungif
libmpeg2
openssh
openssl
udhcp
Shell and utilities
Cramfs Utilities
HTTP library and more
Ext2 Filesystem Utilities
GPL’d Flash Library
FreeType Text Rendering
JPEG image library
Javascript library
MPEG Audio Decoder
PNG image library
GIF decompression
MPEG decoder library
SSH Application
SSL Library
DHCP Client/Server
zlib compression library
zlib
December 28, 2005
Revision 2.8
5.3 Linux Support
23
5.3.8 Notes about userland root filesystems
The cramfs option is one of three primary ’bundled’ targets:
1. cramfs - Creates a single file as a read-only, gzip-compressed image of
a filesystem tree. When you can nail down the content of your filesys-
tem, this is a great choice, providing the fastest boot time (around 7
seconds on a PXA-255) and complete immunity to corruption. This
filesystem is often used in conjunction with read-write filesystems (ram
disk for volatile data, or VFAT for semi-static data).
Requires cramfs support in the kernel (Miscellaneous Filesystems—Compressed
ROM file system support).
2. jffs2 - Creates a single file as a read-write, gzip-compressed image of
a filesystem tree. This is useful for placement in flash, and is fairly
immune to corruption at the cost of extra time for validation at boot
(typically 30-45 seconds for a 32MB filesystem).
Requires JFFS2 support in the kernel (Miscellaneous Filesystems—Journalling
Flash File System v2).
3. mmcinitrd/mmcinitrd.u-boot - Creates a single file as a read-
write, uncompressed image of a filesystem tree suitable for use as an
initial RAM disk (initrd).
It requires the following options in the kernel:
Loopback device support
Device Drivers—Block Devices
Initial RAM Disk support Device Drivers—Block Devices
In addition, this target makes a bunch of other choices for you. Since
this is a bit involved, discussion of the steps is deferred to Section
The Makefile instructions for each of these is at the tail-end of the
userland Makefile (userland/Makefile).
Refer to that file for details, but the bundled image for each is created
by performing a single command specifying an output file (the image), a
path name to a directory tree, and the devices.txt file.
Typical usage for the initrd target is to have the boot loader load the
to the Linux kernel through the bootm command.
Both the cramfs and JFFS2 images may also be mounted directly from
flash EEPROM using Linux MTD block devices. U-Boot’s support for passing
December 28, 2005
Revision 2.8
5.3 Linux Support
24
Linux boot command line parameters to the kernel also helps here. Typical
usage includes is of the following form, which supplies both the MTD partition
information and the root filesystem reference:
mtdparts=phys_mapped_flash:1024k(armboot),256k(params),-(rootfs1) root=/dev/mtdblock3 rootfstype=cramfs
In English, this reads as something like:
MTD partitions are 1MB (named armboot), 256K(named params),
with the remainder of flash named rootfs1. The root filesystem
is in the third partition, and its’ type is cramfs.
Mounting a JFFS2 image is done in the same manner, except the rootfstype
parameter has a value of jffs2.
The U-Boot boot loader supports copying data from RAM to flash for
upgrades and such. Refer to the unprotect, erase, and cp commands for
details.
A third means of mounting one of these root filesystems is to use a loop
device. In Linux jargon, a loop device is a file that contains a filesystem
within it. Both the initrd and cramfs images may be used in this fashion as
shown in the following examples.
Mount a cramfs file (by far the simplest case).
~ $ sudo mount -o loop -t cramfs ~/cramfs.img /mnt/cramfs
Mount an ext2 image (Only slightly harder because mmcinitrd is actually
gzipped and needs to be gunzip’d first).
~ $ cp -f mmcinitrd mmc.img.gz
~ $ gunzip mmc.img.gz
~ $ sudo mount -o loop -t ext2 ~/mmc.img /mnt/ext2
To mount a JFFS2 image a bit more is needed. Your kernel needs to
have mtd and mtdblock support compiled in or installed as modules. Then,
a mtdram device can be created, you can copy the JFFS2 data to it and
mount the device.
