Major changes from 1.0 to 1.1

+ Everywhere: Can compile two versions of the code, under gcc or bcc
(Bruce Evan's cc), so that 16-bit boot PROMs can be made. See
netboot-16/README.16 for more details.

+ Everywhere: Removed duplicate defines, e.g. everybody defined their
own ID for ethernet address size (6). Put IDs for magic numbers in
various places.  Still some inconsistency, ETH_ and ETHER_ are used.

+ Everywhere: Added prototypes of functions to netboot.h.  Removed
unused variables.

+ Makefiles: Rewritten.

+ netboot.h: Added define for TFTP_MAX_PACKET = 512. Increased TIMEOUT
for tftp packets to 180 (about 10 seconds) so that tftp servers would
be able to retransmit blocks.

+ main.c: Rewrote tftp(). Original was intended to get only one block
and the strain on the structure due to modifications was showing.
Increased config_buffer size to TFTP_MAX_PACKET+1 to avoid special
casing data length=512.

+ linuxloader.c: Cleaned up the code in some places, especially
linux_tftp.  Moved the bootp reply block into bss space instead of a
fixed location such as 0x90000.

+ ns8390.c: Removed one bug regarding packet length. packetlen was
wrongly shortened when packets wraparound the ring buffer.

+ 3c509.c: Changed some of the gotos to returns.  Removed ARP code since
that's already done elsewhere.  Removed interrupt enable and Rx early
notification (we can't do anything with the packet until it's complete
anyway).

Major changes from 1.1 to 1.2

+ Small bug in makerom.c. Extra semicolon shortened for loop and made
double checking code invalid.

+ Added version identification to startup string.

+ Gathered external declarations into netboot.h.

+ New compile time option for netboot-32. If INT19H is not defined,
then boot ROM takes control as soon as BIOS scans it. This may solve
the problem of some BIOSes not calling the boot ROM at INT19H. This
may be when disks are detected by the BIOS, i.e. the machine is not
truly diskless.

+ Added a new program, test.c, for making test ROMs to verify that
the BIOS recognises the ROM.

+ New directory contrib contains contributed code. Currently contains
masq by Gerd Knorr: make a boot floppy without DOS
comboot-1.0 by Adam J. Richter: also make a boot floppy without DOS.

+ New mknbi-1.4 from Gero Kuhlmann and Markus Gutschke. This one handles
the new bzImage format.

Bumped up version to 2.0 because we are so listed in the Linux 2.0
distribution and this would avoid confusion.

Major changes from 2.0 to 2.1

+ Added LSM for transname-patch to contrib.

+ Added patch for serial console from Claus Heine
<claus@momo.math.rwth-aachen.de>.

+ Claus Heine contributed patches to the 2.0.21/22 kernel sources to
allow NFS swap.

+ Markus Gutschke provided fixes to start2.S so that main can return to
the ROM code if user doesn't want to ROM boot.

+ Added code to main.c to timeout on the prompt and assume Y or N
for the answer. Timeout and answer configurable.

+ Made NE1/2000 probe addresses configurable from Makefile. Fixed up
autoscan code.

+ Made NFS_BOOT a configurable option. By undefining it, only tagged
file load is supported but ROMs are under 8k.

+ ROMSIZE is not used by makerom now. -s flag controls size of ROM
at runtime.  So don't need to recompile makerom if ROMSIZE changes.

+ Updated netboot-16 for the latest dev tools from the Linux-MT
project. Use the size.c from there.

+ Got rid of _main() in main.c, instead using #ifdef ELF in start2.S.
ELF is preferred now anyway.

+ Changed bcompare to standard bcmp and reversed sense of result.
Reason: to use standard library if available, e.g. Netboot-16.

+ Common Makefile for 32 bit and 16 versions, with differences in
Config files.

Major changes from 2.1 to 2.2

+ New device driver for the Crystal Semiconductor CS89x0 chipset family.
  (because of legal problems, this code is currently in the "contrib"
  directory.)

+ Added support for loading BOOTP extension files (c.f. RFC1533).

+ If we have to go thru a gateway, then use the one that has been used for
  the BOOTP daemon. If the BOOTP daemon is directly accessible, then use
  the first entry in the "gw" gateway list instead.

+ For all retries, back off according to RFC951 by randomizing the timeouts
  and exponentially increasing them until an average of one minute is
  reached.

+ Warn if code will not fit into chosen ROM size. Happens only if the
  autosizing in the Makefile somehow fails.

+ Modified the code for enabling the A20 gate; this could increase
  compatibility, but I still have to hear from users.

+ The copyright message will now reflect, which features have been
  enabled at compile time.

+ Added support for selecting different boot images from a menu. As a side
  effect, this can be used to upgrade the ROM without having to burn a new
  image (c.f. README.VendorTags).

+ Display a "message of the day" that is provided by the BOOTP daemon.

+ The sanity check for detecting a Linux kernel was too strong to properly
  recognize a 2.1.x kernel; this has been fixed.

+ Trys to negotiate for 1432 octect blocks, if the TFTP daemon knows
  about RFC1783.

+ We can optionally boot from local disk, if the BOOTP server cannot
  be contacted.

+ Restructured directories: got rid of netboot-freebsd to reduce confusion
  with Gero Kuhlmann's netboot, moved documentation to doc and renamed
  directories to be more in line with Unix conventions. Edited documentation.

+ Created a dispatch table for NIC routines so that we can include one
  or more drivers in one binary. Renamed all entry points in the driver
  sources. Created two new files, config.c, which holds the dispatch
  table and can be conditionally compiled; and nic.c which contains the
  extern variables referenced by the drivers.

+ Automatically decide what size ROM is needed by doing a size and then
  choosing the correct startup object. Currently caters for 8k, 16k and
  32k ROMs.

+ Optionally include a simple interpreter for ANSI escape sequences. This
  allows for fancier boot menus.

+ Remove patches directory which is mostly relevant to 1.x kernels only.
  Everybody should use 2.x kernels now. If for some reason you need those
  patches, get an older release of etherboot.

+ Fix Makefile to work even if . not on PATH.

+ Sadly, 32 bit ROMs are now > 8kb. We will try to reduce the size by
  conditionals in a later release. For now, use version 2.1 if your NIC
  won't take > 8kb ROMs.

+ Release as 3.0b1 because of the large amount of changes.

+ Fixed a few problems with the code for ANSI escape sequences and
  added optional support for displaying graphics.

+ Optional support for password protected boot images.

+ Optional support for booting from block devices (floppy, hd).

+ The bootp server can pass additional parameters to the loaded kernel
  image (currently, this only  applies to Linux)  and the user can  be
  allowed to edit a commandline; the latter is optionally protected by
  a password scheme.

+ Set the warm-boot flag when the main routine is entered.

+ Release as 3.0b2

+ Added code for updating the FlashCard EPROM over the network
  (contrib/flashimg).

+ Added very simple code for turning a ROM image into a network loadable image.
  This is useful for debugging, but some users without flash EPROMs might
  appreciate the possibility to load a fully fledged image from a very basic
  configuration.

+ Optional support for compressing the ROM images. Please read 
  doc/COPYING.compressor before using this feature.

+ Sync'd our source tree with Gero's netboot-0.5; this means that you now need
  the as86 from the ELKs project in order to recompile all of the files. This
  should not affect normal usage, though.

+ Added "mknbi-blkdev" for booting from local block devices.

+ Fixed some bugs in ppmtoansi.c and bootmenu.c

+ Renamed reference compressor implementation to compressor.exp otherwise
  make tries to use it and it should not be turned on by default.

+ Zero'ing BSS in 16-bit version has to be done to _end, not to A0000
  because it's executing in a segment, not in flat address space in 16
  bit mode.

+ Ken Yap contributed a quick and dirty Perl script for people who use
  netboot to test ELKS.  So far I'm the only one I know of; maybe the
  others are silent. :-(

+ Release as 3.0

Major changes from 3.0 to 3.1

+ 4 versions of etherboot can be built for a NIC: .com for testing and
  .rom for burning into EPROM, and corresponding compressed versions:
  .lzcom and .lzrom.

+ The loaders are now separate programs which are prepended to the
  etherboot binary. This allows them (plain and uncompressing versions) to
  be maintained separately and gives a bit more RAM to the etherboot code.

+ No need to define ROMSIZE in the build. makerom automatically discovers
  the ROM size needed and fills in the size field in the ROM. This
  simplifies the build procedure.

+ 16 bit versions use the same loaders as the 32 bit version. Also fixed
  two bugs in the 16 bit versions: (1) a non-8086 instruction in zloader,
  (2) setting warm-boot flag in main which zapped some code.

+ Fixed a documentation bug on the ANSI escape sequences.

+ Include netboot-0.5.3 distribution from Gero Kuhlmann.

+ Included some contributions from Dickon.Reed@cl.cam.ac.uk: Running
  display of Kbytes loaded, line of delimiters after loading complete,
  a temporary hack to address timing problems with the 3C509, and some
  Makefile cleanup. The first two changes need to be enabled in Config
  with defines.

+ mknbi-blkdev seems to have been left out by Gero Kuhlmann. Add to
  contrib directory. You probably have to make a symlink to it from
  the netboot-0.5.2 directory.

+ Release as 3.1

Major changes from 3.1 to 3.2

+ 16 bit version now can load to extended memory, if it exists. On
  a 8086/8 this will silently fail.

+ Cleaned up interface between main body of code and NIC drivers. No
  global variables referenced in NIC driver, everything is passed
  through a structure. Only the probe function is visible outside,
  pointers to the others are returned in the structure.

+ Implemented autoprobe for 3c503. Also simplifies code at same time.

