Date: Thu, 12 Oct 1995 10:21:56 +0200 (IST)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: "Iota (Tom Wenisch)" <FRITZW AT URIACC DOT URI DOT EDU>
Cc: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Re: Wierd problems with VESA interface under DPMI


On Sat, 7 Oct 1995, Iota (Tom Wenisch) wrote:

> I'm having a truly strange problem getting some VESA-based graphics code to
> work.  The strange thing is, when stepped through in GDB, the code runs
> flawlessly, but when run from the command line, the code crashes DOS, WIN95,
> and every configuration and memory manager I have on my system.

Use __dpmi_int() instead of the function you've used; it zeroes out the 
SS, SP and the FLAGS registers before calling the DPMI server, which you 
didn't do in your code.  Note that this is explained in the DJGPP FAQ 
list (available as faq102.zip at the same place you get DJGPP), in 
section 18.3:

18.3  Q: My program crashes/doesn't do what it should when I call
         _go32_dpmi_simulate_int().
      A: You should zero out some of the fields of the
         _go32_dpmi_registers structure before you call
         _go32_dpmi_simulate_int().  Random values in these fields can
         cause your program to behave erratically.  The fields in point
         are SS, SP and FLAGS.  When SS and SP are zeroed, the DPMI host
         will provide a stack for the interrupt handler which is at least
         30-word long (most DPMI hosts provide much larger stack); this is
         usually enough, but if it isn't, then you should point SS and SP
         to larger buffer in conventional memory (possibly part of the
         transfer buffer).