~
~
$
$
sudo /sbin/insmod mtdram total_size=32768 erase_size=256
Using /lib/modules/2.4.23_pre8-gss-r2/kernel/drivers/mtd/devices/mtdram.o
sudo dd if=jffs2.img of=/dev/mtd0
10809+1 records in
10809+1 records out
sudo /sbin/insmod mtdblock
~
$
Using /lib/modules/2.4.23_pre8-gss-r2/kernel/drivers/mtd/mtdblock.o
sudo mount -r -t jffs2 /dev/mtdblock/0 /mnt/jffs2/
ls /mnt/jffs2/
~
~
$
$
bin etc lib linuxrc opt proc sbin sysfs tmp usr var
December 28, 2005
Revision 2.8
5.3 Linux Support
25
5.3.9 mmcinitrd.u-boot
The mmcinitrd.u-boot userland Makefile target has a lot of parts, but its’
goal is simple
Provide an application developer a means of staying focused on
development without the possibility of trashing a flash.
It presumes the existence of an SD card formatted with the VFAT filesys-
tem, and a cramfs image on the SD card (in the root as cramfs.img). The
mmcinitrd.u-boot file is also typically loaded on the SD card, but that isn’t
strictly necessary, as long as it is available and handed to the Linux kernel.
Through a series of steps, it links the /bin, /lib, /usr, /var, /sbin,
and /share directories from within cramfs.img, leaving the root of the
filesystem read-write (and volatile), with /mmc referring to the root of the
SD card.
Furthermore, it presumes the existence of a script or executable named
linux init in the root directory of the SD card.
This is done both as an example and as a useful way of nailing down the
static pieces of a package (in the cramfs.img file) and allowing read-write
access to the filesystem during application development.
The linux init script on the SD card may be modified to start an app
directly, without any risk of boot failure.
Look at the file /etc/init.d/rcS for the details of how this is accom-
plished.
5.3.10 Javascript stuff
ary Devices scripting application.
5.3.11 Login and SSHD support
By default, the Userland build tool creates a password file /etc/passwd
with a root password of BoundaryDevices.
This is only needed when connecting over sshd.
Use the menuconfig make target to change this.
December 28, 2005
Revision 2.8
26
6 Development Tools
6.1 minidebug
minidebug is a small (under 16k) debugger designed to fit completely within
the instruction cache on the PXA-255 processor to allow testing of boards
even in the absence of ROM or RAM.
It also includes features to download over either serial or Ethernet, al-
lows the display and manupulation of registers and memory, and supports
controlled execution through breakpoints and data watchpoints.
Upon entry, minidebug generally displays a dot (.) prompt, sometimes
pre-pended by a string that looks like $S00#b3. Fear not. The $S00#b3
string is used to allow minidebug to work in conjunction with the gdb de-
bugger on the attached system.
The following is a list of commands that can be issued at the dot prompt.
Note that this list can also be retrieved through minidebug by entering a
question mark (?).
command params
description
BC
BE
BS
address
address
address
Breakpoint clear
Breakpoint examine
Breakpoint Set
BURN
E
D
address range Burn image at address range to flash
address Examine and modify memory
address value Deposit
DL
DLW
G
address
address
address
Start XModem for serial download
Download wireless
Go
GL
GG
R
SSID
T
Go Linux
address
Go no cache clear
Display content of registers
Set Wireless SSID string
Trace
TT
V
Trace no cache clear
address range Verify content of flash
WC
WR
WRW
WW
?
address
address
address
address
Watch clear
Watch read
Watch read/write
Watch write
Show this list of commands
December 28, 2005
Revision 2.8
6.2 JTAG system-level debugger
27
6.1.1 mdebug
The mdebug image adds Ethernet and wireless download capabilities using
R
the Blast protocol to the Neonꢀ . The SSID and DLW commands above are
only valid when mdebug is present.
The following is an example of the use of mdebug and DLW. Note that
the first commands used download mdebug to address A1C00000 and run it
from there. Also note that the use of DLW requires a DHCP or BOOTP
server for IP address assignment.
DLW example
. dl a1c00000
CCCCCCCCCCC
enter binary file name: mdebug
CCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCC.....................
...................................................
73620 bytes, 72 packets,
OK A1C00000-A1C12000
0
retrys
.
