From: "Lawrence Rust" <lvr AT nospam DOT softsystem DOT co DOT uk>
Newsgroups: comp.os.msdos.djgpp
References: <5 DOT 0 DOT 2 DOT 1 DOT 2 DOT 20020609104552 DOT 009e64e0 AT pop DOT gmx DOT net> <advaa1$cgm$1 AT antares DOT lu DOT erisoft DOT se>
Subject: Re: Question about djgpp's int-wrapper
Lines: 63
X-Priority: 3
X-MSMail-Priority: Normal
X-Newsreader: Microsoft Outlook Express 5.50.4807.1700
X-MimeOLE: Produced By Microsoft MimeOLE V5.50.4910.0300
Message-ID: <DILM8.896$ca2.121610@newsfep1-win.server.ntli.net>
Date: Sun, 9 Jun 2002 17:51:23 +0100
NNTP-Posting-Host: 62.253.142.12
X-Complaints-To: abuse AT ntlworld DOT com
X-Trace: newsfep1-win.server.ntli.net 1023641507 62.253.142.12 (Sun, 09 Jun 2002 17:51:47 BST)
NNTP-Posting-Date: Sun, 09 Jun 2002 17:51:47 BST
Organization: ntl Cablemodem News Service
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com

"Martin Stromberg" <eplmst AT lu DOT erisoft DOT se> wrote in message
news:advaa1$cgm$1 AT antares DOT lu DOT erisoft DOT se...
> Martin Steuer (martinSteuer AT gmx DOT de) wrote:
> : The problem i mean is that the DPMI Specification says that
non-priviledged
> : code may not be able to alter the IF via 'popf' or 'iret'.
> : Therefore the DPMI Spec. suggests to insert a 'sti' before an 'iret' in
> : hardware interrupt handlers, because 'sti' is a priviledged instruction
it
> : will be trapped and the DPMI Host will then execute an 'sti' in Ring0.
>
> 1. Where does it say that?

http://www.delorie.com/djgpp/doc/dpmi/ch4.4.2.html


> 2. IIRC, if sti is a priviledged instruction (because of IOPL < 3),
> then iret and popf are too. So why would an emulated sti set IF and
> not an emulated iret or popf?

If you look at Intel's IA32 architecture manual:

ftp://download.intel.com/design/pentium4/manuals/24547107.pdf

you'll see on page 3-49 that in protected mode, iret only restores
EFLAGS(IF) if CPL <= IOPL.  This is also true of popf.  So if the code is
run on a system such as Windows NT, 2K or XP where CPL is 3 and IOPL is 0
then the task will be left with virtual interrupts disabled.  The system
will remain responsive but the task will not receive any more interrupts
:-(.

However, sti signals GP(0) if CPL > IOPL, which on NT is handled by the
NTVDM and will enable virtual interrupts to the task.

IMHO this is a design fault in the IA32.  The correct action would have been
to signal GP(0) if EFLAGS(IF) is altered.  This gives the host OS (NTVDM)
the chance to update the state of virtual interrupts for the task.

This design fault prevents many protected mode games, like Doom and
DukeNukem3D, from running with sound on NT, Win2K and XP.  In many cases the
sound driver disables interrupts around sound card h/w accesses with code
like...

pushf
cli
...
popf ; Restore IRQ state

Here the cli traps to NTVDM which obligingly disables the virtual interrupt
state.  However the popf that should restore the interrupt state does
nothing so virtual interrupts remain disabled.

If you want to run these games with a sound card emulator like SoundFX
(www.softsystem.co.uk/sndfx202.zip) then it's necessary to patch the code to
remove the cli opcodes.

--
Lawrence Rust
Software Systems, www.softsystem.co.uk

The problem with Windows XP - http://www.arachnoid.com/boycott