Etherboot 4.6.12

A fix to BIOSes that are strict about using the boot entry vector in
preference to the INT19H vector.

Finally fixed DHCP option limit I hope. It was requesting large
packets but the parsing was limited by the constant in the call to
decode_rfc1533().

There is a multiple rx buffer fix for lance.c which may fix problems
people might be having with it; it's reported to work well now under
WMware (simulated Lance).

There is an EEPRO/10 (ISA adaptor) driver now.

.lilo images can now be booted with Syslinux. It would be interesting if
someone would try booting it from loadlin.

There's a contributed script to convert floppyfw (www.zelow.no/floppyfw)
floppies to tagged images.

Various documentation additions and changes.

There are a couple of small bug fixes to mknbi, one for compilation
problems on some old Perl 5 interpreters, the other for a bug that
prevented it from letting the kernel do IP autoconfig.

4.6.11 contains a prefix file that turns an Etherboot image into an
image that LILO is happy to register as a bootable "Linux" kernel,
courtesy of Marty Connor and Gary Byers of thinguin.org.

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

If you have problems with the kernel trying to mount the tftp directory
(this can be different from the NFS root directory), then you might need
the patch to mknbi-1.0/first-linux.S.

There is a possible fix for a bug affecting EEPRO100s. The controller is
not turned off before handing over control to the image as
eepro100_disable is empty. Under heavy traffic, the EEPRO100 can pick up
stray packets which then corrupt the memory area used by the kernel. The
fix was to implement eepro100_disable, using the command that the Linux
driver uses, PartialReset.

Another couple of entries for the NIC file: a ST Micro made Tulip clone,
stmicro (anybody have a better suggestion?), and 82559er, a new Intel
chip (again, anybody have a better suggestion?).

Patch for mknbi to stop using the tftp root directory as the fallback
rootpath. IMO it should use the rootpath option or nothing.

4.6.10: Brown paper bag bug found in start32.S. Clock tick interrupts
not enabled by going into real mode now and then. Please do not use
4.6.8 or 4.6.9, some drivers will not work.

The major changes in 4.6.9 are bug fixes supplied by Klaus Espenlaub for
NFS booting with *BSD, the Lance driver and the serial driver. There is
a correction for a typo in the device ID of the Macronix MX98715 which
could have created non-functional ROMs.

The major change in 4.6.8 is a fix for a bug in the DHCPREQUEST code
that sent a DHCPREQUEST packet with a UDP source port of 0. Some DHCP
servers and some gateways reject or ignore such packets.

The major change in 4.6.7 is a fix for a long-standing bug in loader.S that
probably prevented booting on some BIOSes.

The major change in 4.6.6 is a new driver for Winbond W89C840 by Igor V.
Kovalenko.

A long-standing bug in first-linux.S was discovered by Chris Johns.
Sometimes ES != CS and this caused the INT15 doing the ramdisk moving to
fail. You can upgrade mknbi separately instead of fetching the whole
Etherboot distribution.

4.6.4 has experimental changes to support running at 0x88000
instead of 0x98000 (RELOCADDR) to avoid Disk On Chip drivers. You have
to edit Makefile to change RELOCADDR and also to select an assembler
because the loaders have to be rebuilt. Remember that if you choose a
RELOCADDR of 0x88000, you can only load images up to the 512kB mark or
above the 1MB mark.  Let me know how it goes.

It seems that forcing 16 bit bus width for NEPCI causes Etherboot
to fail on the RTL8029. Therefore the #define has been changed to
NS8390_FORCE_16BIT and should only be turned on for those NEPCI NICs
that falsely detect 8 bit bus width when it should be 16. The bus width
detection code needs a rewrite sometime.

Various mknbi tools in contrib and netboot-0.8.1 have been merged into
one tool, mknbi. The tools from netboot-0.8.1 are now distributed
separately.  mknbi handles ROM, Linux, FreeDOS, and DOS images. It
should be reasonably straightforward to support other images.  mkelksnbi
is obsolete, as ELKS has a Makefile rule for making a tagged image.  If
anybody uses mknbi-blkdev, please let me know, as I would prefer not to
have to port it to mknbi, if I can avoid the work. :-)

There is an unconfirmed report that the Lance PCI driver requires
-DUSE_INTERNAL_BUFFER to work on real NICs (as opposed to the one
simulated in VMware).

The Lance driver has been fixed to work with VMware.  The change should
only affect PCI variants of the Lance chip.  Please test it if you can
(both with real and virtual Lance PCI cards).  Note that Etherboot crashes
if you try to use any of the "Boot Local" functions (which basically just
do a int 0x19) in a VMware session.  This seems to be a VMware bug, as we
don't clobber any memory claimed by the BIOS.  A possible, but not
generally recommended workaround is to use a RELOCADDR of 0x88000.