$S00#b3
g
a1c00000
Reset A0008000
R0: 00000000 R1: 0000014C R2: 00000001 R3: 00000060
R4: A1F1D540 R5: A1F22B1C R6: A1E9BECC R7: 00000002
R8: A1E9BFDC R9: A1E9BE88 SL: 00000000 FP: A1E9BE10
IP: A1E9BE14 SP: A0003400 LR: A0008000 PC: A0008000
CPSR 600000D3 FP0: 0000000000
.
.
dlw a0008000
Boundary Devices
SMC91C11xFD
1
%s: PHY=LAN83C183 (LAN91C111 Internal)
%s: PHY remote fault detected
%s: Ethernet Link Detected
%s: PHY 100BaseT
%s: PHY Half Duplex
valid mac address
00:50:C2:06:30:8F
..........DISC:received 0x012C bytes of reply
done
REQ:received 0x012C bytes of reply
done
router at 192.168.0.1
DNS server at 68.2.16.25
DNS server at 68.2.16.30
DNS server at 68.6.16.30
DHCP success, using IP 192.168.0.14
ready to receive file
enter binary file name: cramfs.img
...........................................................
................
transmitted in 52 seconds
.[eof]
lost 0x00000000 packets
[eof] in 52 seconds
sent 19783680 bytes of file to 192.168.0.14
Error free!!!
0x012DE000 bytes written to buffer at A0008000 A12E6000
6.2 JTAG system-level debugger
The jtag executable provided by Boundary Devices is based on the one
Our main goals in developing the jtag program were to aid in hardware
debugging and to allow the first flash EEPROM image to be burned onto
December 28, 2005
Revision 2.8
6.2 JTAG system-level debugger
28
a new device. That said, we also use it extensively as a terminal emulator
during development and have added a number of extensions for that purpose.
The current release supports the PXA250, PXA255, PXA270, and SA1100
(lart untested). It checks the IDCODE register and uses the appropriate
BSDL structure.
6.2.1 Requirements
The jtag executable runs either under Linux or Cygwin.
Under Linux, there are no known dependencies except for libc and lib-
stdc++.
stalled. This driver makes the ioperm() and iopl() system calls available
under Windows for access to the serial and parallel ports. Note that after
the cygwin package is installed, you still need to enable the driver through
the use of the ioperm executable
For the cmd.exe inclined:
c:\> c:\cygwin\bin\ioperm.exe -v -i
or for the bash-inclined:
user@machine ~/u-boot-1.1.2
$ /bin/ioperm.exe -iv
Either way, the output should be something like the following.
Installing ioperm.sys...
OpenSCManager
CreateService
OpenService
ok
ok
ok
ok
StartService
ioperm.sys is already running.
6.2.2 Startup Options
jtag -t Generate a square wave on the processor pins.
This option allows pins to be checked in a sequence defined by the
hardware file. A ’+’ or ’-’ keypress will scroll forward or backward
through the list. Also, pin name can be entered directly. Entering
GP0 will generate a square wave on GP0. A ’?’ will list matching pin
names. Entering GP? will list all gpio pins.
December 28, 2005
Revision 2.8
6.2 JTAG system-level debugger
jtag -i Identify the flash part used
29
This option tries to identify the part number of the Flash EEPROM.
Currently supported parts are 28F160F3B, 28F320J3A, 28F128J3A,
28F320C3B, and 28F320S3, though not all have been tested. It should
be relatively easy to add new parts.
jtag -f Generate the appropriate signals to program a flash.
This option is rarely used, since we normally program the flash through
the minidebug software.
jtag -c Download code to the mini and main instruction cache.
This option is used to load a file into the instruction cache. Usually
-x, -e, -or -d option is used to load minidebug. The -d option just
loads minidebug. The -x option then proceed to dowload a file over
the serial port using xmodem. The -e option dowloads a file using
ethernet (wireless and wired support.) The -ssid option can be used
to specify a wireless essid value to pass to minidebug.
jtag -s Terminal emulator option.
The parallel port is still searched because [Ctrl A] B can be used to
send a JTAG break and attempt to return control to minidebug.
jtag -N Burn the entire flash.
This option can be used to burn a flash for the first time.
It first downloads the file mdebug to ram address A1800000.
Then it executes an ethernet download of the file totalflash.
