Sender: bill AT delorie DOT com
Message-ID: <387EF75A.C883B18B@taniwha.org>
Date: Fri, 14 Jan 2000 23:15:54 +1300
From: Bill Currie <bill AT taniwha DOT org>
X-Mailer: Mozilla 4.61 [en] (X11; I; Linux 2.2.14 i686)
X-Accept-Language: en
MIME-Version: 1.0
To: djgpp-workers AT delorie DOT com
Subject: Re: MWDPMI exception handling
References: <Pine DOT OSF DOT 4 DOT 20 DOT 0001131047590 DOT 31342-100000 AT sirppi DOT helsinki DOT fi>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Reply-To: djgpp-workers AT delorie DOT com

Esa A E Peuha wrote:
> /* Pass exception to client */
>         addl    %ebp, %ebp      /* Item size now 8 */
>         cs
>         addl    _current_exception_table, %ebp
>         pushl   $except_return_point    /* Return EIP */
>         cs
>         pushl   4(%ebp) /* Handler CS */
>         cs
>         pushl   (%ebp)  /* Handler EIP */
>         movl    $gdt_code32_sel, %ebp
>         xchgl   %ebp, 12(%esp)  /* Return CS */
>         lret

If you can, check the CPL at this point and the DPL of the handler (as
on the stack). If DPL < CPL (numericly), you will get a GPF. Note, I
can't see why this would be the case. 

Hmm, are you sure the GPF is happenning at the lret? It may be happening
at the lret (iret? been too long) of the handler. It is illegal to lret
(or iret) from (eg) cpl=3 to cpl=0.

What needs to happen is the dpmi server needs to lret into the correct
cpl, call the handler, then ljmp through a gate to the rest of the dpmi
server.

eg

	pushl $handler_stub_cs  /* alias for dpmi seg, but has correct PL */
	oushl $handler_stub
	lret
handler_stub:
	lcall (handler)
	ljmp	dpmi_handler_gate:junk 
dpmi_handler_gate_dest:
	...

Note, this is highly simplified (you need to handle the stack changes,
etc), but it should give you an idea.

Basicly, whats needed is two (4 if pl 1 and 2 supported) extra
descriptors: one that is a pl3 alias of the dpmi server segment (you
might be able to get away with this using conformant segments), and one
that is a call gate pointing to the return location within the server.
Hmm, I suspect conforming segments were designed for this reason.

HTH
Bill
-- 
Leave others their otherness. -- Aratak