ipxe/contrib/baremetal/marini.txt

From: "Paolo Marini" <paolom@prisma-eng.it>
Subject: Etherboot on bare metal
Date: Tue, 10 Apr 2001 23:19:19 +0200
Organization: Prisma Engineering srl

Hi Ken,
I have ported Etherboot on an embedded, biosless platform and would like
to contribute the code.

Essentially, the hardware I was running Etherboot is a Pentium based
embedded system, with an Intel Chipset, *but* without serial, VGA,
keyboard etc., only an 82559 Intel (custom) Ethernet controller (I debug
it with the etheral Ethernet packet analyser and an emulator).

What I did was:

  a.. integrate the init.s file within the firmware, with GDT
(re)initialisation (a simple and single entry point taking control of
the boot process)
  b.. provide some stupid BIOS stubs in order to let the OS boot and
still belive that an INT10 call goes to the BIOS
  c.. provide some basic functions to Etherboot, like timer (I used the
Pentium TSC internal counter)
  d.. hardwire in the code information about the RAM size
The BIOS stubs are enough to boot Linux, pSOS and QNX with bootp. QNX is
somewhat difficult to load, because the i82559 driver tries to find the
component using the BIOS32 calls, so I had to patch it.

what i I got from the original firmware is the PCI initialisation and
resource (I/O, interrupts, memory) allocation.

I send you what I changed, that is, the initialisation code and the
misc.c file containing the timer, and the makefile (I don't remember
exactly the options I used to compile all).

Of course, it is only a good starting point for anyone wanting to
implement a bootp client on a biosless platform; some integration work
still needs to be done.

Ciao
Paolo

And in a subsequent email:

I worked with version 4.6.12, but the real modifications involve the
init.S file, which I think is quite sstable between releases.  I forgot
to say that my entry point (symbol _start in init.s) assumes the
processor is already in protected mode.

[The only difference between main.c and misc.c from those in Etherboot
4.6.12 seems to be the deletion of eth_reset(). This may be of use to
others trying to make these changes work on more recent releases. Ken]