If successful, it then burns the flash using the minidebug(mdebug)
command BALL (burn all).
December 28, 2005
Revision 2.8
6.2 JTAG system-level debugger
30
6.2.3 Control Keys
Once running, the jtag program responds to a number of command se-
quences, all beginning with [Ctrl A] .
[Ctrl A] B Send a break
[Ctrl A] S Send a file using XModem
[Ctrl A] L Toggle logging to jtag.log
[Ctrl A] T Send an ascii file
[Ctrl A] P Choose baud rate
[Ctrl A] Q Quit
[Ctrl A] R Hardware reset
6.2.4 Blast protocol
When used with the mdebug image, the jtag program recognizes the start-
of-download request sent by the device, and will prompt the user for a file
name to send. Refer to the example in the mdebug section for details.
6.2.5 Quick-start download and burn
If you have a minidebug for your platform in the current working directory,
the following sequence shows the process of using it to download and burn
a new u-boot image.
Start debugger.
$
$
cd ..
./jtag -d
ioport 3bc wrote 5d read ff
using printer port at 378
IDCODE: 69264013
Halt released
-
0110 1001001001100100 00000001001
1
Waiting for stub
LDIC finished
This uses the program minidebug on the arm to download to ram
using the serial port(xmodem protocol) or blast the file using
ethernet
^A
^A
^A
Q
I
T
for quit, ^A
for sending an RGB bitmap with xmodem, ^A
to send an ascii file
B
external break, ^A
S
for sending
P
a
baudrate
file with xmodem,
DBG-Vector Trap A0008000
R0: 00000000 R1: 0000014C R2: 00000000 R3: 00000003
R4: 0000001E R5: 81A0F288 R6: AAA00010 R7: 000BD784
R8: 00000000 R9: 81A18774 SL: AAA0001C FP: 81A1606C
IP: 80039094 SP: A0003400 LR: 8006C8CC PC: A0008000
CPSR 600000D3 FP0: 0000000000
.
To download using serial, use the ’dl address’ command.
Hit [Ctrl A] S to send the file (assumes u-boot.bin in the current direc-
tory). After issuing the DL command, the minidebug will begin sending
C’s. These are the start commands for XModem, and signal the readiness
December 28, 2005
Revision 2.8
6.2 JTAG system-level debugger
31
to receive a file. Use the [Ctrl A] S sequence to instruct jtag to prompt
for and send a file using XModem.
To abort the operation, either when prompting for a filename or before,
use [ctrl-C].
. dl a1f00000
CCCCCCCCCCCCCC
enter binary file name: u-boot.bin
CCCCCCCCCCCCCCCC............................................
....................................
81292 bytes, 80 packets, 0 retrys
OK A1F00000-A1F14000
To burn a range of data from RAM to the start of flash, use the ’burn’
command like this. Note that the end address was given above at the end
of the DL response.
. burn a1f00000 a1f14000
Sector 04000000 Erasing Programming Verifying...
Success
December 28, 2005
Revision 2.8
6.3 TeraTerm blast extensions
32
6.3 TeraTerm blast extensions
As an alternative to the jtag executable, Boundary Devices has also pro-
duced an extension to the TeraTerm open-source terminal emulator with
R
support for the Blastꢀ protocol.
It has the following benefits over the use of jtag:
• Does not require Cygwin and ioperm
R
• Because it’s a Windowsꢀ graphical application, it’s a bit simpler to
use and has a file-chooser dialog.
The drawback is that it does not support the jtag hardware connection
or any of the associated features (can’t force a hardware reset, can’t recover
a machine with a trashed flash).
We recommend its use only for non-development needs, or when cabling
the jtag is inconvenient (e.g. during production).
December 28, 2005
Revision 2.8
6.4 Using U-Boot Networking
33
6.4 Using U-Boot Networking
transfer files across a network. As shown below, the dhcp command is
typically used to perform both a BOOTP/DHCP request and transfer a file.
$
$
$
$
set bootfile nk4.nb0
set serverip 192.168.0.26
dhcp
Using MAC Address 00:50:C2:06:30:8F
BOOTP broadcast
1
DHCP client bound to address 192.168.0.14
TFTP from server 192.168.0.26; our IP address is 192.168.0.14
Filename ’nk4.nb0’.