+ Removed ARP response code in ns8390.c. Don't think we need to respond
  to ARP requests because other machines will do gratuituous ARP when
  boot code sends out bootp request. Are there cases where this is not
  true? Gateways?

+ Added 1 second timeout to routine that clears the keyboard buffer
  in case there is no keyboard.

+ Added a skeleton driver that can be used as a template for new NIC
  drivers.

+ ./lzhuf in Makefile so that it will run even if . is not on path.

+ Fixed comboot to work on 286s also. Won't work yet on 8086/8.

+ Fixed mknbi-blkdev to configure properly under netboot-0.7.

+ Added David Munro's PCI code adapted from Linux. Currently has entries
  for PCI NE2000 clones. Generalised it to probe other PCI cards later on.

+ Moved twiddle() outside NIC driver except where used to provide a delay.

+ Fixed bug in Makefile spotted by Ton Biegstraaten. Should prepend
  ZLOADER to make all.lzcom, not LOADER.

+ Charlie Brady donated a NE2100 (LANCE) card, so Ken Yap wrote a driver
  for it. Should work for other LANCE (7990, etc) based cards with some
  modifications.

+ Markus Gutschke wrote rom-scan, and it is in contrib/.

+ Hack rom-scan.c so that a DOS version can be compiled.

+ Removed all.* targets from Makefile. The NIC specific loaders should be
  used in preference as the all.* loaders can run out of memory.

+ Added a disable routine to dispatch table so that cards can be turned
  off before the loaded code is executed.

+ Make INT19H the default.

+ Rogier Wolff persuaded AW computer systems to contribute the Intel
  EtherExpressPro 100 driver. The binary to hex converter in contrib/
  is also from Rogier Wolff. Part of the work is sponsored by BitWizard
  NL (www.bitwizard.nl).

+ Distribute with a subset of netboot-0.7.2 that doesn't have the bootrom
  portion.

+ Release as 3.2.

Major changes from 3.2 to 4.0

+ Merged in Vlad Lungu's patches for DHCP support, ifdef'ed by DHCP_SUPPORT.

+ XID matching fix also provided by Vlad Lungu.

+ Merged in William Arbaugh's patches to make eepro driver work properly.

+ Add to contrib/ better bin2intelhex from Jean Marc Lacroix.

+ Patches from Jim Hague (thanks!) for the following:

* Added PIO mode for 3c503 to ns8390.c. Changed the card detection to
  detect shared memory or PIO and use the selected one, and removed
  a jumper check that failed on the Bull (no jumpers).

* Added more #ifdefs to ns8390.c to include only code relevant to the
  card being compiled, and removed unnecessary run-time card vendor
  behaviour branches - it's all now #ifdefd.

* Added a -3 parameter to makerom to set the last two bytes to 0x80.
  These are the values they have in the 3Com Etherboot image that was
  in the Bull. (Also on the 3c503 card I have - Ken.) Also altered the
  Makefile to add this parameter when building a 3c503 image.

* Modifies lzhuf.c, objdump86.c & size86.c to work on either-endian systems.

* Adds a -DT503_AUI config paramter to let you choose AUI or BNC on
  3c503s.  Previously it defaulted to AUI, and you had to change the
  code to alter it.

* Changed the BCC include directory to /usr/bcc/include. If you use the
  BCC include files they don't define u_char, u_short etc. I've added
  these into linuxdef.h, ifdef'd on BCC.

* Adds a trivial Linux 3c503 driver patch to the contrib directory
  to let it spot these 'ere Bull things.

+ Charlie Brady confirmed that the Lance driver works with PCI so there
is a new lancepci driver now.

+ Removed support for NFS_BOOT; only TFTP booting supported now.

+ Removed support for linear images; only tagged images supported now.

+ Removed PRIORIZEBOOTPKERNEL and BOOTPKERNELONLY. Bootp reply must
specify kernel name.

+ Bug fix for 16 bit version of 3c509 driver: sign bit propagation bug.

+ Revert to non-pausing versions of out[bw] and inb for NEx000 driver
because of reports of timimg problems on some cards.

+ New driver for NI6510. Just a simple tweak of the lance driver for
different ID bytes and different register offsets. According to the
NI6510 driver in Linux, the NI6510 Etherblaster is more like the NE2100
and would be detected by the NE2100 driver. So use that one instead.

+ New drivers for 3c507 and NI5210, both of which use the i82586 chip. It
works properly for both now. I needed to have enough receive buffers to
make sure i82586 never goes into an out of resources state. 3c507 driver
has one quirk, it only responds after second bootp request. I seem to
remember something about this problem of losing the first packet after
initialisation in early Linux discussions.  I am also now convinced
that Intel designers have weird minds. BTW, NI5210 driver assumes 8k
RAM because if you put the ROM on the NIC you can only have 8k RAM. If
you are putting the ROM off-board and you want 16k, well, talk to me.

+ Call nic_disable routine just before jumping to loaded image. This
does nothing in most drivers, but may have side effects as the nic_reset
routine used to be called instead.

+ One of those "why I didn't think of it before" ideas: A modified version
of comboot, called floppyload, that is prepended to the .rom image rather
than the .com image and then both written raw to a floppy for testing the
bootrom. All we have to do is jump to an entry point in (z)loader that
skips the INT19H stuff. Relocation will happen automatically.  Now the
.com images and comboot are superfluous, unless one is masochistic enough
to want to try to test under DOS.

Added .fd0 targets to Makefile. Saying make <card>.fd0 will make
floppyload.bin and <card>.rom, and cat both to /dev/fd0. Naturally the
drive must be writable and you must have a floppy in the drive.

+ New driver for Tiara (Fujitsu EtherStar). This was one of the easiest
drivers to write. But the chip apparently has some quirks; there is no
Linux driver for it in the standard distribution and the email address
of someone who wrote an alpha driver is invalid. Perhaps I'll bump into
an AT1700 (which has a similar chip) one day and be able to reuse a lot
of the code.

+ Updated contrib/mkelksnbi for ELKS 0.0.68.

+ Confirmed that the SMC8216 driver works. 8416 not tested, it's a PnP
card. Can anybody confirm this?

+ New contributed software: p910nd, a tiny printer daemon suitable for
diskless hosts.

+ Made ANSIESC work for Etherboot/16.

+ Fixed Makefile and Config.* so that it works with old binutils.

+ Included a subset of netboot-0.7.3.

+ Released as version 4.0.

Changes between 4.0 and 4.1

+ Patches by Andrew Coulthurst for eepro100b.

+ Patches by Doug Ambrisko for booting Windows95 after answering N to
the boot from network question. Added conditional code to cope with
broken DHCP server and TAG 128.

+ Put version and driver identifier at end of ROM image if it fits,
to help identify ROMs in future.

+ Capture ROM segment address and length to help choose between multiple
NICs later.

+ Don't clear all of memory because it will destroy return address
on stack.

+ Changes to comboot-1.2 (although obsolete) to run on 8088s.

+ New version 0.2 of p910nd, a non-spooling printer daemon.

+ Patches by Alex Harin to prepended loaders and makerom to generate PnP
ready ROMs. Modified makerom to automatically detect PnP and PCI headers
and do the right thing. Added option to change the vendor and device IDs.

+ Augmented documentation for 2.1 and above kernels. Kernel now wants to
mount /tftpboot/<hostname in bootptab> rather than /tftpboot/<ip address>
as the root FS.

+ Changed all the outb* and outw calls in drivers (except eepro) to be
OUTB* and OUTW, then defined macros to translate to out[bp]* for both
Etherboot/32 and Etherboot/16, instead of pasting macros from Linux
include files because apparently they've changed in 2.1. At the same
time reversed the arguments in start16.S for outb and outw to match the
Linux convention. Unfortunately the out[bw] usage came that way from
FreeBSD. Someday I'll reverse the arguments in the C files properly.
For any new driver writers, you should use the Linux order now.

+ New contributed utility, disnbi for decoding and extracting network
boot images.

+ Martin Atkins contributed mntnbi for mounting DOS NBIs.

+ Peter Dobcsanyi contributed vendor and device IDs for the Netvin
NE2000/PCI clone.

+ adam@mudlist.eorbit.net contributed RARP code as alternative to
BOOTP/DHCP. Activated by RARP_NOT_BOOTP define.

+ Added link to Claus-Justus Heine's NFS swap Web page and updated the
contrib directory.

+ Disabled max packet length check in ns8390.c. Caused spurious Bogus
packet messages in some cases and doesn't seem that useful a sanity
check anyway.

+ Daniel Engstrom contributed a SMC9000 driver.

+ Didier Poirot contributed an Etherpower II (EPIC 100) driver.

+ Added bug fix by Attila Bogr for bootmenu.c and patch to main.c to
remove looping menus on failure. Also code for ARP replies and TFTP
block retransmit (#ifdefed because controversial).

+ Code cleanup of tftp and tftpd also by Attila Bogr.

+ Nathan R. Neulinger fixed a bug with block being declared short instead
of u_short in tftpd.c, which limited transfers to 32k blocks. Fixed
problem with field tu_block being declared as signed short in many
platforms by including fixed version of tftp.h. Fixed tftp also for
good measure.

+ New mini-HOWTO on a "ssh terminal".

+ Andreas Mack pointed out that eepro100 doesn't compile on 2.1 and
2.2 kernels. Removed unnecessary include of bios32.h and reliance on
definition of virt_to_bus in kernel headers in eepro100.c, epic100.c
and lance.c.

+ David Sharp contributed a Tulip driver written for FreeBSD netboot.
Ken Yap ported to Etherboot. Not tested yet because no hardware.

+ Replaced references to arptable[ARP_CLIENT].node to nic->node_addr in
eepro100.c and epic100.c as they should be.

