Penguin
Note: You are viewing an old revision of this page. View the current version.

The Compaq Evo T20 is a thin-client device targetted at Windows TS or Citrix TS services.

Specifications:

  • 300 MHz Geode CPU
  • 64, 128 or 256 MB ram
  • USB only

Some of these units support PXE booting. If you happen to have one, it is trivial to turn it into a linux thin client - just make use of a PXE netboot environment, such as PXES or diskless.

If you don't have a PXE boot model, or you wish to run linux locally, you could try hacking the firmware (Note, this resource seems to have been relagated to the google cache)

General notes

Video depth

It appears that the driver XOrg uses for this unit can't support a bit depth of 24bits, but it does support 16bits.

Interestingly, if you allow the bootp to wait, it times out after an hour and tries to tftp /tftpboot/kernel on port 10069.


From Google Cache:

Linux on Compaq EVO T20 HOWTO

Dag Sverre Seljebotn Evo T20 Hacking Team

Copyright © 2004 by Dag Sverre Seljebotn Revision History Revision 0.3 11/08/2004 Revised by: DSS Revision 0.2 10/08/2004 Revised by: DSS Revision 0.1 09/08/2004 Revised by: DSS

Introduction

I did it!!

I'm really happy now. I've still got problems with my NFS setup so I haven't run X yet, but if anything difficult pops up there I'll update the howto. This should get you well started.

The version I have is 48/64, meaning 48 MB flash and 64 MB RAM. Changes for the other versions will probably be needed.

Note to HP/Compaq/WYSE: If you don't like what I have done here just email me and I'll promptly take the page down.

Uploading firmware

The first problem we're facing is that the BIOS won't boot from anything else than the soldered-on flash. Luckily there's a firmware update mechanism, "NETXFER". The EVO T20 tries to flash itself when you hold down the "P" key when applying power to it (a power outlet with a power switch is a must when working on this). "NETXFER" will appear on the screen, then it will try to connect with a BOOTP server.

You can download the windows tool for updating the firmware from hp.com (combined BOOTP+TFTP server, it will need an ancient Java implementation running on the machine, search the web), or (very very much preferred), set up Linux to update the firmware

*

Configure a BOOTP server to run at port 10067 and serve out an IP address and a filename (to the firmware file). Easiest is to use the stock ISC DHCP with something like:

range dynamic-bootp 192.168.0.100 192.168.0.140;

in the file "/etc/dhcpd.conf" (the point was the dynamic-bootp), then start it with:

dhcpd -p 10067 -cf custom-config-file-for-this-purpose

*

Run a TFTP server at port 10069, handing out the firmware file specified by the filename in the file "dhcpd.conf". Register the port in the file "/etc/services", then copy the line in inetd.conf for the TFTP and put it on that port as well.

It might look easier to use a Windows box, but getting it to work in Linux takes max half an hour and is very well worth it!

Then we need the stock firmware for modifications, get it from hp.com. It should be about 48 MB, and comes packed in an EXE-file that should execute just fine with the program WINE.

In the firmware, the data for the flash chips is located at offset 834710. That was the short story, if you don't have the 48 MB version it most likely won't be that so here's how I found out:

The firmware is a concatenation of setup code and a package. Inside the package, the filesystem is in two files called "filesys0" and "filesys1" (at the end of the package).

First do a flash of the stock firmware. Note the files that are transferred and their sizes... the filesystems are in "filesys0" and "filesys1". Search for those names in the firmware using an hex editor (i.e. ghex2). Somewhere before "filesys0" the size of "filesys0" is as a DWORD (four bytes) (you found the size when transferring, right?). Now, the four bytes before this is the offset from the start of the package.

To find the start of the package (except by just figuring it out in the hex editor), basically find the size of the setup code before it, which is the sizes of the files "TFTP.dat" and "ulc_code.tmp" that can be found in the windows Netxfer download. Other Evo versions might have other setup code, so make a new firmware using the setup code bundled with the program NETXFER
mkutc.bat firmware-file

It will "reformat" firmware-file with the setup code we know and stick it in the file "bootp.bin".

