[LispM-Hackers] Loaded ersatz ucode and N928.LOAD on x86
Fri Mar 8 16:04:01 2002
"James A. Crippen" wrote:
> Yeah, but doesn't this cause the problem of needing to know how much
> memory you have? I mean, we don't have virtual memory yet, so if you
> stick something up at 1GB but you only have 256MB then you'll lose.
> I'm not certain here, since I've no experience down there at the
> highly ferrous level.
Actually, the Etherboot guys are pretty on-the-ball. They have a flag
you can set in the load record for a particular set of bytes that tells
the bootloader how to interpret the load address for the set of bytes.
One of the options says "make it relative to the end of physical memory
(whatever that is)."
Then your boot program grabs a "far pointer" off the stack and chases
down the records to find out where things got put. I forget whether you
have to redo the math, or whether the bootloader (as a courtesy) puts
the right address in. I'd have to check the code (I used to stick the
classpath zipfile in high memory and use it as a zip-format ramdisk).
> Ancient? Wheezing? I have a 486DX2/50 (with PCI no less!) that wants
> to talk to you.
Oh, goody. It has a floating-point unit (or so "The Undocumented PC"
tells me). Hmm... If we're going to do 28 bit floating point, are we
going to do it in software, or are we going to use any floating-point
unit we have. I am not keen on writing a floating-point library.
==== John Morrison
==== MAK Technologies Inc.
==== 185 Alewife Brook Parkway, Cambridge, MA 02138
==== vox:617-876-8085 x115