From: Regis BOSSUT <bossut AT isen DOT fr>
Newsgroups: comp.os.msdos.djgpp
Subject: [BUG] Total crash under MS-DOS when using itimer + workaround
Date: Mon, 13 Oct 1997 12:03:57 +0200
Organization: IEMN - Dépt ISEN - Acoustique
Lines: 36
Message-ID: <3441F20D.4F8D@isen.fr>
NNTP-Posting-Host: hpac1.isen.fr
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Precedence: bulk

Hi all !

I've been spending two days trying to figure out why my small
program was crashing my PC. Here is the context :

- A program allocating a much memory via malloc,
- using signal(SIGALARM,timeout),
- using setitimer() for a recurrent 2 second interrupt,
- doing some disk and screen I/O.

Under a DOS box in Win95, the program works fine. Under MSDOS 7.00,
the program works fine, up to the end. Just before being prompted for
a new command, I get a General Protection Fault in RMCB with error
0xa4 (I don't have a list of errors...)

Use of fsdb or gdb is not of any help (same type of crash).

Allocating less memory made the program work some times.

After a two day brainstorming, I thought about a timer HW interrupt
not being serviced correctly, so I added a setitimer() to disable
the timer just before exiting, by means of atexit(). Now the program
works fine !

By carefully checking the source code about timers and exit routines,
I discovered that the timer HW int. is not disabled before exiting.
This should be done around keyboard and cbrk HW ints. disabling, in
go32/dpmiexcp.c.

I'm not DPMI aware enough to add this functionality, but I'm sure that
it would not be difficult for the author(s) to correct this. Also,
I have a workaround...

Hope this will contribute to DJGPP performance.

/Régis