+ Greg Beeley of LightSys Technology Services contributed a 3c905b
driver. Be sure to read the release notes in 3c905b.txt.

+ Gnter Knauf suggested making the prompt strings more generic and to
put a newline after the answer. Beware, N now means Network boot and
not No to network boot.

+ Alex Nemirovsky contributed some patches for BIOSes that use an extended
space at the top of 640k. Also some code for BIOSes that don't implement
BIOS32 correctly or at all.

+ Use PCI extension BIOS header only for PCI cards, all others use legacy
extension BIOS header.

+ Klaus Espenlaub contributed various cleanup patches to the code.  Also
introduced Rainer Bawidamann's code, see next paragraph.

+ Rainer Bawidamann contributed a Realtek 8139 driver.

+ Simplified rules for building .bin files, use -b of as86 so we
don't need ld86 or objdump86 now. But later found that a fixed as86 is
required. So supply preassembled binaries. The keen hackers can get the
fixed tools.

+ Georg Baum contributed a Schneider & Koch G16 driver. Only the
32 bit version works at the moment; even though the 16 bit version
compiles, it won't work because the current code assumes flat memory
addressing. Anybody who needs the 16 bit version should feel free to
fix it.

+ Reduce size of ROM image loaded by floppyboot.bin from 64kB to 32kB
for a slight speedup in loading. None of the images are even 32kB anyway.

+ Updated some of the documents in doc/sgml.

+ Klaus Espenlaub sent in a totally revamped start32.S, using the code16
and code32 directives in recent GNU as (so if this file won't assemble
maybe your as version is not recent enough). Also patches to the menu
handling code.

+ jluke@deakin.edu.au sent in a fix for the WD/SMC8013 long ago which
I finally got around to verifying.

+ start16.o and start32.o are supplied for those people who have problems
compiling start*.S with as/as86.

Released as Etherboot-4.2.0

+ One line fix to 3c509b by Greg Beeley for Wake-On-LAN support.

+ Added patches by Klaus Espenlaub that I forgot.

+ Made AS_PSEUDOS not the default, assume that people have sufficiently
up-to-date GNU as. If not they can uncomment that line.

Released as Etherboot-4.2.1

+ Disable 3c509 after loading finished or port may be unusable.

+ Woops, forgot to up the version's last digit in the last release.

+ Stephan Bauer sent in a device ID for config.c for 21142 chip Tulips.

+ Ifdef out input overrun recovery code (it's really only needed for
NE2000s) and use SHMEM by default for 3c503s to bring the ROM size down
below 8kB.

+ Added code to start32.S to detect < 386 and exit so it doesn't hang
the computer. Doesn't print a message yet.

+ Changed README.sgml to recommend that TFTP use a separate directory
from NFS, /tftpdir. Added sample configuration file for DHCPD and caveat
about the name of the root directory when using DHCPD.

+ Revamped atnetboot.sgml: use mtools most of the time, write about a
few example applications.

+ Quick Perl script for converting bootptab to dhcpd.conf.

Released as Etherboot-4.2.2

+ Forgot to put new version of src/start32.o. Not urgent though, the
extra code only guards against Etherboot/32 being executed on < 386.

+ Supply version preassembled start32.o with ANSIESC and FLOPPY defined
for those with a deficient as. Define dummy handleansi routine in
ansiesc.c so that it will still link even if this start32.o is used.

+ Edited *.asm so that they can be assembled by either as86 (ELKS version)
or nasm.

+ Changed Makefile so that one can choose between no as86, as86 or nasm.

+ Added Mark Burazin's conditional code for Compex RL2000 PCI NIC.

+ Increase delay during probe phase to 10ms as 1ms is too low for some
3c509 boards.  Donald Becker's Linux driver hints that > 2ms is needed
to be safe. Also call t509_disable in case board was active.

+ Update README.sgml date and version.

Released as Etherboot 4.2.3.

+ rtl8139.c: Reduce the number of transmit buffers to reduce footprint.
Remove polling loop in *_loop(), caller already does the looping.

+ main.c: Fix code indentation in bootp().

+ Small corrections to documentation.

Released as Etherboot 4.2.4.

+ Matthias Meixner found a longstanding bug in rtl8139.c where it was
testing the wrong bit for the existence of a packet in the buffer.

+ Added paragraph to documentation about caveats for hosting NFS root
on a different architecture.

Released as Etherboot 4.2.5.

+ Updated mklnim for RH6.0 which uses a different floppy image and
requires 'network' to be appended to kernel parameters.

+ Reversed sense of #ifdef DHCP_SUPPORT to #ifndef NO_DHCP_SUPPORT and
so forth, i.e. the default is with DHCP support.

+ Suggest reducing options for rtl8139 driver to minimise footprint in
release notes.

+ Jim McQuillan provided changes to support the SMC1211, which uses the
RTL8139 chip.

+ Changed lret in start32.S to int $0x19 so that it doesn't depend on
the return location being there.

Released as Etherboot 4.2.6

+ Succeeded in booting FreeDOS with the kernel in the tagged image rather
than on the ramdisk, i.e. the boot process jumps directly to the kernel
in memory. The utility is mkfreedosnbi in contrib/.

+ Additions to the documentation.

Released as Etherboot 4.2.7

+ Matt Hortman pointed out that the BIOS clock value returned by INT1AH
rolls over at midnight, causing delay loops that cross midnight to
fail. Fix was to keep track of midnight crossings in currticks so that
currticks' return value is monotonically increasing.

+ Added paragraph to documentation about editing PCI vendor and device
IDs in Makefile if needed.

+ Merged in FreeBSD support code supplied by Doug Ambrisko. Thanks!

Released as Etherboot 4.2.8

+ Cleaned up the FreeBSD support in osloader.c. Symbols to use in Config
are to select image type, not FreeBSD specific now.

+ Added simple signature checking code in floppyload.asm in case somebody
forgets to append a ROM image or tries to load a non-ROM image.

+ Merged in 3c90x and multiple PCI bus support by Steve Smith. Thanks!

+ Included NT-Diskless-Terminal HOWTO by Pavel Tkatchouk.

+ Included snapshot of H. Peter Anvin's rewrite of tftp.

+ Combined tftp and tftpd directories, and touched up Makefile to reflect
normal Linux install directories.

Released as Etherboot 4.2.9

+ NT-Diskless-Terminal HOWTO renamed to Diskless-From-NT.

+ One line fix to mknbi-dos to recognise FAT16 filesystems as legal.
Updated first.S in mknbi-dos so that it can be assembled by nasm and is
also suitable for FreeDOS with the right define. Still backward compatible
with as86. Bug fix at line starting getnm3:.

+ Updated first.S in mknbi-linux so that it can be assembled by nasm also.

+ Locate bootp data block at 0x93C00-0x93FFF to free up 1024+ bytes in
[0x98000-0x9FFFF]. Etherboot/32 only, business as usual for Etherboot/16.

+ Patch from Attila Bogr to make CONGESTED a compile option for
TFTP_TIMEOUT. Also corrected spelling of his name in README.

+ Fixed start32.S to assemble correctly with gas version 2.95. Must
define GAS295 in CFLAGS.

+ Add contrib/3c90xutil containing a utility for handling 3c90x EEPROMs
in situ.

+ Update p910nd to 0.3. client.pl now checks if hostname is known and
uses more convenient routines from Socket module.

+ Moved the Flashcard directory into a separate package as it's of
limited interest.

Released as Etherboot 4.2.10

+ Fixed bug introduced in 4.2.10 in code for handling bootp extension
files.

+ Change in 3c90x.c so that it can compile under FreeBSD.

+ Patch to tulip.c from Nick Lopez to handle Macronix 98715 (Tulip clone).

Released as Etherboot 4.2.11

+ Prefer RFC1533_GATEWAY to giaddr for routing tftp packets.

+ Print out both relay address and gateway address.

+ Stored arptable ipaddrs in network byte order. Makes everything simpler
and we can get rid of convert_ipaddr. Now %I in printf and inet_ntoa
have to deal with NBO. We reduce the footprint by ~100 bytes.

+ Renamed setip to more standard inet_ntoa and change argument type to
suit, using in_addr.

+ Replaced bcmp by memcmp, bzero by memset, and bcopy by memcpy. Allows
us to use optimised versions in inline assembler.

+ Replaced OUTB and OUTW by outb and outw with arguments reversed.
Finally all the out macros are the same form as for Linux.

+ Replaced \r\n and \n\r in output strings with \n since putchar
now prints \r before \n. Then changed printf("\n") to putchar('\n').
Like Unix convention for strings now.

+ Got rid of test.c and Makefile rules connected with it.

+ Added patch to eepro100.c by Matt Hortman to correct PCI latency.

+ Merged in Marty Connor's ntulip.c. Now supports Macronix 98715 and
Linksys LNE100TX.

Released as Etherboot 4.2.12

+ Anders Larsen contributed mkQNXnbi, for generating tagged images from
QNX kernels.

+ Bernd Wiebelt contributed code to request vendor tags in DHCP.

+ Fixed more bugs introduced in 4.2.10 in code for handling bootp
extension files.

Released as Etherboot 4.2.13

+ Marty Connor reduced RTL8139 footprint by using only one transmit
buffer instead of 4. Now RTL8139 is reliable.

+ Moved initialised data declaration from ns8390.h to ns8390.c.

+ Contributed utilities for wake-on-LAN: wol.c (Bob Edwards) and wake.pl.

+ Makefile for 3c90xutil/romutil.c. -O is essential when compiling.

+ Happy year 2000!