Load address: 0xa0030000
Loading:
T
T
#################################################################
#################################################################
#################################################################
#################################################################
#################################################################
#################################################################
...
#################################################################
#################################################################
done
Bytes transferred
$
=
23068672 (1600000 hex)
First of all, the bootfile environment variable is used in the example
above to define the file to transfer. By default, the boot file is computed
using a hex representation of the IP address assigned to the device.
’192.168.0.14’
=> ’0E00A8C0.img’
Used with a tftp server that allows symlinks, this provides a convenient
way to define per-device boot files.
The second thing to note in the example is the use of the serverip
environment variable. This variable defines the IP address of the TFTP
server, in this case ’192.168.0.26’. If your DHCP server allows setting of
the si addr field in the DHCP response (refer to RFC2131 for details), this
value can be automatically provided.
The third thing of interest is the load address (0xa0030000). This value
is defined in neon.h in the CFG LOAD ADDR macro. It may be overridden
through the use of the loadaddr environment variable.
The CONFIG EXTRA ENV SETTINGS macro in configs/neon.h may be
used to assign the proper compile-time defaults for the environment variables
listed above.
The DHCP/BOOTP/TFTP process is relatively fast, even using a slow
protocol like TFTP. The 23MB transfer above took 20 seconds. Much faster
than swapping MMC cards. Slower than mdebug/jtag under Linux, but
faster than Cygwin jtag and blast.
Any server software that supports RFC1350 should work. The stan-
dard tftpd daemon under Linux is a good choice. Under Windows, the free
Tftfpd32 by Philippe Jounin is a very nice tool.
December 28, 2005
Revision 2.8
34
7 Configuration Notes
7.1 Display configuration
R
The Neonꢀ supports a variety of LCD panels. The following section de-
scribes the process of configuring the board for a known, currently supported
on a new panel.
If you know the type of panel at compile time, you can place a selection
The CONFIG EXTRA ENV SETTINGS macro is used to define a compile-time
choice. If you are using EEPROM to store environment settings, these can
be saved in the environment as well as described below.
Name
Resolution Description
qvga portrait
hitachi qvga
sharp qvga
hitachi hvga
sharp vga
240 x 320
320 x 240
320 x 240
640 x 240
640 x 480
800 x 480
Hitachi Quarter VGA 3.5” panel
Hitachi High-Brightness Quarter VGA
Sharp Quarter VGA
Hitachi Half VGA
Sharp 10.4 inch VGA
Hitachi Half VGA
hitachi wvga
crt1024x768
1024 x 768 HP SVGA
For example:
#define CONFIG_EXTRA_ENV_SETTINGS "panel=hitachi_hvga" "\0"
Note that this is automatically done as a part of the make neon config
step.
The boot loader settings for the LCD panel will carry through to the
Linux and Windows CE drivers.
R
If you’re using the Neonꢀ with a new panel, you’ll need to determine
and define the following fields for the panel.
December 28, 2005
Revision 2.8
7.1 Display configuration
35
field name
name
type
string
description
used to identify the panel
pixclock
number Divisor for the pixel clock. Generally
3 for QVGA, 1 for higher resolution.
number Horizontal pixel count
xres
yres
number Vertical pixel count
act high
hsync len
left margin
right margin
vsync len
number Clock polarity, 0 (default) or 1
number Horizontal sync pulse
number Idle pixels before leftmost pixel
number Idle pixels after rightmost pixel
number Vertical sync pulse
upper margin number Idle rows before topmost
lower margin number Idle rows after bottom
active
crt
rotation
number Active Matrix (1) or Passive (0)
number digital LCD(0) or Analog CRT(1)
number landscape(0) or portrait(90)
Once you have collected this information, a corresponding entry must be
added to the list of panels.
u-boot-1.1.2/common/lcd_panels.c
To allow the testing of these settings and the use of a different display
without re-compiling, the lcdp boot loader command is available. It may
be used in one of the following ways:
command string description
lcdp
Show the current lcd panel settings
lcdp ?
Show the list of currently supported lcd panels
lcdp panelname Select and initialize panelname
lcdp + Add a new panel (prompts for details)
Note that the boot loader text display will not be updated properly if
the X and Y resolution don’t match the current default display. Use the
bmp commands to test the new panel configuration after using the lcdp +
command string.