Now, hopefully you've found the offset of the file "filesys0", you should find the Windows NT master boot record there (with strings referring to your lack of proper hard disk setup, etc.). Now mount it as a loop device
losetup -o 834198 /dev/loop0 firmware.bin
Now we can make any changes we want on the device "/dev/loop0", treating it like a hard drive, and it will immedeately be reflected in the firmware. You can even mount 512 bytes further on the device "/dev/loop1" and mount the NTFS file system the Evo uses
losetup -o 834710 /dev/loop1 firmware.bin mount /dev/loop1 /mnt
I like to zero out everything but the MBR at this stage
dd if=/dev/zero of=/dev/loop bs=512 seek=1

just to remove every trace of Windows NTe. Then we must install our own stuff, I picked GRUB (because it was the one I first got to the assembler level of in my efforts to make the bootloaders work on the Evo...).

Making GRUB work

Getting GRUB running was quite a challenge for me, I had to do my share of debugging through the assembler code of GRUB (which boils down to inserting assembler code for printing debug messages, then go through the three minute flashing process of the Evo, see the results, repeat. The flashing is the killer...).

Instruction for getting GRUB sources from CVS are at the home page over at FSF (you want Legacy GRUB, not GRUB 2). Some patching will be needed...most are for relocating GRUB to disk sector 1 and onwards (and is a well documented configuration of GRUB, I just like to do it in the assembler when I've got the source anyway), while the "gate A20" problem is real patching.

Relocation: (Why are we doing this? Well, it turns out that the BIOS ignores sector 0/the MBR, and always executes sector 1. Why escapes me.)

First let's patch file "stage1.S" located in "grub/stage1/stage1.S":

Change the location of file "stage2" from sector 1 to sector 2, that's at the stage2_sector label.

Also the BIOS doesn't pass the boot device to GRUB so we must manually specify 0x80 (first hard drive) in place of GRUB_INVALID_DRIVE. Do this at the boot_drive label.

The changes to the file "stage1" can be done with a hex editor instead, see the hacking section in the GRUB docs.

That's it for file "stage1". Then we need to patch file "stage2" located in "grub/stage2/stage2.S":

First we must tell the stage2 preloader to load from sector 3 instead of sector 2. We do this by changing the blocklists at the end of the file "stage2/start.S". Just change the number after blocklist_default_start to 3.

Then it is time for the "gate A20" issue. A very good description of the problem that you might want to read is here:

http://www.win.tue.nl/aeb/linux/kbd/A20.html

Basically, since the EVO T20 is "legacy free", it doesn't have a keyboard controller, and GRUB hangs when trying to use it to set the Gate A20. Now, it seems that Gate A20 is always enabled on the EVO T20 (as it should have been on all computers, really).

Bottom line is, disable the A20 handling in GRUB and it will all work. The function itself is located in the file "stage2/asm.S", and is called from files "builtins.c" and "common.c" in the same dir. Comment it all out.

Because we have no keyboard (only USB, won't work in GRUB or any bootloader that I know of because most BIOSes emulate a regular keyboard during startup) we must precode GRUBs actions. "dhcp --with-configfile" is my strong suggestion, as it lets you specify the GRUB boot options from a DHCP server.

Create a new file "presetmenu"
default 0 timeout 0 title Load config from DHCP dhcp --with-configfile

(Why I have to write it as a menu instead of just the dhcp command is beyond me, but I must or it will enter an infinite DHCP loop...)

That's it for the patches. Let's configure it, my suggestion

./configure \

--enable-preset-menu=presetmenu \ --enable-natsemi \ --enable-pci-direct \ --disable-ext2fs --disable-fat --disable-ffs --disable-ufs2 \ --disable-minix --disable-reiserfs --disable-vstatfs --disable-jfs \ --disable-xfs --disable-iso9660\ --disable-md5-password --disable-serial

The important ones is the presetmenu, to enable direct PCI (we have no PCI access from BIOS), and the natsemi network card.

Now for installation.. make sure the loop device is set up (see first section), then
dd if=grub-src/stage1/stage1 of=/dev/loop0 bs=512 seek=1 dd if=grub-src/stage2/stage2 of=/dev/loop0 bs=512 seek=2
Finally, stick this somewhere in dhcp.conf on the server
option option-150 "(nd)/grub-menu.lst"
And put something fancy in the file "grub-menu.lst". My LTSP (Linux Terminal Server Project) setup
default 0 timeout 0 title LTSP root (nd) kernel /lts/kernel-version/bzImage-kernelversion init=/linuxrc rw root=/dev/ram0 initrd /lts/kernel-version/initrd-kernelversion.gz

Running X

Running X was a relatively quick job compared to the madness above. Only a single problem occured: Xorg detects the driver to be "geode", which is actually wrong. What we want is the "nsc" driver. And, National Semiconductor (seems like they made every chip in that box, not that strange).