Released as Etherboot 4.4.0

+ VIA-Rhine driver contributed by Paolo Marini. Footprint reduced by
Marty Connor.

+ Netgear FA310TX (Tulip clone, LC82C168 chip) support added by Marty
Connor.

+ Support for 3C905C added by Marty Connor.

+ mklnim updated to support SuSE 6.x also.

Released as Etherboot 4.4.1

+ New configuration file scheme to specify what ROMs are built.

+ Adam Fritzler contributed 3c529 (MCA version of 3c509) support in
driver.

+ Marty Connor wrote a version of 3c90xutil/romutil for 905C NICs.

+ Gnter Knauf contributed a wake on LAN CGI script.

+ Notes in contrib/eepro100notes on how to flash the EEPROM.

+ Happy Australia Day, 2000-01-26.

Released as Etherboot 4.4.2

+ hwilmer@gmx.de found a probe error in 3c507.c, shouldn't mask memory
size with 0xffffL, this fails with 64kB memory.  Also changed to use
real-time-clock to timeout on initialisation of i82586 rather than
relying on decrementing a variable (which is CPU speed dependent).

+ James Pearson pointed out that ~ should be escaped as &tilde; in
vendortags.sgml. Also cleaned up the 8-bit characters while I was at it.

+ Gnter Knauf pointed out that if ASK_BOOT is defined, then key input
echos twice. Also, CR doesn't select the default. Problem was that getchar
converts all \r to \n then main.c was testing against \r instead of \n,
and also that getchar() was echoing when it should leave it to the caller.
Deleted code from getchar() and made it a function with no arguments.
Also sent me a new version of mp-form.pl with mp-form.txt. Also suggested
the good idea that the gateway and relay fields not be displayed if they
are not used.

+ Bob Edwards sent in patches to ntulip.c to autodetect 100 Mb media
for 21142/3 NICs.

+ Okuji Yoshinori found a small bug in main.c: for (retry1 = 0; retry < ..
Don't know what effect it had.

+ Klaus Espenlaub sent patches which make the RTL8139 driver more
reliable, see notes in rtl8139.c.

+ Russ Nelson gave permission to release cs89x0 driver with Etherboot.
It's normally under GPL. Moved cs89x0 driver to src directory, removed
cs89x0.sgml.

+ Clarified copyrights of files in distribution.

Released as Etherboot 4.4.3

+ Klaus Espenlaub sent in some patches to overcome the 64K block rollover
problem in tftp.

+ Bob Edwards, Paul Mackerras and Marty Connor worked out fixes to
ntulip.c to handle 21142/3 Tulips properly.

+ Karsten Tinnefeld sent in a Makefile fragment for doc/sgml/Makefile
to generate DVI and PS files using sgml2latex.

+ tulip.* renamed to otulip.*, ntulip.* renamed to tulip.*

+ Added FAQ and Writing an Etherboot Driver sections to README.

+ Woops, if and of swapped in dd command in atnetboot.sgml.

Released as Etherboot 4.4.4

+ FUKUHARA Makoto supplied a patch for 4.4.4 for FreeBSD ELF booting
with large tftp blocks which I missed putting in.

+ Small patch by Marty Connor to tulip.c to handle fast Tulips better.

+ Wrote small loader to load images from .com files.

+ Shusuke Nisiyama contributed a 3c595 (and may work for 3c590) driver.

Released as Etherboot 4.4.5

A major reorganisation by Klaus Espenlaub. Here are his notes, verbatim:

The biggest change is the almost rewritten Makefiles (both in src and in
doc - the one in doc is still reasonably simple).  If you want to get
an overview - basically one Makefile now builds both /16 and /32 ROMs.
This involves quite a few make variables (similarly in Config), but it
should be mostly self-explaining.  I also converted the suffix rules to
pattern rules, because it is the only way to keep the build directories
for /16 and /32 apart.

Generally all /32 code is compiled in bin32, all /16 code is compiled
in bin16, and the code that works for both is in bin.

The names (object files, images, but NOT the final ROMs) used during
the build process for some PCI cards have changed (especially lancepci
and nepci).

I eliminated the driver.a file by adding yet more ugly code to genrules.pl
(no, I'm not very proud of the code I added there - I hate perl, but
at least it works), which now generates all sorts of make rules and
dependency information.  This eliminates the redundant recompile of all
other drivers if you specified a specific target at the make command line.

There are now automatic checks for the generated ROM size (3K safety
margin are used to allow for 1K BIOS EBDA and 2K stack).  This should
avoid most of the cases in which the stack overwrites code and/or data,
causing strange crashes.

On to the real code - I rewrote both the ELF and a.out loading (the
blocksize is now properly handled for all formats), and now most of the
code is identical.  Someone might want to merge the two.  I verified that
both a.out and ELF FreeBSD loading still works by booting the FreeBSD
2.2.8 and 3.4 install kernels (the ELF kernel you gave me didn't work
for whatever reason).  Actually I tested everything except NBI, but that
code is completely unchanged.

I eliminated config.h and created cards.h - just have a look and you
should get the idea.  The PCI/ISA prototype matching hack is now buried
in there.  All PCI drivers (and skel.c) have been converted to the new
include file.

The only file that is really new is nfs.c, which contains the code where
I started my journey.

The loader code is now merged into one file for both the normal and
compressed case.  Some bugs have been fixed along the way (the normal
loader copied too much data, potentially crashing the machine).  BTW: I
removed the block copy from comboot, because the first thing the loader
code does is - to copy the code to the right place.  Also copying it to
0x80000 prevents compressed ROMs (though there is no way in the Makefile
to create such a beast).  Also the stack changing code was wrong -
the overflow handling is missing: 0x9fxxxx-0x800000 is truncated to
fit in the 16 bit register, which is not quite what one would expect.
The stack pointer must be in the same segment, so for overflows just
load 0.  Anyway, we don't need a new stack for 4 bytes (the return
address).  Also the shifts won't work on an 8086/8088.  The ROM segment
and length code is also done once more in loader.S.  Finally the org
0x100 didn't work with my version of as86 - it prepended 256 null bytes.
Don't get me wrong - I see that comboot was just a quick and dirty hack.
It's now working...

I also fixed floppyload to read exactly the right amount of data (not
up to 65520 bytes more like the Linux code did) and ripped out some
unused code.

3c509.c: includes fixed, const added, static added, removed redundant code

3c90x.c: includes fixed, const added, shortened messages to save space,
made the transceiver selection bootrom fix optional (default: off),
updated 3c90x.txt

ansiesc.c: de-tftpified (now uses "download")

config.c: moved the driver prototypes to separate file, added a few const

cs89x0.c: ansified, const added, fixed timeout handling

eepro100.c: includes fixed, const added, static added, USE_INTERNAL_BUFFER
hack, passing globals as a parameter removed, timeout handling fixed,
comments fixed

epic100.c: comments fixed, const added, static added, USE_INTERNAL_BUFFER
hack, removed unused variables, changed debug code to use macro instead
of variable, allow broadcasts to be received (after all how did it
work before???)

etherboot.h: dual/serial console "cleanups", added NFS support, fixed
INTERNAL_BOOTP_DATA for pathologic cases, removed prototype within comment
(how did that end up there?), moved prototypes for config.c code here,
added consts all over the place

floppy.c: added static, changed cleanup code

floppyload.S: renamed from .asm

genrules.pl: I didn't write this :)

i82586.c: fixed includes, added const

lance.c: fixed DMA for PCI cards, removed leftovers from Linux driver,
USE_INTERNAL_BUFFER hack

linux-asm-io.h: added prototypes

linux-asm-string.h: added prototypes

loader.S: renamed from loader.asm, more comments see elsewhere in
this mail

loader.inc: removed

main.c: includes fixed, static and const added, Etherboot/32 now prints
relocaddr, hooks for NFS support, drain the Rx queue in some strategic
places (after sleep), changed interface to await_reply: timeout handling,
new cleanup code

misc.c: ansified, twiddle code changed, strcasecmp enabled for
Etherboot/16 (smaller than the 16 bit libc version), clear A20 before
kernel is started (only for NBI), dual console fixes, removed bogus
comment about interrupt enabling using iskey()

nic.h: include guard added, added const

ns8390.c: fixed includes for PCI version, added const, ansified, renamed
internal functions

osdep.h: removed #include <asm/byteorder.h> - probably all #ifdef
__linux__ and __FreeBSD__ may be removed now.

osloader.c: see elsewhere in this mail. removed loads of debugging code.

otulip.c: removed unused variables, USE_INTERNAL_BUFFER hack, ansified,
added const

otultip.h: added const

pci.c: fixed signed comparison

pci.h: include guard added, const added

rtl8139.c: static added, const added, globals are no longer passes as
parameters, fixed probe loop

serial.S: removed the Makefile hack to calculate the divisor and replace
it by a CPP hack

sk_g16.c: added static/const

skel.c: updated

smc9000.c: fixed includes, added const, fixed timeout, removed unused
variables

start16.S: changed stack pointer reloading

start32.S: cleaned up GAS295 support, fixed CPU detection, fixed memsize
for > 64MB, added basememsize()

tiara.c: fixed includes, added const, removed unused variable

tulip.c: fixed includes, added const, decreased oversized frame tolerance,
USE_INTERNAL_BUFFER hack, converted to C comments, fixed prototypes,
removed unused variable

via-rhine.c: fixed includes, converted to C comments, removed unused
functions, fixed timeout, USE_INTERNAL_BUFFER hack, removed unused
variable, const added

lzhuf.c was modified not to output ':' characters, because I find it
useful to search for them in a make log to spot all error messages.

