Mail Archives: djgpp/1995/01/31/17:50:01
Charles Sandmann (sandmann AT new-orleans DOT NeoSoft DOT com) writes:
> The source to access.c is in dj112m3.zip - note that the flags are
> not being set to zero in the register structure before the call.
> I think this is the problem. Both DPMI and VCPI are obviously
> protecting you by clearing some bit that should not be set. Sorry,
> I am too busy to test this theory, but it looks like that is the
> cause.
I just discovered this in my mailbox from October last year; looks
like it got lost!
------- Start of message section of RFC 934 digest (1 message) -------
Return-Path: <buers AT DG1 DOT CHEMIE DOT UNI-KONSTANZ DOT DE>
Message-Id: <9410231843 DOT AA09505 AT DG1 DOT CHEMIE DOT UNI-KONSTANZ DOT DE>
From: buers AT dg1 DOT chemie DOT uni-konstanz DOT de (Dieter Buerssner)
To: djgpp AT sun DOT soe DOT clarkson DOT edu
Subject: Bug in access, with fix
Date: Sun, 23 Oct 94 19:43:54 +0100
access() will fail sometimes.
As a previous discussion showed, the flags in the register structure
must be set, before calling go32_dpmi_simulate_int.
The trivial patch does this. Of course it might be even better to use memset
to clear the whole register structure, but, after only clearing the flags,
the bug never showed again.
Dieter
*** access.oc2 Mon Sep 12 00:26:16 1994
- --- access.c Sun Oct 16 05:07:32 1994
***************
*** 21,27 ****
int access(const char *fn, int flags)
{
_go32_dpmi_registers r;
! r.x.ss = r.x.sp = 0;
r.h.ah = 0x43;
r.h.al = 0;
r.x.dx = _go32_info_block.linear_address_of_transfer_buffer & 0xf;
- --- 21,27 ----
int access(const char *fn, int flags)
{
_go32_dpmi_registers r;
! r.x.flags = r.x.ss = r.x.sp = 0;
r.h.ah = 0x43;
r.h.al = 0;
r.x.dx = _go32_info_block.linear_address_of_transfer_buffer & 0xf;
------- End of message section of RFC 934 digest -------
- Raw text -