Using LTSP, all I had to do was change "server" from "auto" to "nsc", and it would work.

More info from a comparable system: http://www.larwe.com/technical/geode_linux.html

The "vesa" driver also works, but nsc has some advantages: From a quick glance it looks slightly faster, it detects refresh rates much better and it displays a black screen during startup instead of garbage.

I couldn't get either driver to run in anything better than 16 bit color no matter the resolution (the nsc driver complains it doesn't know 24 bit, but 32 bit I just get the standard "No screens found"). Getting it to run.

More stuff that could be done

I probably won't do any of these, considering that they're not all that useful, but here's the ideas anyway

*

Now that we can have Linux running we probably have full access to the BIOS code. Reflashing it with a different boot picture would be good (it still has that "Windows Powered" logo when booting...grrr). Considering that most of the time you won't see this (see point 3) it's not very useful, but then again it's probably not terribly difficult to do if the right tools exists.

*

Look into video. Try to allocate more video memory for higher modes, get it to run in 24/32-bit, etc.

*

The boot phase is strange: Orange LED shows during very basic initialization. This happen when it gets power (pressing power button not required). After this, it progresses with booting with the LED and screen output turned off. Then it pauses right before Linux is about to boot. Ok, with a splash screen in the Linux kernel it could look good since it takes you past all the text stuff, but still: Why does it pause at all? How does it detect that Linux is about to boot, and is there any way to make it go all the way, so that when the power button is pressed we have an X session ready and waiting?

I have Linux appliances projects out there (commercial) running on Geode, and the driver is there, so if there is a higher potential than this it can probably be done.

Conclusions

Like I said, a more thorough investigation is probably in order, but this is all I'll do for now. Please mail me any experiences you have with it...also, if someone running the Evo T20 the intended way (Windows with Rapport on server) is reading this: It would be interesting to know the highest resolution/color they can set on it.


Resources


CategoryHardware

tried the stuff above - i'm not really a hacker - but ... yes it works !

my way:

since i saw a cheap t20 on ebay i've looked for information to run linux/bsd on this machine, then i found the solution from Dag Sverre and got a t20 with 96/128 memory configuration fom ebay (30$).

first i've downloaded the firmware-image (U96CPQ163.bin) from hp-support-site and compared it with the 48/64-imgefile from Dags machine: the offset differs, while it seems to be exactly the same stuff is contained and loaded up before starting up the NTe-Image (don't ask me).

  1. losetup -o 834236 /dev/loop0 U96CPQ163.bin (for the entire file)
  2. losetup -o 834748 /dev/loop1 U96CPQ163.bin (for the contained ntfs-filesystem)
  3. mount /dev/loop1 /mountpoint -t ntfs
  4. cd /mountpoint
  5. ls -l

first i tried to manipulate the bootloader with grub for nt

http://marc.herbert.free.fr/linux/win2linstall.html#grub-for-nt

but neither i can't edit the boot.ini file nor i can't write a "menu.lst"-file to the filesystem (file truncated, no rights ...)

... probing on a standard-machine: the default-entry in boot.ini seems to be useless, while the NT-startup-routine allways tries to boot up the NT-environment, also when i put the c:\grldr="Start GRUB" to this position (it would be fine, if somebody could explain me the sense of the default-entry in boot.ini) (... another problem there's no keyboard available on a t20 at this moment to change to the grub entry - with legacy keyboard support this no problem ... but for my evo-t20 - no way!)

so i tried "Making GRUB work"

commented out the A20-calls in "builtins.c" and "common.c" patched the other files as described (maybe i have forget/ignored to comment out the routine itself in "asm.S" - don't know if this is nessesary) and wrote the preset-menu (may be i failed here - i wrote all in the same row - shit!)

after compiling, patching

  1. dd if=my-compiled-grub-src/stage1/stage1 of=/dev/loop0 bs=512 seek=1
  2. dd if=my-compiled-grub-src/stage2/stage2 of=/dev/loop0 bs=512 seek=2

and flashing (with netxfer)... and rebooting ...

