Re: [Etherboot-developers] mknbi-1.2-7rc2 released

[Rohit J. <ro...@pr...>]

Greetings,

On Wed, Feb 27, 2002 at 11:20:53AM +1100, Ken Yap wrote:
> >Heh, you just got patches to do some of that in the context of the
> >PXE patches to emulate PXE enough for the FreeBSD PXE boot loader to
> >work!  Maybe grub could be re-worked to access the hardware via PXE and
> >not just look like a PXE image.  Randoms thoughts are what if
> >Etherboot had a PXE network driver writen in terms the PXE stuff
> >that was sent in?  That would seem to get us most of the way there.
> >"pxegrub" could use the Etherboot "PXE network driver" to talk
> >to the Etherboot PXE emulation layer.
> 
> At first glance I don't think it can be modified to do pxelinux. The
> FreeBSD patches just do "PXE-style" booting, which means a small
> secondary loader is loaded starting at 0x7c00 and this loads the main
> image. There is no requirement for it to go through the "official" PXE
> interrupts, in fact this one just does a longjmp to the Etherboot
> routines. Now that's not necessarily a bad thing, but it means you can't
> take pxelinux and use that. So either:

Actually the PXE API entry point is a 'C' function defined in main.c.

int pxeemu_entry(struct v86 *x_v86_p, int *x_pxeemu_v86_flag,
	  	 int x_pxeemu_func_nr, vm_offset_t x_pxeemu_func_arg)

Its address is contained in the PXENV+ structure (pxenv->PMOffset). The
nbp (network boot program) uses this address to invoke PXE calls. This is 
analogous to the real mode PXE API calling mechanism. This facility
is non-standard and had to be invented as the PXE standard provides no
way (at least none I could figure out) to access the PXE API from 32bit 
protected mode.

The address of the PXENV+ structure is passed to the nbp in the %ebx register
as required by the standard. The nbp is loaded and executes at 0x7c00 again as 
required by the standard.

The *jmp magic is to allow the emulation code to utilize the nbp 
infrastructure to make vm86 calls, as _real_to_prot and _prot_to_real are 
unusable once the nbp switches to protected mode. The idea is that when we 
want to make a vm86 call we fill in the v86 structure, set the v86_flag, save
our state (setjmp) and return immediately. The caller checks this flag on 
return from the API call, if set it executes the vm86 action defined by the 
v86 structure and calls the API function again with the flag set. The API 
entry function notices that the flag is set and does a longjmp to return to 
the saved state. 

For this to work correctly the emulation code and the nbp must not use the 
same stack, which is no problem as the nbp defines its own stack area as 
does ETHERBOOT. We switch stacks on API entry and exit (using *jmp routines). 

Currenty the v86 facility is used only by console_putc to provide video
console feedback.

IMO it should be relatively easy to modify pxelinux to use this PXE emulation,
especially so if we are willing  do away with all the v86/*jmp stuff. This
would mean no video console output (com console is ok) from the emulation 
layer once the nbp is running. Should be okay as the nbp will provide feedback
using its own mechanisms.


regards, 

rohit