Mailing-List: contact cygwin-help AT cygwin DOT com; run by ezmlm
Sender: cygwin-owner AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com
Message-ID: <3EBDFA13.1010708@coppice.org>
Date: Sun, 11 May 2003 15:21:55 +0800
From: Steve Underwood <steveu AT coppice DOT org>
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.3) Gecko/20030314
MIME-Version: 1.0
To: cygwin AT cygwin DOT com
Subject: Re: A problem with mmap on Win98
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

>On Tue, May 06, 2003 at 04:07:49PM +0800, Steve Underwood wrote:
>> I am using the current version of cygwin (1.3.22). When I build gcc to 
>> cross-compile for the TI MSP430 MCUs the result works OK on 2000 and XP 
>> machines. On some Win98 and Me machines it works, but on others it 
>> crashes with a segmentation violation. I traced this to a problem with 
>> anonymous mmap. If I keep calling mmap on the problem machines it keeps 
>> returning the same memory pointer twice. If I patch GCC to check for a 
>> repeat of the same memory pointer, and allocate again until it gets a 
>> different pointer, my GCC build seems to run OK on these problem machines.
>> 
>> The results are not random. They follow a pattern. If I allocate 65536 
>> byte blocks, I consistently get the same pointer twice, before I get a 
>> pointer 65536 bytes greater. If I allocate 16384 byte blocks, I get four 
>> pointers incrementing by 0x4000 each. Then I get a repeat of the fourth 
>> pointer. Then I get four more properly incrementing pointers.
>
>Weird, especially that it only happens on 9x...

Its weird that there is trouble with windows? :-)

Its only on some Win9x machines, too. I did my original software testing and
produced a binary installer on a Win98SE machine that worked without a hitch.
Then various users started complaining it would not work for them. Now I have
found a Win98SE machine that exhibits the problem, so I can follow up on it.
Perhaps the different behaviour relates to the system patch level, or something.

>> I tried downloading the source code for cygwin, to build it, and do a 
>> little debugging of my own. However, the make files build most of the 
>> code, and then fail saying there is no rule to make 
>> /usr/lib/w32api/Makefile needed by /lib/. The make dependancies are a 
>> little intertwined, and I gave up trying to find the cause of this problem.
>
>Did you build in the source dir?  Don't do this.  For more info see
>http://cygwin.com/contrib.html

I tries building in the suorce dir. Then I read the instructions, so I tried again
using another dir. In both cases I got the same error.
>
>> Can anyone help?
>
>Please create a *brief* testcase in C which encounters the above mmap
>problem and send it to this list.  I'll have a look, then.
>
>Did I mention that the testcase should be as short as possible? 

Is gcc-core-3.2.3 built for C support a small enough test case? :-)

Seriously, it may be better if I debug this a little, as I have a machine
exhibiting the problem. I have no idea what proportion of Win98 machines succeed
and fail.

Does anyone have pointers to solving my cygwin build problem? I have the latest
cygwin (updated this week) installed on a WinXP machine, including (of course) the
development tools packages. I downloaded the source for the current version of the
cygwin DLL. I configured to build in another directory, and got the error message
described above.

Regards,
Steve


--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/