... the machine waited for me at the grub command-prompt (no preset-menu) ... without having a keyboard - kacke!

i've tried some etherboot images

http://rom-o-matic.net

(but which kind of it?) after testing with version 5.4.1 as a floppy bootable ROM-Image (.zdsk) and Binary ROM-Image (.zrom) i tried a HD-Image (experimental)(.zhd) but forget to enable "direct-PCI-Support" (machine boots up (hooray) ... and - no wonder - can't find a NIC) ... after enabling Direct-PCI i got a compiler-error

i've tried version 5.4.2pre2 with these settings:

1. natsemi:dp83815 -- 0x100b,0x0020

2. HD (experimental) Hard Disk Partition Image (.zhd)

3. default-values + CONFIG_PCI_DIRECT

and got a "eb-5.4.2pre2-natsemi.zhd"

now i have patched the firmware with this image

  1. dd if=eb-5.4.2pre2-natsemi.zhd of=/dev/loop0 bs=512 seek=1

then i have flashed the patched U96CPQ163.bin to my t20 ... rebooted ... and:

BINGO! ### a DHCP/PXE-Request on my sniffer-screen and the machine replies on ICMP-Requests

i hope the hardest part is done. next days i will try to boot some LinTS-images and my own bsd-stuff for building a diskless FreeRADIUS-Server and a RDP/X-Client ... will tell you about my experiences

floaty

changing bootscreen und booting up pxes-prebuild client over the network

bootscreen:

the bitmap from bootup is ugly :

http://mitglied.lycos.de/szeissig/bilch.bmp

my screen is not really beautifull but inspired (from this web-site) :

http://mitglied.lycos.de/szeissig/bilch-patch.bmp

patching the bitmap is easy:

  1. note the filesize of your startup.bmp while flashing firmware
  2. the bitmap is before your (unused) mbr (remember the offset - for me starts at 834236) the filesize is 14120 bytes
  3. you find the bmp-header: BM(7......
  4. make your own screen.bmp (640 x 480 x 8 BPP)
  5. patch the new screen with dd or a hex-editor
  6. flashing - finished

booting the prebuild-client:

