X-Recipient: archive-cygwin AT delorie DOT com
X-Spam-Check-By: sourceware.org
Date: Mon, 18 Jun 2012 16:51:48 +0200
From: Corinna Vinschen <corinna-cygwin AT cygwin DOT com>
To: cygwin AT cygwin DOT com
Subject: Re: Problem with MAP_PRIVATE for fork() using pypy
Message-ID: <20120618145148.GD26243@calimero.vinschen.de>
Reply-To: cygwin AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com
References: <1339287593 DOT 67918 DOT YahooMailNeo AT web162901 DOT mail DOT bf1 DOT yahoo DOT com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <1339287593.67918.YahooMailNeo@web162901.mail.bf1.yahoo.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Mailing-List: contact cygwin-help AT cygwin DOT com; run by ezmlm
Precedence: bulk
List-Id: <cygwin.cygwin.com>
List-Unsubscribe: <mailto:cygwin-unsubscribe-archive-cygwin=delorie DOT com AT cygwin DOT com>
List-Subscribe: <mailto:cygwin-subscribe AT cygwin DOT com>
List-Archive: <http://sourceware.org/ml/cygwin/>
List-Post: <mailto:cygwin AT cygwin DOT com>
List-Help: <mailto:cygwin-help AT cygwin DOT com>, <http://sourceware.org/ml/#faqs>
Sender: cygwin-owner AT cygwin DOT com
Mail-Followup-To: cygwin AT cygwin DOT com
Delivered-To: mailing list cygwin AT cygwin DOT com

On Jun  9 17:19, Uwe F. Mayer wrote:
> I have recently patched pypy (http://www.pypy.org) to compile under
> Cygwin. Pypy implements a Python interpreter with a Just In Time (JIT)
> compiler. At runtime pypy uses a MAP_PRIVATE anonymous mmap to
> allocate memory for its JIT. This seems to be cause of fork() calls
> failing. A sample error message is included below. I have discussed
> this problem at length with the pypy developers. The discussion is
> online at:
> 
> http://www.mail-archive.com/pypy-dev AT python DOT org/msg02448.html 
> 
> Finally I hacked the pypy sources and replaced the MAP_PRIVATE
> anonymous mmap allocation with MAP_SHARED at the place where pypy
> allocates its JIT temporary memory, and the code runs fine (but that's
> not a solution for pypy, as their JIT child and parent processes
> likely will drift apart for any real non-toy program and do need
> separate memory space). From what the pypy folks say, this is not a
> pypy issue but a Cygwin issue.
> [...]
> PS Attached is an output of cygcheck, and the relevant portion of an strace.

Unfortunately the "relevant portion" of the strace is not at all relevant
to see what happens.  The error message is weird.  This should not occur
at all if the memory address still exists in the parent.  I can't reproduce
it with a simple testcase (see below) either.  I ran it on a 64 bit system
so that the allocated memory area is in the large address area, just as in
your scenario.

If you could create a testcase in plain C which allows to reproduce the
behaviour, it would help a lot to track down the cause.


Corinna


#include <stdio.h>
#include <sys/mman.h>
#include <unistd.h>

int main ()
{
    char *p;
    int sz = 2 * getpagesize ();
    int offset;
    int magic = 0x33;
    int pid;
    int rv;

    p = (char *) mmap (NULL, sz, PROT_READ|PROT_WRITE,
		       MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
    if (p == MAP_FAILED)
      {
        printf ("mmap() = MAP_FAILED\n");
	return 1;
      }
    printf ("mmap() = %p\n", p);

    offset = getpagesize () + 0x42;
    p[offset] = magic;

    pid = fork ();
    switch (pid)
    {
    case (-1):
        printf ("fork() failed\n");
        break;
    case (0):
        printf ("child alive\n");
	sleep (1);
        printf ("child touching %p\n", &(p[offset]));
        printf ("child M[%p] = 0x%x\n", &(p[offset]), p[offset]);
	p[offset] = 0x44;
        printf ("child M[%p] = 0x%x\n", &(p[offset]), p[offset]);
	pid = fork ();
	switch (pid)
	{
	case (-1):
	    printf ("fork() failed\n");
	    break;
	case (0):
	    printf ("grandchild alive\n");
	    sleep (1);
	    printf ("grandchild touching %p\n", &(p[offset]));
	    printf ("grandchild M[%p] = 0x%x\n", &(p[offset]), p[offset]);
	    sleep (2);
	    printf ("grandchild M[%p] = 0x%x\n", &(p[offset]), p[offset]);
	    break;
	default:
	    printf ("child alive\n");
	    sleep (2);
	    printf ("child touching %p\n", &(p[offset]));
	    printf ("child M[%p] = 0x%x\n", &(p[offset]), p[offset]);
	    p[offset] = 0x55;
	    printf ("child M[%p] = 0x%x\n", &(p[offset]), p[offset]);
	    break;
	}
        break;
    default:
        printf ("parent alive\n");
        sleep (5);
        printf ("parent touching %p\n", &(p[offset]));
        printf ("parent M[%p] = 0x%x\n", &(p[offset]), p[offset]);
        break;
    }

    return 0;
}

-- 
Corinna Vinschen                  Please, send mails regarding Cygwin to
Cygwin Project Co-Leader          cygwin AT cygwin DOT com
Red Hat

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