[Etherboot-developers] Re: Disk on chip work-arounds

[Ken Y. <ke...@nl...>]

[Further discussion directed to eth...@li...,
please subscribe if you are not already on the list.]

>Hi Ken
>
>I saw this posting which sums up my problem exactly. I am
>using etherboot 4.6.10 and wonder if any of the work-arounds
>that you (might of) provided to mar...@hp... worked?
>
>		"By: ken_yap ( Ken Yap ) 
>                           RE: DiskOnChip [ reply ]   
>                         2000-Jul-05 07:14 
>                         This is a hard one and gets asked
>often. 
>                         The problem appears to be that DOC
>uses some 
>                         memory just under 0xA0000 and this
>clashes 
>                         with Etherboot use. 
>                         You can't just redefine RELOCADDR,
>this has 
>                         only the effect of shifting the
>base address 
>                         of the code down to 0x96000 or
>0x94000 to 
>                         allow for more features, but the
>clash still 
>                         exists. 
>                         The solution may be to move
>Etherboot to 
>                         0x88000 and not load kernels larger
>than 
>                         544kB or to load bzImage kernels
>(above 1MB). 
>                         To do this, the formulas in the
>assembler 
>                         files need to be edited. 
>                         Another wrinkle is that the
>decompressor will 
>                         need to be moved down to 0x70000
>but at least 
>                         this doesn't affect the kernel
>size. 
>                         If you are willing to experiment,
>contact me 
>                         by email and I will give you some
>things to 
>                         try"

A couple of versions ago, I put in some small changes to support running
from the 0x80000 segment, but it still uses some memory areas above
0x90000.  I don't have DOC, but the changes didn't break any existing
usage.  Subsequently Stuart Lynne (sorry if my memory's got your name or
affiliation wrong) of Lineo communicated to me privately that DOC uses
areas above 0x90000 and that to have any chance of success, everything
would have to be moved out of 0x90000, probably down to 0x80000.  This
would also include areas above 0x90000 used by kernel startup segment.
Apparently there are patches floating around the Linux kernel world for
this. The details are my email at home, which I can post later.

If you are serious about this and are willing to invest some time in
hacking the code, I'm willing to talk you through this. But you would
have to do the experimentation, as I don't have DOC.