As always, the source code is available. The two modules used to support
dynamic display selection are:
• common/cmd lcdpanels.c - defines U-Boot commands
• common/lcd panels.c - display initialization
7.1.1 What display is currently selected?
The lcdp command is used for a variety of purposes including querying the
currently selected display.
December 28, 2005
Revision 2.8
7.1 Display configuration
36
$ lcdp
------------------------------------
name
pixclock
xres
: crt1024x768
: 65000000
: 1024
: 768
: 1
: 200
: 24
: 161
: 6
yres
act_high
hsync_len
left_margin
right_margin
vsync_len
upper_margin
lower_margin
active
: 3
: 29
: 0
7.1.2 What displays are supported...?
The lcdp command followed by a question mark will list the currently sup-
ported displays. As shown in the following example, the list is extensive
(and extensible, as we’ll show later).
$ lcdp ?
------------------------------------
name
: hitachi_qvga
pixclock
xres
yres
: 0
: 320
: 240
: 1
: 64
: 1
: 16
: 20
: 8
act_high
hsync_len
left_margin
right_margin
vsync_len
upper_margin
lower_margin
active
: 3
: 1
------------------------------------
name
pixclock
xres
: sharp_qvga
: 0
: 320
: 240
: 1
: 8
: 16
: 1
yres
act_high
hsync_len
left_margin
right_margin
December 28, 2005
Revision 2.8
7.1 Display configuration
37
vsync_len
upper_margin
lower_margin
active
: 20
: 17
: 3
: 1
------------------------------------
name
: hitachi_hvga
pixclock
xres
yres
: 1
: 640
: 240
: 1
: 64
: 34
: 1
: 20
: 8
: 3
act_high
hsync_len
left_margin
right_margin
vsync_len
upper_margin
lower_margin
active
: 1
------------------------------------
name
pixclock
xres
: sharp_vga
: 1
: 640
: 480
: 1
: 64
: 60
: 60
: 20
yres
act_high
hsync_len
left_margin
right_margin
vsync_len
upper_margin
lower_margin
active
: 34
: 3
: 1
------------------------------------
name
: hitachi_wvga
pixclock
xres
yres
: 1
: 800
: 480
: 0
: 64
: 1
: 39
: 20
: 8
act_high
hsync_len
left_margin
right_margin
vsync_len
upper_margin
lower_margin
active
: 3
: 1
------------------------------------
December 28, 2005 Revision 2.8
7.1 Display configuration
38
name
: crt1024x768
pixclock
xres
yres
: 65000000
: 1024
: 768
: 1
: 200
: 24
: 161
: 6
: 3
act_high
hsync_len
left_margin
right_margin
vsync_len
upper_margin
lower_margin
active
: 29
: 0
$
7.1.3 Select a supported display
If you supply a supported panel name on the lcdp command line, the display
controller will be reset with the associated parameters.
$ lcdp hitachi_wvga
found panel hitachi_wvga
panel: 800x480x8
$ lcdp
------------------------------------
name
: hitachi_wvga
pixclock
xres
yres
: 1
: 800
: 480
: 1
: 64
: 1
: 39
: 20
: 8
act_high
hsync_len
left_margin
right_margin
vsync_len
upper_margin
lower_margin
active
: 3
: 1
The selection takes place immediately, so if you have a panel connected,
you should see valid output on the display.
Note that if you change resolutions, the display memory will likely have
mis-aligned data in it. Displaying a bitmap on the display through the use
of the fatload and bmp commands will remedy this situation. Refer to
init.script for an example.
If you want to make your selection stick through a reset, you can save it
through the set and save U-Boot commands.
December 28, 2005
Revision 2.8
7.1 Display configuration
39
$ set panel hitachi_wvga
$ save
Saving Environment to Flash...
Un-Protected 1 sectors
Erasing Flash...
Erased 1 sectors
Writing to Flash... done
Protected 1 sectors
$ reset
resetting ...
