From owner-freebsd-hackers  Wed Oct  7 07:43:23 1998
Return-Path: <owner-freebsd-hackers@FreeBSD.ORG>
Received: (from majordom@localhost)
          by hub.freebsd.org (8.8.8/8.8.8) id HAA01756
          for freebsd-hackers-outgoing; Wed, 7 Oct 1998 07:43:23 -0700 (PDT)
          (envelope-from owner-freebsd-hackers@FreeBSD.ORG)
Received: from pcnet1.pcnet.com (pcnet1.pcnet.com [204.213.232.3])
          by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id HAA01743
          for <freebsd-hackers@FreeBSD.ORG>; Wed, 7 Oct 1998 07:43:20 -0700 (PDT)
          (envelope-from eischen@vigrid.com)
Received: (from eischen@localhost)
	by pcnet1.pcnet.com (8.8.7/PCNet) id KAA12225;
	Wed, 7 Oct 1998 10:43:10 -0400 (EDT)
Date: Wed, 7 Oct 1998 10:43:10 -0400 (EDT)
From: Daniel Eischen <eischen@vigrid.com>
Message-Id: <199810071443.KAA12225@pcnet1.pcnet.com>
To: dbeck@freesoft.hu, freebsd-hackers@FreeBSD.ORG
Subject: Re: pthreads and semctl problem
Sender: owner-freebsd-hackers@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG


> This is the first time I post to this list. 
> Big hello to everyone :)
> 
> I started to port an application from DEC Unix to FreeBSD and
> an unexpected problem cropped out. The app runs on 3 threads
> and uses system V semaphores for synchronizing access to a
> shared memory block between unrelated processes. (It 
> communicates with unrelated processes using that shared
> memory block.) There are a couple of other reasons for why 
> I couldn't use pthread mutexes in this case. 
> 
> The problem is I found that when I block one thread with the
> semop call, all the other threads will be blocked. Why ? Is it
> a problem in the C library or the kernel ?
> 
> I compiled the program with the -pthread flag. Maybe I missed
> some other flags or switches ?
> 
> The same code works under DEC and Linux :(

FreeBSD pthreads currently doesn't wrap semop(), so any attempt to
wait for the semaphore will hang.  It doesn't look hard to fix
the threads library for semop; I might try implementing this when
I get some free time.

You can add a hack to your application to make it work by using
the IPC_NOWAIT flag on the semop call, and then if the thread
doesn't get the semaphore, issue a pthread_yield() which will let
other threads run.

Someone else reported this problem also.  Search the GNATS problem
database to see if it's there.  If not, submit a PR so it doesn't
get lost.

Dan Eischen
eischen@vigrid.com

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message