http://downloads.2x.com/pxes/1.1.1/PREBUILT/

  1. download *.nbi file for etherboot
  2. set up dhcp / tftp (for me tftpd32 ... yes!, quick & dirty!)
  3. boot the machine (have a look on your brand new start-up screen) - fertich
P192168100251# dmesg
Linux version 2.4.32-2pxes (diego@thindev) (gcc version 3.3.5 (Debian 1:3.3.5-8)) #3 Tue Feb 21 18:02:22 EST 2006
BIOS-provided physical RAM map:
 BIOS-e801: 0000000000000000 - 000000000009f000 (usable)
 BIOS-e801: 0000000000100000 - 0000000007d10000 (usable)
125MB LOWMEM available.
On node 0 totalpages: 32016
zone(0): 4096 pages.
zone(1): 27920 pages.
zone(2): 0 pages.
DMI not present.
ACPI: Unable to locate RSDP
Kernel command line: rw root=/dev/nfs ramdisk_size=17080 initrd=pxes-1.1-1PB.squash ro root=/dev/ram
Initializing CPU#0
Working around Cyrix MediaGX virtual DMA bugs.
Detected 300.683 MHz processor.
Console: colour VGA+ 80x25
Calibrating delay loop... 599.65 BogoMIPS
Memory: 106808k/128064k available (1351k kernel code, 20868k reserved, 528k data, 108k init, 0k highmem)
Checking if this processor honours the WP bit even in supervisor mode... Ok.
Dentry cache hash table entries: 16384 (order: 5, 131072 bytes)
Inode cache hash table entries: 8192 (order: 4, 65536 bytes)
Mount cache hash table entries: 512 (order: 0, 4096 bytes)
Buffer cache hash table entries: 4096 (order: 2, 16384 bytes)
Page-cache hash table entries: 32768 (order: 5, 131072 bytes)
Working around Cyrix MediaGX virtual DMA bugs.
CPU:     After generic, caps: 00808131 00818131 00000000 00000001
CPU:             Common caps: 00808131 00818131 00000000 00000001
CPU: NSC Geode(TM) Integrated Processor by National Semi stepping 02
Checking 'hlt' instruction... OK.
POSIX conformance testing by UNIFIX
ACPI: Subsystem revision 20040326
ACPI: Interpreter disabled.
PCI: Using configuration type 1
PCI: Probing PCI hardware
PCI: Probing PCI hardware (bus 00)
PCI: Fixup for MediaGX/Geode Slave Disconnect Boundary (0x41=0x9c)
Linux NET4.0 for Linux 2.4
Based upon Swansea University Computer Society NET3.039
Initializing RT netlink socket
Starting kswapd
devfs: v1.12d (20021015) Richard Gooch (rgooch@atnf.csiro.au)
devfs: boot_options: 0x1
Squashfs 2.2-r2 (released 2005/09/08) (C) 2002-2005 Phillip Lougher
intelfb: Framebuffer driver for Intel(R) 830M/845G/852GM/855GM/865G chipsets
intelfb: Version 0.7.7, written by David Dawes <dawes@tungstengraphics.com>
tridentfb: Trident framebuffer 0.7.5 initializing
keyboard: Timeout - AT keyboard not present?(ed)
keyboard: Timeout - AT keyboard not present?(f4)
pty: 256 Unix98 ptys configured
Serial driver version 5.05c (2001-07-08) with MANY_PORTS SHARE_IRQ SERIAL_PCI enabled
RAMDISK driver initialized: 16 RAM disks of 17080K size 1024 blocksize
NET4: Linux TCP/IP 1.0 for NET4.0
IP Protocols: ICMP, UDP, TCP, IGMP
IP: routing cache hash table of 512 buckets, 4Kbytes
TCP: Hash tables configured (established 8192 bind 16384)
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
RAMDISK: squashfs filesystem found at block 0
RAMDISK: Loading 17078 blocks [1 disk] into ram disk... done.
Freeing initrd memory: 17128k freed
VFS: Mounted root (squashfs filesystem) readonly.
Mounted devfs on /dev
Freeing unused kernel memory: 108k freed
usb.c: registered new driver usbdevfs
usb.c: registered new driver hub
usb-uhci.c: $Revision: 1.275 $ time 18:01:58 Feb 21 2006
usb-uhci.c: High bandwidth mode enabled
usb-uhci.c: v1.275:USB Universal Host Controller Interface driver
PCI: Setting latency timer of device 00:13.0 to 64
usb-ohci.c: USB OHCI at membase 0xc00cc000, IRQ 6
usb-ohci.c: usb-00:13.0, Compaq Computer Corporation ZFMicro Chipset USB
usb.c: new USB bus registered, assigned bus number 1
hub.c: USB hub found
hub.c: 2 ports detected
usb.c: registered new driver hid
hid-core.c: v1.8.1 Andreas Gal, Vojtech Pavlik <vojtech@suse.cz>
hid-core.c: USB HID support drivers
mice: PS/2 mouse device common for all mice
hub.c: new USB device 00:13.0-1, assigned address 2
hub.c: USB hub found
hub.c: 4 ports detected
hub.c: new USB device 00:13.0-1.2, assigned address 3
input: USB HID v1.10 Keyboard [  USB Keyboard] on usb1:3.0
input: USB HID v1.10 Device [  USB Keyboard] on usb1:3.1
hub.c: new USB device 00:13.0-1.3, assigned address 4
input: USB HID v1.10 Mouse [05e3:1205] on usb1:4.0
natsemi dp8381x driver, version 1.07+LK1.0.17, Sep 27, 2002
  originally by Donald Becker <becker@scyld.com>
  http://www.scyld.com/network/natsemi.html
  2.4.x kernel port by Jeff Garzik, Tjeerd Mulder
eth0: NatSemi DP8381[56] at 0xc883d000, 00:80:64:1e:48:f3, IRQ 10.
eth0: link up.
usb-uhci.c: $Revision: 1.275 $ time 18:01:58 Feb 21 2006
usb-uhci.c: High bandwidth mode enabled
usb-uhci.c: v1.275:USB Universal Host Controller Interface driver
inserting floppy driver for 2.4.32-2pxes
Floppy drive(s): fd0 is unknown type 7 (usb?), fd1 is 360K PC
floppy0: Unable to grab IRQ6 for the floppy driver
SCSI subsystem driver Revision: 1.00
Initializing USB Mass Storage driver...
usb.c: registered new driver usb-storage
USB Mass Storage support registered.
Uniform Multi-Platform E-IDE driver Revision: 7.00beta4-2.4
ide: Assuming 33MHz system bus speed for PIO modes; override with idebus=xx
isapnp: Scanning for PnP cards...
isapnp: No Plug & Play device found
Soundblaster audio driver Copyright (C) by Hannu Savolainen 1993-1996
sb: No ISAPnP cards found, trying standard ones...
SB 4.12 detected OK (220)
Too much work in interrupt on uart401 (0x330). UART jabbering ??
parport0: PC-style at 0x3bc [PCSPP(,...)]
parport1: PC-style at 0x378 [PCSPP(,...)]
parport2: PC-style at 0x278 [PCSPP(,...)]
lp0: using parport0 (polling).
lp1: using parport1 (polling).
lp2: using parport2 (polling).
P192168100251#