Finally all symlinks have gone

I fixed sgml/vendortags.sgml (maybe it's more a workaround): sgml2txt
didn't like the ~~' sequence and ignored a whole line, making the .txt
file incomplete.  Also freebsd.txt was converted to sgml.

My standard patch to netboot-0.8.1/configure (to remove the "unknown") is
again in place - how is this file created? By hand?  I added serial/dual
console support to mknbi-linux.

Oh, and something I almost forgot: Etherboot now works under VMware.
VMware has a rather large EBDA (3K instead of 1K like most other
BIOS versions), so one might have to sacrifice a few nifty options.
Note that the Makefile check cannot detect this - maybe we should add a
runtime check, too?  One thing that doesn't work with VMware is the "Boot
Local" stuff - it's not terribly useful if you have to boot from disk,
but it shouldn't crash the machine (I've seen several triple faults,
depending on RELOCADDR and the stack usage).  I tracked it down to the
fact that VMware crashes on int 0x19 if the memory area 0x98000-0x9f3ff
is overwritten.  If one changes RELOCADDR to 0x8f000, everything works
nicely.  Someone should report the bug to VMware.  Also proper boot ROM
support should be added to VMware and they should upgrade the virtual
card to 100Mbps (don't laugh - they limit the bandwidth to 10Mbps even
if you have a real 100Mbps card!).  Maybe one could even use the flash
utility for the recent AMD PCnet cards...

Released as Etherboot 4.5.5

+ More fixes from Klaus Espenlaub, based on bug reports by Doug Ambrisko
and others. Small cleanups in tulip.c. He also added a BACKOFF_LIMIT
parameter to the exponential backoff for retry intervals.

+ Christoph Willing sent in a fix for PCI NE2000s which forces 16-bit
transfer mode for PCI. This might make it work on some PCI NE2000s that
failed before, please test if you can.

+ Moved commercial links off main web page into web page of its own.

+ Merged in TRY_FLOPPY_FIRST code from <as@bart.nl>.

Released as Etherboot 4.5.6

+ Frank Mehnert spotted a bug in tulip.c where the sole transmit buffer
was not tagged as the last one in the chain.

+ Hacked serial.S to check if a serial interface is present and to disable
the routines if not, so that Etherboot does not hang on a machine without
a serial interface, even if the option has been compiled in.

+ Wrote mkromnbi for making a network bootable image from a ROM image
(for using Etherboot to test another Etherboot driver).

+ Krzysztof Halasa found a small bug in nfs.c in the handling of BOOTP
extension files.

+ Some instructions on adding Etherboot to a main BIOS were contributed
by Dirk von Suchodoletz.

+ Some commentary on cbrom.exe versions posted by Rapp Informatik Systeme
GmbH to the Netboot mailing list included.

Released as Etherboot 4.5.7

+ Patrick Auge pointed out that -DT503_SHMEM option should be not used
by default, as the comments say.

+ Heinrich Rebehn pointed out that the documentation needed to be updated
for the new bin32 and bin16 prefixes for targets.

+ Christoph Plattner found that node_addr in struct nic should be unsigned
char * or problems occur in the handling of the MAC address in cs89x0.c.

+ Greg Hudson pointed out that inet_ntoa is misnamed, it should be
inet_aton.

+ Vsevolod Sipakoff suggested that the probe address list for NE2000
should match that of Linux. Done, with the deletion of 0x360 which often
conflicts with the parallel port at 0x378.

+ Created mknbi, a Perl utility that replaces mknbi-dos, mknbi-linux,
mkfreedosnbi, and mkromnbi. Removed mk{freedos,rom,elks}nbi/ from
contrib/.  (ELKS has its own tagged image creation method now.)

+ Removed obsolete/ from contrib/.

Released as Etherboot 4.5.8

+ Some small enhancements to mknbi. Merged in disnbi and mklnim. Changed
version to 1.0.

+ netboot-0.8.1 directories distributed separately now.

Released as Etherboot 4.6.0

+ Nick Lopez discovered that the Davicom 9102 is Yet Another Tulip Clone
and a few appropriate initialisations will make it work.

+ Matthew Reimer sent in patches for contrib/3c90xutil.c/bromutil.c to
compile and run under FreeBSD.

+ Krzysztof Halasa found that the number of buffers in epic100.c needed
to be raised to 2 each for operation at 10 Mb for the PCI EtherPower
9432B-TX using the SMC 83C171.

+ Klaus Espenlaub fixed the root path handling in main.c, it wasn't
requesting option 17 in DHCP. Also removed one prototype warning in
tftp().

+ Daniel Shane sent in code for for adding identifiers to a DHCP request,
using a user specified DHCP option. The patch is in contrib/dhcpid.

+ Bug fix in mknbi-1.0 for the ipaddrs=rom option, warnings about badly
formed specification to ipaddrs= or unresolvable names.

Released as Etherboot 4.6.1

+ Stuart Lynne sent in patches to allow - to mean reuse kernel filename
in menu specifications. Also fixed DHCP request to ask for tags 129 and
130 (Ken added this), as the mknbi man page promised.

+ Fixed URL for Richard Ferri's LUI.

+ Added URL for Gregory R. Warnes' ClusterNFS.

+ Changed RFC2132_MAX_SIZE option in main.c to allow maximum size DHCP
replies.

+ Matthias Schniedermeyer submitted an optional feature to use numbers
instead of letters for bootmenu entries.

+ Brought some FAQs up to date with mknbi-1.0.

+ Wrote perl script mklrpnb for making netboot image from Linux Router
Project floppy.  Tested on Coyote Linux (based on LRP).

Released as Etherboot 4.6.2

+ Re-enabled tulip.c:tulip_disable() code.

+ Minor fix to SHOW_NUMERIC code in bootmenu.c.

+ Minor spelling corrections to vendortags.sgml.

+ andreas.kabel@slac.stanford.edu pointed out that to be RFC951 compliant,
the broadcast BOOTP request near line 705 of main.c should use port
BOOTP_CLIENT and thus stand a better chance of going through firewalls.

+ Christoph Plattner found that the 16 bit bus width fix for NEPCI cards
introduced in 4.5.6 breaks operation for other NEPCI cards. So now the
#define symbol is NS8390_FORCE_16BIT, to be used when the NIC requires
16 bit bus width but Etherboot guesses wrong. This needs to be fixed
later to autodetect adapter bus width.

+ Shusuke Nisiyama sent me the PCI IDs for 3C595 NICs so that ROM images
could be generated for them.

+ Jim McQuillan sent in modifications to tulip.c to reject bad packets.

+ Klaus Espenlaub suggested changing the filename substition macro in the
IDENT16 and IDENT32 strings to $(@F) so that only the filename portion of
the pathname is inserted.  Otherwise the contributed romid program breaks.

+ Added author ident and copyright status to contrib/bin2intelhex.c.simple
at Rogier Wolff's request.

+ Gnter Knauff sent updates to romid and mp-form.

Released as Etherboot 4.6.3

+ Marty Connor contributed patches to tulip.c to support the LinkSys
LNE100TX v4 NICs.

+ Added rmrd.com to mknbi-1.0 directory.

+ Added experimental changes to allow RELOCADDR to be changed to 0x88000
to avoid Disk On Chip drivers. See RELNOTES and Makefile for details.

Released as Etherboot 4.6.4

+ Chris Johns found a long-standing bug in first-linux.S. ES would
sometimes != CS, making the ramdisk moving fail.

+ Made --harddisk work for mknbi-fdos.

Released as Etherboot 4.6.5

+ Bug in mknbi-dos, last sector omitted in ramdisk image.

+ Use ceil() for computing number of rootdir sectors.

+ --harddisk now works for mknbi-dos also.

+ Should work on FAT16 partitions now.

+ New driver for Winbond W89C840 by Igor V. Kovalenko.

+ Modification to loader.S by Steve Smith for some PCI BIOSes that don't
handle INT19H well.

+ Small bug in osloader.c, j not defined in two places if DELIMITERLINES
defined.

Released as Etherboot 4.7.6 (developmental)

+ All the changes from 4.6.5 to 4.6.6 back propagated except for the
loader.S change.

+ Removed -nostdlib from ld flags. Seems to be a vestige from when we
were calling gcc as it's not an ld option.

Released as Etherboot 4.6.6

+ Contributed one-line patch to allow booting from SCSI disks also (sd).

+ Grzegorz Jablonski sent in a fix for long-standing bug in loader.S that
prevented booting on some BIOSes.

+ More updates to documentation.

Released as Etherboot 4.6.7

+ Rick Kennell pointed out that there is another place in main.c where
the DHCP request is sent with a source port of 0, which may not elicit
replies from some DHCP servers or pass gateways. Changed to
BOOTP_CLIENT.

+ Depca driver finally works.

+ Changed unsigned int len; to int len; in 3c595.c transmit routine
which was generating a warning from gcc.

+ Removed warning about ljmp indirect without * in start32.S by putting
in the *. Checked that generated code has not changed by comparing
binaries. Later: But only for gas 2.9.5 and above, gas 2.9.1 doesn't
like *.

+ Wanted to do the same for the lcall's in pci.c but discovered that
gas 2.9.1 chokes on *, and as I don't want to introduce more #ifdefs
into pci.c, we'll just live with the warnings until gas 2.9.1 is dead
or the incorrect syntax is rejected by a current gas release.

+ Reversed #define so that GAS291 indicates gas 2.9.1. Rationale: gas
2.10 probably behaves like gas 2.9.5 so the default should be the new
behaviour.

+ NI5010 driver finally works. Weird NIC, has only 2kB memory on board
so if you don't switch to the receiver immediately after transmitting
you will lose the reply packet. (This is a very ancient board, I was
doing the driver for kicks.)

