From: muller AT cerbere DOT u-strasbg DOT fr
Message-Id: <3.0.6.32.20000218215350.009cf860@ics.u-strasbg.fr>
X-Sender: muller AT ics DOT u-strasbg DOT fr
X-Mailer: QUALCOMM Windows Eudora Light Version 3.0.6 (32)
Date: Fri, 18 Feb 2000 21:53:50 +0100
To: djgpp-workers AT delorie DOT com
Subject: Re: Bug in  crt0.S
Cc: sandmann AT clio DOT rice DOT edu (Charles Sandmann)
In-Reply-To: <38AD8997.295D9B17@inti.gov.ar>
References: <200002171707 DOT SAA16280 AT cerbere DOT u-strasbg DOT fr>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Reply-To: djgpp-workers AT delorie DOT com
Errors-To: dj-admin AT delorie DOT com
X-Mailing-List: djgpp-workers AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

At 15:04 18/02/00 -0300, you wrote:
>Pierre Muller wrote:
>
>>    Hello,
>>
>>   I  wrote the Run Time Library of the Free Pascal port  to DJGPP  system.
>> As such, I changed crt0.S to allow starting a program without any use of
the
>> C library.
>>
>>    Due to internal problems, I disabled the multiblock feature in the
>> startup code.
>> Thus our RTL uses Unixy sbrk with selector base changes.
>
>How it affects? is this problem found only when using Unixy sbrk?

  I don't know,
it really seems that its only when the base is changed!

>>    Recently, we got more and more problems with our IDE
>> (a freeware equivalent of the Borland editor BP) under Windows.
>> These crashes allways appeared in the VMOUSE VXD and completely crashed the
>> computer!
>
>I have a TV application + Win95 system that have this problem (ramdonly).
But is on
>exit.

 But is the mouse already disabled ?
Any other code using the alias could have the same bug...
doesn't  exceptn.S also use djgpp_ds_alias for the keyboard real mode
stuff (the one that catches Crtl-C) ??

  I first sent this message to Eli,
but he told me to send it to the list and to Charles Sandmann
But I forgot this, I did it right now!

>>   These errors were even more frequent on Win98 and Win 2000 !!
>> After quite long and hard debugging, I finally realized that
>> there is a major bug in the startup code:
>>
>>    when the function 0x503 of DPMI is called,
>> only the cs and ds selector are updated in the 16 byte sbrk16.asm code.
>> The djgpp_ds_alias selector (that is used for the RMCB
>> to avoid problems with fake exceptions like Crtl-C) is updated
>> later in the 32 bit code... but its updated after the
>> interrupts are reenabled !!!
>
>Ugh!
>
>>   If the mouse callback is called in the interval between the 0x503
>> function call and the update
>> of the alias selector base address => you get the crash!!
>>
>>   I include here a diff to version 2.03 of crt0.S file that also fixes the
>> problem for the C library!
>> Could someone please test this with RHIDE
>> because the recent problem about linking could be due to that fact
>> if (and I don't know if this is true) the linker is called internally !
>
>You mean it could be because the RHIDE's RMCB is called during ld startup?

  Its a possibility but I have never really understood how
other DJGPP exes are run from a DJGPP exe !