Date: Wed, 7 Jan 1998 13:29:24 GMT
From: cmatraki AT eleceng DOT ucl DOT ac DOT uk (Chris Matrakidis)
Message-Id: <199801071329.NAA24840@lupus.eleceng.ucl.ac.uk>
To: djgpp-workers AT delorie DOT com
Subject: Re: djgpp v2.02 alpha 980101
Precedence: bulk

> > I *can't* make it 0x10ffff.  The 80386 CPU chip doesn't have that much
> > resolution.  The size of segments larger than 64k must be multiples of
> > 4K.  The closest 4k to 0x10ffff is 0x110000.
> 
> You *can*. You even cannot do otherwise.
> You are right but the resolution for 20+ bits addresses, but a segment
> which has a size of N has a limit of N-1. The guys who wrote the DPMI
> spec's required (not in that many words, IIRC) you should pass the limit
> with the 12 LSbits set. Under CWSDPMI this causes the actual limit to be
> 4K more than you intended it to be, i.e 0x110FFF.
> 

From the DPMI spec v0.9 section 8.8:

      o    Segment limits  greater than  1 meg  must be  page
           aligned.     That  is,  limits  greater  than  one
           megabyte must have the low 12 bits set.

(also on page 63 of the v1.0 spec, with a small change)