$S00#b3
Reset A0008000
U-Boot 1.1.2 (Jun 10 2005 - 22:31:50)
U-Boot code: A1F00000 -> A1F20500 BSS: -> A1F54520
RAM Configuration:
Bank #0: a0000000 64 MB
Flash: 32 MB
panel hitachi_wvga found: 800 x 480
...
7.1.4 Define and test a new display
If you add a plus sign to the lcdp command line, you’ll be prompted for all
of the parameters needed to define a display.
$ lcdp +
name: myDisplay
pixclock: 65000000
xres: 800
yres: 600
act_high: 1
hsync_len: 200
left_margin: 24
right_margin: 161
vsync_len: 6
upper_margin: 4
lower_margin: 29
active (0|1) : 1
------------------------------------
name
: myDisplay
December 28, 2005
Revision 2.8
7.2 Memory size configuration
40
pixclock
xres
yres
: 1694498816
: 800
: 600
: 1
: 200
: 24
: 161
: 6
: 4
act_high
hsync_len
left_margin
right_margin
vsync_len
upper_margin
lower_margin
active
: 29
: 1
As with switching to a known panel, the settings take effect immediately
upon completion of the command. This can be a very quick way to add
support for a new display before committing it to the supported list.
Adding an entry into the lcd panels array in common/lcd panels.c
will provide boot-time support.
7.1.5 Saving settings to Flash EEPROM
All of the descriptions above are useful, but don’t address the issue of persis-
tence. That is performed through the use of the ’panel’ environment variable
The following example shows the process.
$ set panel crt1024x768
$ save
Saving Environment to Flash...
Un-Protected 1 sectors
Erasing Flash...
Erased 1 sectors
Writing to Flash... done
Protected 1 sectors
7.2 Memory size configuration
R
The Neonꢀ supports either 32 or 64MB of RAM.
Most of the default boot loader configuration assumes at least 32MB of
RAM is available. In particular, the TEXT BASE variable in board/neon/config.mk
links the uboot.bin image at 31MB from the start of RAM.
December 28, 2005
Revision 2.8
7.3 Upgrading U-Boot
41
Use the PHYS SDRAM 1 SIZE variable in include/configs/neon.h to
specify the actual size for your hardware.
The Windows CE image supports either, but defaults to 32MB. Set the
RAM SIZE 64 MB environment variable in your project to indicate that 64MB
should be present.
The RAM size set in the boot loader is passed to the Linux kernel.
7.3 Upgrading U-Boot
u-boot.bin) on an SD/MMC card, you can upgrade it by first unprotecting
and erasing the first sector of flash, then copying the new image to address
zero as shown below.
$ mmcinit
...
registering device
$ fatload mmc 0 a0008000 u-boot-neon.bin
reading u-boot-neon.bin
134264 bytes read in 271921 ticks, (73 ms),
adler == 0xf0cde398 in 24546 ticks, (6 ms)
$ protect off all
Un-Protect Flash Bank # 1
$ erase 0 3ffff
Erased 1 sectors
$ cp.b a0008000 0 $filesize
Copy to Flash... done
$ cmp.b a0008000 0 $filesize
Total of 134264 bytes were the same
$ reset
After reset, you should see the new build date in the U-Boot banner.
December 28, 2005
Revision 2.8
7.4 Touch Panel Calibration
42
7.4 Touch Panel Calibration
Under Linux, the flash sector at address 0x140000 is used to store the touch-
screen calibration settings. If you’re using bdScript startup code, the cali-
bration routine will launch upon first boot if not defined.
Under Windows CE, the touch screen settings are stored on the MMC
card in a file named touch.txt. You’ll need to use the mouse to launch the
touch calibration program.
December 28, 2005
Revision 2.8
7.5 Ethernet MAC Addresses
43
7.5 Ethernet MAC Addresses
Normally, Neon boards come with their MAC addresses pre-programmed
during assembly and test. This is done by using the U-Boot mac command
as shown below.
Invoked without an argument, the command will display the current
MAC address. Used with a single parameter (MAC address with colons
separating each pair of hex digits), the command will allow (re)programming
of the MAC address.
$ mac
mac address ff:ff:ff:ff:ff:ff
$ mac 00:50:c2:06:30:b8
setting mac address to 00:50:c2:06:30:b8
done
December 28, 2005
Revision 2.8
|