+ Implemented a low-overhead timer routine for implementing timeouts in
drivers. This timer should be used instead of hacks with integer loops
(which are CPU speed dependent) or calling currticks(), which has
relatively high overhead because it reads the BIOS timer, and has a
resolution of only 1/18th of a second.  Timer 2 of the 8254 timer chip
can be loaded with an unsigned 16-bit value that will be decremented at
about 1193 counts per ms (constant TICKS_PER_MS in timer.h). So the
maximum timeout that can be implemented with a single call is about 54
ms. Call load_timer(u16) to load and start the timer (0 == 65536). Call
int timer2_running() to check if it is still active. timer2_running can
be interleaved with other tests, e.g. on the NIC hardware registers.
When the timer runs down to 0, it will return 0. If you just need to
delay a short time, call waiton_timer2(u16) to load and delay that many
ticks.

+ Removed slowdownio() from sources. It's called in only one place and
is followed by a call to currticks() which will result in a large delay
anyway. It's also not clear if the strategy in slowdownio (a couple of
local jumps) will even work on fast CPUs with instruction caches.

+ Removed support for %i (alias for %d) in printf. Nobody uses it, if
they do they should just edit the format string to use %d, and removing
it means one less non-standard format in printf; we have %X, %b and %I
as it is.

+ Simplified currticks() by accessing BIOS variables directly instead of
calling timeofday BIOS interrupt.

Released as Etherboot 4.6.8

+ I'm pissed off with the inconsistency between gas 2.9.1 and gas 2.9.5
re ljmp * (indirect jump). Some assemblers claiming to be 2.9.5 dislike
the * (RH6.2's is one). I'm going to just make it ljmp and live with the
warning message until the gas 2.9.1 syntax is totally dead.

+ Various fixes from Klaus Espenlaub: Fix for NFS booting from *BSD
platforms, fix for Lance driver, interrupt was being turned on, fix for
serial.S and code cleanups.

+ Paul Robertson of Locsoft found that if the top half of %esp contains
garbage just before the first call to real_to_prot in start32.S, it can
affect booting on some BIOSes. Inserted andl to clear top half.

+ Sleep for 2 seconds after failure to load file to avoid pounding the
server if the file does not exist and thus shutting down the tftp
service (inetd has rather poor rate controls).

+ Shredda of gmx.de reported a discrepancy between the Macronix MX98715
device IDs used in pci.h and NIC. Marty Connor has ruled that the id in
NIC is a typo.  If booting works from floppy but not from ROM, this may
be the reason.

Released as Etherboot 4.6.9

+ Oops, a brown paper bag bug in start32.S. Forgot that the currticks()
needs interrupts enabled to work. Calling the BIOS allowed interrupts to
happen by going into real mode (interrupts are disabled in protected
mode). So inserted prot_to_real and real_to_prot pair inside
currticks().

Released as Etherboot 4.6.10

+ Marty Connor and Gary Byers of thinguin.org contributed a LILO prefix
file that makes the Etherboot image look like a Linux kernel to LILO and
thus bootable from the LILO prompt.

+ Ifdefed out code in mknbi that falls back to the tftp directory for
root-path. The result is that the options passed to the kernel specifies
the tftp directory as nfsroot even if there is no root-path option in
the DHCP or BOOT server configuration. This interacts badly with recent
Linux kernels which actually pay attention to the root-path handed down
and use that as nfsroot. People who were expecting the diskless kernel
to use the default /tftpboot/<name of client> as nfsroot would have been
surprised to find it trying to use the tftp directory (which often
happens to be /tftpboot, although I recommend that it be different, to
improve security).

+ Eric Biederman found an off-by-one bug in the ELF loader.

+ Christoph Plattner pointed out that the comments for COMCONSOLE in
Config are wrong, the serial port is at 3F8, not 378.

+ A couple of updates to wol.c and wake.pl by Gnter Knauff.

+ The Davicom 9009 and 9102 driver was contributed by Sten Wang of
Davicom. It will be merged into tulip.c later.

+ Implement eepro100_disable. There have been reports of memory
corruption after Etherboot has handed over to the booted image due to
the live controller.

+ Matt Hortman reported a new Intel NIC, 82559ER, with different IDs.
Added to file NIC.

+ Ranjan Parthasarathy reported another Tulip clone from ST Micro with
different IDs.  Added to file NIC.

Released as Etherboot 4.6.11

+ Stefan Lesicnik sent in a report from Intel which explains what is
probably wrong with the PnP header. BEV should point to the mainline
vector, not to the part that installs the INT19H vector.

+ Hopefully finally fixed DHCP option limit problem. It was requesting
large packets but not parsing them fully due to the length limit passed
into decode_rfc1533(). Thanks to shredda for testing this.

+ Added the # modifier to printf, it prefixes 0x to %x and %X making
printf formats shorter throughout. 0x%[xX] changed to %#[xX] in lots of
files.  Now if only I could make b,x,X the standard hhx,hx,x and get rid
of I.

+ Andreas Neuhaus provided patches for multiple rx buffers for lance.c
which made it work again with VMware.

+ Perl script to convert floppyfw floppies to netbootable images.

+ Marty Connor made some small changes to liloprefix and Makefile to
make the LILO images SYSLINUX bootable also.

+ Make the default return value for _poll in skel.c 0 so that when
driver writers implement _transmit first, it will not hang on garbage
return values from _poll when it's called to flush the input queue
before the first transmit.

+ EEPRO/10 driver now works. s timer routines came in useful.

+ Anders Larsen sent in a patch to 3c90x.txt which makes it clearer.

+ Added more stuff to the documentation.

+ Fixed bug in mknbi that always did the equivalent of --ipaddrs=rom
no matter what. Also removed undef from my variable list in TruncFD.pm
so that it won't have problems with Perl interpreters.

Released as Etherboot 4.6.12

+ Jim McQuillan sent in a patch for first-linux.S where it was assuming
the argument in tag 129 (additional parameters) is a null terminated
string, when it's a length counted string. A new routine, addkarg was
created to handle this.

+ eepro100 should handle newer NICs with 256 byte EEPROMs now. This
includes the onboard NICs on some motherboards, see file NIC. Thanks to
Stephan Lauffer for helping with the fixes. WARNING: This code may have
a bug that causes the onboard EEPROM to be corrupted. We believe we have
found and removed the bug but please proceed with care.

+ DHCPDISCOVER was sending out one byte too many for PARAM_LIST.

+ In DHCPDISCOVER send "Etherboot" in VENDOR_CLASS_ID option (60).  Will
add code later to check for "Etherboot" in vendor encapsulated options.

+ Used s timer routines in 3c509 for more accurate timing and hence
better hardware detection. Use COMMAND_IN_PROGRESS bit to detect
transmit complete instead of waiting for a fixed amount of time. Get rid
of eth_vendor and associated tests, it doesn't serve any useful purpose
since the driver was modularised long ago and the 3c509 detection status
is stored outside of the driver now. Got rid of some unused global
statics in 3c509.c, leftovers when the drivers were monolithic. Wait 2
seconds after enabling TP interface to give it time to come up. This
allows us to get rid of T509HACK in main.c.

+ Get rid of eth_vendor and associated tests in cs89x0.c, same reasons
as for 3c509.c.

+ Moved the rest of the VENDOR_ and FLAG_ defines into ns8390.h, as
ns8390.c is the only file that uses them now.

+ Use lower 32 bits of node address + current time for xid (network byte
order). More likely to be distinct from other clients than just the
current time, which is similar for all clients booted at about the same
time.

+ Support for 16-bit code has been removed.  XTs and ATs are pretty much
dead now and in fact many Etherboot/16 drivers have been broken for a
while but nobody noticed. This should make some of the code easier to
maintain.  If you really wanted 16 bit support, use an older version of
Etherboot, maybe 4.4 or something like that, not sure when things
started breaking for 16 bit mode.

+ Should not store IP and UDP headers at BOOTP_DATA_ADDRESS. Redefined
bootp_t without IP and UDP headers. Now requested size of bootp packet
matches storage available. Do not add sizeof(iphdr) + sizeof(udphdr) to
bootp pointer in start32.S:xstart now. start16.S:xstart was broken
because it did not do this addition but nobody noticed.

+ Removed array kernel_buf, saving 128 bytes and replaced with
KERNEL_BUF, a pointer into the bp_file of the bootp_reply structure at
BOOTP_DATA_ADDR. Note: this depends on the server not sending Option
Overload which would use the sname and file fields for options, but we
don't request this option so it shouldn't.  Removed char *kernel,
instead check KERNEL_BUF[0] just before booting and if null, use
fallback filename.  (This is needed for future extensions to booting
protocol.)

+ Define a shorter tftpreq_t type for making requests instead of using a
full sized tftp_t packet to reduce stack usage.

+ Defined macros for htonl/htons/ntohl/ntohs for cases where the operand
is a constant, saving a function call.

+ Started on first32.c, a protected mode 32-bit version of
first-linux.S, which should be far easier to read and maintain. Will
boot basic Linux kernels correctly but doesn't handle ramdisk or
kernel arguments yet.

+ Updated nfs-swap documentation in contrib/nfs-swap to point to
Claus-Justus Heine's new web page.

Released as Etherboot 4.7.13 (development)

+ Added more IDs for eepro100 variants taken from the Linux 2.2.18
source. Should handle the EEPROM properly now, a few defines were wrong
in 4.7.13.  Loop counter timeouts in eepro100.c replaced with hardware
timeouts.  Don't loop waiting for packet in _poll, return 0 immediately.

