[Qemu-devel] Re: Problem with DOS application and 286 DOS Extender application

[Gerhard Wiesinger]

Hello,

After some fortune I found out that also Turbo Debugger 286 doesn't work 
under plain DOS 6.22 (without any memory mananger just pressing F5) or 
with some memory mananagers (HIMEM.SYS, EMM386, QEMM386).

Error message is:
Error 266 loading D:\DIR\TD286.EXE into extended memory.

So it looks like that there is a major issue with extended memory. Any 
ideas how to fix or how to find the problem and fix it?

Version is latest seabios and QEMU from git as of now (own builds).

I'm pretty sure that it is the same reason that the 286 DOS Extender 
application doesn't work.

For full reference of the previously discussed have a look here:
http://www.mail-archive.com/address@hidden/msg29518.html
http://www.mail-archive.com/address@hidden/msg29465.html

Thnx.

Ciao,
Gerhard

--
http://www.wiesinger.com/


On Wed, 14 Apr 2010, Jan Kiszka wrote:

> Jamie Lokier wrote:
> > Gerhard Wiesinger wrote:
> > > It is a non public, proprietary application which uses the Ergo Computing
> > > 286 DOS Extender. I guess some other application which use the same DOS
> > > extender have the same problem. So best thing is to find another
> > > application which uses the Ergo Computing 286 DOS Extender, too.
> >
> > The 286 was obsolete 20 years ago, although code depending on it
> > persisted for some years after.
> >
> > I'm fairly sure the number of people using (or trying to use) Qemu
> > with 286-specific code is very small indeed, so unfortunately for a
> > 286 problem, you will need to help reproduce it as much as you can for
> > it to be fixed.
>
> In some scenarios, we use QEMU in emulation mode for such a legacy guest
> (16-bit protected mode), but we mostly run it in KVM mode these days. It
> works fairly well under QEMU, but also we did not explore all corner cases.
>
> > Note that Qemu doesn't emulate segments properly even for 32-bit x86
> > code, and 16-bit (286) code depends on that all the more.  That may be
> > the problem.
>
> More precisely: QEMU does not check for segment limits. This can be a
> problem with buggy or pedantic guests, but usually one tried to avoid
> triggering this anyway. I once wrote a crude patch to add this, but it
> had significant performance impact and did not properly make use of the
> TCG to optimize the checks. You'll find it in the archives (but I guess
> it no longer applies).
>
> > Or it may be the "reset using keyboard controller and BIOS" method
> > used to switch from protected mode to real mode on a 286 is not
> > implemented properly, or is not supported by the BIOS properly.
> >
> > Or it may simply be a bug in 16-bit task segment switching or
> > something like that, which is quite complex and so rarely used that it
> > might never have been properly tested.
>
> Task switching looks fairly stable in QEMU (in contrast to KVM where we
> just ran into some more corner cases).
>
> > Did you try running the application under Bochs, which has a more
> > accurate emulation of very old x86 CPUs?
> >
> > -- Jamie
>
> That said, having some test case to reproduce the issue is essential.
> I'm willing to have a look if you can provide such thing (publicly or
> privately). Before that, you could already try building QEMU with
> --enable-debug and run it with "-d exec,int". The generated
> /tmp/qemu.log may point out where things go wrong (usually where faults
> starts to occur).
>
> Jan