Date:	Tue, 7 Dec 1999 12:16:15 +0200 (IST)
From:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
X-Sender:	eliz AT is
To:	Peter Johnson <pljohnsn AT uiuc DOT edu>
cc:	djgpp-workers AT delorie DOT com, Charles Sandmann <sandmann AT clio DOT rice DOT edu>
Subject:	Re: NT4 DPMI BUG (303h, Allocate Real-Mode Callback)
In-Reply-To:	<4.2.1.19991206190011.00af0210@ews.uiuc.edu>
Message-ID:	<Pine.SUN.3.91.991207121512.29709K-100000@is>
MIME-Version:	1.0
Reply-To:	djgpp-workers AT delorie DOT com
X-Mailing-List:	djgpp-workers AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

I cc: the developers' list and Charles Sandmann, since I think this is
something that should be discussed.  A summary of this issue is
attached below, for those who didn't read this thread on c.o.m.d.

On Mon, 6 Dec 1999, Peter Johnson wrote:

> Another question/problem:  I'm going to have to not use the 
> __dpmi_allocate_real_mode_callback function in the new versions of the 
> _go32_* functions because __dpmi_allocate_real_mode_callback always passes 
> the selector for the data structure as ds, and I need to have it be the 
> custom allocated selector.
> 
> Ideas?

How about adding a new function, called, say,
__dpmi_allocate_real_mode_callback_with_selector, which can be passed
the selector as an additional argument?

(Btw, __dpmi_allocate_real_mode_callback uses CS, not DS, since DS
might be invalid when the real-mode callback is called, e.g. due to
some pending signal.)

One other issue about the solution proposed by Peter: it sets the
limit of the additional selector to 32h (since it only spans the
real-mode registers' data structure.  This makes the segment be
byte-granular (as opposed to page granularity used by DJGPP for all
other selectors).  Can this cause any problems?  Is it safer to define
a page-granular segment?

Summary for those who didn't read this thread: Peter is working on
changing _go32_dpmi_allocate_real_mode_callback_* functions to work
around a bug on NT whereby only the lower 16 bits of the offset of the
callback address are used; NT ignores (or garbles) the upper 16 bits.
For this to work, Peter allocates an extra descriptor whose base
address is such that the offset is zero, and passes the selector of
that descriptor to the DPMI host.

I asked Peter to make this special code be executed only when the
program actually runs on NT, to minimize possible adverse effects on
other platforms (shortage of selectors, etc.).

- Raw text -