+ first32.linux works. Does kernel arguments and ramdisk but doesn't do
appended parameters from menu selections, which should be replaced by a
more elegant menu scheme anyway.  Needed gateA20 routines in
mknbi-1.1/first32.c otherwise cannot access extended memory.
first32.linux should be able to handle memory > 64 MB, which the old one
couldn't.  Support for first32pm call protocol added.

+ 3c595.c changed to use hardware timer for delays. Transmit routine
waits for a fixed period after transmitting. Changed to check
S_COMMAND_IN_PROGRESS bit. It also contains some of the same unused
variables as 3c509.c and mentions 3c509 in some comments. Cleaned up.

+ Hmm, how come this wasn't fixed long ago? Should discard BOOTP/DHCP
replies that are not to broadcast or own MAC address. I guess xid caught
practically all of the non-matching packets. (Later: Actually the NIC
should filter out packets we don't want, but leave test in anyway, it's
the last test.)

+ Removed last vestiges of ETHERBOOT32 and ETHERBOOT16.

+ ETHER_ADDR_SIZE => ETH_ALEN, ETHER_HDR_SIZE => ETH_HLEN,
ETHER_MIN_PACKET => ETH_ZLEN, ETHER_MAX_PACKET => ETH_FRAME_LEN.  More
Linuxy and therefore more familiar to programmers.

+ Cleared up confusion with 60/64 and 1514/1518 for minimum and maximum
frame sizes. Practically always the right numbers are 60/1514, except
that some chips count the FCS in the receive length, then we have to use
64/1518.

+ Make __swap32 and __swap16 inline routines available globally as
swap32 and swap16. eepro.c can use swap16 instead of making up one.

+ Make aui field in nic.h an int since it will be padded to a longword
boundary anyway and call it flags so that other drivers can use it for
their own purposes. Currently only 3c503 uses it to indicate AUI xcvr.

+ Make sprintf return number of characters written instead of a pointer
to the last char written to be more consistent with standard C.

Released as Etherboot 4.7.14 (development)

+ Thanks to Mark VandeWettering for the start of HomePNA (networking
over phone lines) support for the AMD 79C978.

+ Bug fixed in first32.c handling of (ip|nfsroot)=X where X is not rom.

+ first32pm.linux works. No need to go into real mode to call first32pm
and then it goes back to protected mode. Paves the way for extension
routines to Etherboot.  Implement program returns to loader flag in
header field.  Added option to mknbi to specify this.

+ first32*.linux: Check tag 128 present and correct before appending tag
129. Also tag 129 should be appended to parameters before substitutions.

+ Merged cleanup_net into cleanup since they are always called together.

+ Floppy booting doesn't need to be passed BOOTP_DATA_ADDR.

+ Clean up variables associated with tagged image loading in osloader.c.

Released as Etherboot 4.7.15 (development)

+ Duplicated 3c900 PCI IDs under 3c595 as some NICs apparently detect
and work with the 3c595 driver but not the 3c90x driver, according to a
report from Dirk Pfau. (The 3c90x series has two modes of operation,
programmed I/O mode descended from the 3c509, good only for 10 Mb
operation, and bus mastering mode, essential for 100 Mb operation. For
network booting, either mode is acceptable.)

+ Removed auto from kernel parameters; it's the default already.

+ Use hardware timer instead of loop counter for transmit timeout in
3c90x.c.

+ Define a jmpbuf type for setjmp and longjmp. Trim size to 7 longs,
that's all that's needed. Standardise the return values from longjmp:
-2: loader error, -1: timeout or ESC, 0...: various meanings to
Etherboot main loader.

+ Ansify function headers in bootmenu.c.

+ Make _int10 return ax | (bx << 16) as result so that these can be
accessed more efficiently in the following statements.

+ Make handleansi take unsigned int instead of unsigned char as
argument, otherwise extra code will be generated to handle this
according to ANSI rules.  (Quite significant saving of 55 bytes.) Rename
it ansi_putc for clarity.

+ Prefix getc, putc and ischar with console_ to make things clearer and
to avoid confusion with the Unix getc, putc.

+ Add menu as a target to mknbi. Started source code for menu extension.
Successfully transferred control to menu at 0x10000 and back. ANSI
colour controls work, at least. Return end needs more work.

+ Started on ELF support in mknbi.

Released as Etherboot 4.7.16 (development)

+ Added atftp 0.2 (ftp://ftp.mamalinux.com/pub/atftp/) to contrib/.
Supposedly contains a tftpd that runs multithreaded, which may help
people having problems with *inetd shutting down tftpds that spawn too
fast.

+ Added a few more Tulip entries to config.c and NIC, not all of them
have been confirmed working.

+ Got ELF format creation in mkelf-linux working now. At least one empty
section header is required to make a valid ELF file.

+ Added code to support non-MULTIBOOT ELF when IMAGE_ELF is selected but
IMAGE_MULTIBOOT is not.  Booting from images created by mkelf-linux
now works!

+ TAGGED_IMAGE is now not always selected. It's just the fallback if
none of TAGGED_IMAGE, AOUT_IMAGE or ELF_IMAGE is selected. Therefore you
must explicitly select TAGGED_IMAGE if you want it, and you have
selected AOUT_IMAGE or ELF_IMAGE.  Startup banner line displays all
image formats accepted.

+ exit() in mknbi/start32.S should copy argument to %eax first.

+ Images with 0xAA55 in bytes 510-511 are no longer accepted, which
should reject invalid formats now, e.g. Linux kernel images, which have
a boot sector in the first block. Strictly this does not conform to the
original netboot spec by Jamie Honan, which specifies that non-tagged,
linear images starting from 0x10000 are allowed, but that format is
pretty useless now. Any decent loading scheme needs a roadmap to the
blocks in the downloaded file, which is what tagged, a.out, or ELF
images provide in the header. Q: What is config_buffer in main.c for?
Nothing else seems to use it. Is it a relic of non-tagged images?

+ Clean up lots of obsolete prototypes in etherboot.h. Ansify lots of
function headers in main.c. Make lots of functions and variables in
main.c static.  Make bootmenu.c:getoptvalue() static.

+ Simple external menu program works!

+ More documentation cleanup, notably editing the compile options to
match what has been done.

Released as Etherboot 4.7.17 (development)

+ Marty Connor did it again! He found a long standing bug in the PnP ROM
header which caused it not to be recognised by BIOSes. Now Etherboot ROM
images are PnP compliant. I hacked makerom.c to fill in the offset of
the ident string for the device string so that the BIOS can even print
out "Etherboot" and the device ident on boot up.

+ Made demo menu program a bit more elaborate with timeout.

+ Igor V. Kovalenko fixed the Winbond W89C840 driver to use the hardware
timer instead of CPU counter loops. Now none of the Etherboot drivers
rely on CPU speed dependent loops.

+ Small fix to contrib/p910nd.c to not use getprotobyname which requires
libnss_files.so, which may not be installed in all environments.

+ Small change to contrib/mkffwnb to allow the user to choose ELF format
(by editing one line) instead of tagged format.

+ Patch to atftp-0.2 to make it accept a filename if the directory is a
prefix of the filename. This is needed so that valid absolute pathnames
will work. Patch sent to atftp authors.

+ Added support for filtering out replies which do not include a Vendor
Class Identifier of "Etherboot" in the Vendor Encapsulated Options. This
can be used to select only the DHCP servers which we want to get
addresses from and reject the rest.  Select the compile option
-DREQUIRE_VCI_ETHERBOOT. This requires ISC DHCPD 2 or 3 AFAIK.  (It's
not documented in DHCPD 2, but it works.) Other DHCP servers may support
VEO. (It's a RFC2132 option.)

+ Jim Thomas suggested a way, other than creating /fastboot, of
preventing fsck from running on NFS root, ln -s /bin/true
/sbin/fsck.nfs. Not verified but should work. Added to documentation.

Released as Etherboot 4.7.18 (development)

+ Pavel Tkatchouk verifies that lance.c can handle PCnet-FAST III
79c973. Added a new entry to lance.c and NIC. NIC entry not verified
yet.

+ Enhanced disnbi to decode ELF images too.

+ Arrgh! There are old BIOSes that rely on the wrong order of the bytes
in the device identifier in the PCIR and PnP structures. Wrote a Perl
program swapdevids.pl to swap these bytes. Apply this to image file just
before programming the EPROM.

+ Marty Connor suggested that the specs state that for PnP ROMs the
unsuccessful return from boot should be int 0x18 rather than int 0x19.
Using int 0x19, selecting L for local device doesn't work. Fixes needed
in both loader.S and start32.S (get lret to work properly, instead of
doing an int 0x19 directly, involved saving ss and sp in real mode
instead of in protected mode).

+ Wrote a Perl program disrom.pl to display key structures of a ROM
image.

+ Added call to binmode() in various Perl utilities so they should work
under other OSes.

+ Added check in makerom.c to warn if 55 AA not found at start of image.
It seems some people don't read the warning not to use the Linux
supplied as86.

+ It seems Z is a recent addition to pack/unpack formats in Perl and
even a Perl as recent as 5.004 doesn't implement it. Change Z5 in
mknbi.pl and TruncFD.pm to a5 since we only need to compare it against
'FAT12' and 'FAT16'.

+ Some errors found in osloader.c in the #ifdef IMAGE_MULTIBOOT
sections.  kernel should be KERNEL_BUF and union info should have
unsigned short s[256];. Also kernel -> KERNEL_BUF for IMAGE_FREEBSD,
obviously few FREEBSD users have tried compiling it.

+ Explain in docs that .com and .(lz)lilo images can be generated and
touch briefly on how to use them.

