Date: Wed, 23 Apr 1997 00:03:05 +0000 (   )
From: "Gurunandan R. Bhat" <grbhat AT unigoa DOT ernet DOT in>
To: Eduard Ralph <EBRALPH AT AUGSBURG DOT BAYNET DOT DE>
Cc: djgpp AT delorie DOT com
Subject: Re: DPMI and Page swapping
In-Reply-To: <5jitme$s6o$1@sparcserver.lrz-muenchen.de>
Message-Id: <Pine.LNX.3.91.970422235915.1341A-100000@aditya.unigoa.ernet.in>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Precedence: bulk



On Tue, 22 Apr 1997, Eduard Ralph wrote:

> When you try to access a page of memory (beign 4k big) which is
> currently swapped to disk, the processor generates an exception. From
> there the handler takes over. It chooses a page which was accessed the
> least, though it can choose any one, swaps that to disk and then loads
> the page from disk. It then proceeds to change the GDT or IDT what
> ever to correct the info and then calls up the instruction which
> caused this fault. Generally it work fine and the software continues
> as if it had never been interrupted.
> The DPMI Server, which takes over this job, must know which are
> swapped out and which are not, because a flag in the Table tells him
> of the status.


yes that is correct. there is a bit in the descriptor "p" (for present 
i think, check the libc.a docs for __dpmi_get_descriptor(...)) which 
tells whether the segment is present in ram or paged. in fact i just 
discovered that virtual memory managers use this bit for their jobs.