+ Donald Christensen contributed translations of floppyload.S and
loader.S to gas syntax. Currently they are in contrib/gassyntax/.  They
potentially allow us to throw away as86 and/or nasm and use GNU tools
throughout, but I have to do some work on them: 1. I have to check what
versions of gas accept the syntax, the 16-bit mode in gas was a recent
enhancement; 2. I have to put back the #ifdef PCI_PNP_HEADER into
loader.S and also bring it up to date to the recent patches.
floppyload.S should be usable as is.

Released as Etherboot 4.7.19 (development)

+ Donald Christensen completed the translation of loader.S and all the
other .S files in the src directory that previously required as86 or
nasm. No more futzing around with precompiled versions. Yipee!

+ Luigi Rizzo contributed a slightly hacked FreeBSD loader that works on
floppy or hard disk. See boot1a.s for details. Makefile rules edited.
Targets renamed .dsk and .lzdsk to indicate they're not floppy specific
now. Documentation updated. I note that the loader is smart enough to
figure out exactly how many sectors have to be read.

+ Renamed comload.S to comprefix.S which describes it better.

+ mknbi tools split out into separate package for independent
development.

+ Use A32 instead of Z32 in unpack format in disrom.pl in case we
encounter old Perl versions (< 5.005).

+ Vendor Class Identifier string that's sent out is now of the form
Etherboot-x.y (13 bytes long).

+ Reduced size of ee_data in davicom.c to 32 bytes because we only need
to access the MAC address in bytes 20:25.

+ Changed type of formal arguments to pci.[ch] routines to unsigned int
except for the last argument, because in ANSI C parameter passing is
like assignment and extra dummy variables and code may need to be
generated if the formal argument type is not the the same size as the
actual argument that gets pushed on the stack. Not insignificant
overhead, reduction from 1853 to 1727 bytes due to change. Changed devfn
and bus members of struct pci_device to unsigned char to enforce limit
on type. Surprisingly this reduced the size further to 1667 bytes.
Probably the compilier could do more optimisations after the last
change.

+ Some drivers hardwired 0 for bus number in calls to pcibios_*
functions. Changed to pci->bus, which is set in scan_bus. Only
people with NICs not on bus 0 would have noticed.

+ Found a couple more old-style pre-ANSI C function declarations. Turned
on -ansi and -pedantic for kicks and fixed some non-ANSIness, e.g.
// comment in #define, text after #endif, casting memcpy arguments to
void *, using void * instead of char *. Some char declarations changed
to unsigned char.

Released as Etherboot 4.7.20 (development)

+ I changed my mind. I think the format in disrom.pl should be Z32,
otherwise it displays binary characters after the valid part of the
string. For people with Perl < 5.005, please upgrade.

+ Moved RELOC down to 0x94000. Turned on USE_INTERNAL_BUFFER by default.
Unless BOOTP_DATA_AT_0x93C00 is defined, use internal bootp strucutre.
This gets Etherboot out of the area just below 0x10000 and the area from
0x93C00-0x93FFF. DHCP packets can now be as large as the Ethernet
payload. Later on, first32.c and the parameter area could expand a bit.
In lance.c had to reduce RX_RING_SIZE (by reducing LANCE_LOG_RX_BUFFERS)
from 16 to 4 to fit the driver. I'd like to hear feedback from lance
users. Is 24kB of receive buffers really needed given that Etherboot
uses stop-wait protocols?

+ Doug Ambrisko sent in some patches for FreeBSD: a correction for one
typo I missed and changes to make compilation under FreeBSD easier.

+ Michael Sinz contributed patches to allow the FreeBSD loading code to
load debugging symbols also.

+ Marty Connor contributed a SiS900 driver. Also pointed out that if
ASK_BOOT <= 0, the prompt and read shouldn't even happen. Conditional
code reworked.

+ Split documentation into user manual and developer manual. Added
material.

Released as Etherboot 4.7.21 (development)

+ Preston Wilson pointed out that ds.internic.net should be replaced by
www.ietf.org in documentation URLs. Also RFC1090 should be RFC1094
(NFS).

+ Marty Connor pointed out some broken links due to the split in
documentation. Also comments don't match code in lance.c for change just
above. Also passed on a bug found by a user of rom-o-matic.net where
tftp was not defined when ANSIESC was defined. My mistake, I made it a
static function when it needs to be exported to ansiesc.c.

+ Christoph Plattner found that the Etherboot startup messes up the
flags while testing for a 386+. This doesn't affect Etherboot but it can
cause the loaded operating system to crash. Fix: pushf before and popf
after the CPU model testing.

+ Jean-Jacques Michel contributed patches to the via-rhine.c driver to
make it work for the VT6102 model as used on some DFE530-TX Rev.A3 NICs.

+ Luigi Rizzo sent in a fixed boot1a.s that actually works on HDs now.

+ Stefan Furtmayr sent in a list of URLs to TFTP servers for NT. Added
to directory contrib/Diskless-From-NT.

+ Paul Whittaker contributed a HOWTO in HTML format on booting with NT
as the server.  Added to directory contrib/Diskless-From-NT.

+ CJ pointed out that the advertised message size is > 1500 bytes.
Define MAX_BOOTP_EXTLEN so that the size of the structure matches the
Ethernet payload size.

Released as Etherboot 4.7.22 (development)

+ Reintroduce the old Via-Rhine PCI IDs into file NIC and give the
corresponding ROM images -old suffixes, in case some people are using
the old chips.

+ Peter Kgel contributed patches to the SiS900 driver to make it work
for the SiS630e and SiS730s.

+ Charles Dobson pointed out that when booting from a PnP BIOS, the code
should not hook or restore the INT19H entry point.  Put hooking code
inside #ifndef PCI_PNP_HEADER and changed entry point in PnP structure
from start19h to blockmove.

+ I have one report that gas 2.91 doesn't assemble loader.S because it
can't handle all the 16-bit code. Since the benefits of using gas for
all assembly code are great, and gas 2.95 has been out there for a long
time, I regretfully say to affected users: upgrade your tools. Otherwise
there may be a workaround by specifying the opcodes literally.  I leave
it to interested parties to try. If you do this, put in an #ifdef
GAS291, naturally.

+ Updated contrib/tftp-hpa to 0.16 from ftp.kernel.org.

Released as Etherboot 4.7.23 (development)

+ Rename nepci entry in file NIC to rtl8029 to avoid giving the
impression that nepci will work for all PCI NE2000 clones. Make the
issue of PCI IDs in ROMs clearer in documentation, both in the
configuration and troubleshooting sections.

+ Tania Oka and Hyun-Joon Cha at about the same time found that
implementing the rtl_disable() routine in rtl8139.c stopped random
crashes in Linux later. It is important to disable the NIC after network
loading.

+ Implemented _disable() routine in w89c840, 3c90x and via-rhine drivers
too. Don't know how to do it for epic100.

+ p910nd-0.4 in contrib/ has -f device option now to specify other
printer ports, e.g.  USB.

+ Robb Main found a bug with the #ifdef logic in loader.S. This may fix
problems with BIOS detection.

Released as Etherboot 4.7.24 (development)

+ Paolo Marini sent in some code to make it work on his bare metal (no
standard BIOS or peripherals) platform. This may be a useful starting
point for some applications. See contrib/baremetal/.

+ Eric W. Biederman contributed a Perl script in
contrib/award_plugin_roms/ to list flash BIOS plugin components.

+ Marty Connor rewrote the Tulip driver to handle many more variants.

+ For PCI ROMs loader.S can now detect if it's being called from a PnP
BIOS and choose to hook INT19H if not.

+ Pass struct *rom_info in priv_data to probe routine. This is to allow
drivers to decide, based on the ROM address, which one of multiple
instances of identical network adaptors to activate. Started on 3c509
code to use this but need to understand 3c509 contention resolution
mechanism first.

+ Link src/lzhuf.c to contrib/compressor/lzhuf.c so that we don't need
to refer to contrib/compressor/lzhuf.c anymore in Makefile.

Released as Etherboot 4.7.25 (development)

+ Minor documentation edits, merged in NIC entries for newly supported
Tulip variants from Marty Connor.

+ One last minute change, Robb Main suggested calling cs89x0_reset()
from cs89x0_disable() to shutdown the hardware cleanly.

+ Updated tftp-hpa to 0.17.

+ Slight mod to lzhuf.c to make compression statistics report shorter.

Released as Etherboot 5.0.0 (production)

+ Donald Christensen found a small bug in osloader.c. Not all context
was cleared on tftp restart which caused restarted tftp loads to fail.

+ Correct a small error in setting %sp when not running at 0x9xxxx.
Now relocation to 0x84000 works.

+ Marty Connor added a generic Tulip entry and renamed the Macronix
entries because PHP doesn't like strings starting with digits (for
rom-o-matic.net).

+ In loader.S, move code to save ROM segment and length to before jump
to new segment, otherwise if MOVEROM is defined, then the ROM segment is
always 0x8000. In etherboot.h, define an inline function to say if a ROM
address is ok to boot from. Allow if < 0xC0000 or matches assigned ROM
address of NIC.

+ Thomas Kessler found a bug in vendortags.sgml re option-NNN tags in
dhcpd.conf, the wrong syntax was presented. However on trying the
option-NNN syntax documented in the dhcp-options man page, it was
discovered that option option-NNN is no longer supported in the old way
in recent versions of ISC dhcpd v3; a new syntax should be used. Added
note to vendortags.sgml to warn users.

+ Incorporated changes suggested by Hannu Martikka to #define
DEFAULT_KERNELPATH in etherboot.h for rarp(), and display the TFTP
server address before filename in Loading: message.

+ Split off documentation into separate package in anticipation of
production/development series split. Moved previous LOG to top level.
Moved distribution section of index.html into separate web page so that
index.html will be less ephemeral.

Released as Etherboot 5.0.1 (production)
