Re: Using signals to unblock calls to msgrcv() in a multi-threaded process (Cygwin v1.7.33)

[Corinna Vinschen <corinna-cygwin at cygwin dot com>]

On Jan 19 17:21, Corinna Vinschen wrote:
> On Jan 19 11:38, Corinna Vinschen wrote:
> > On Jan 19 08:29, SCOTT Damien wrote:
> > > [...]
> > > I have included a simplified version of my code below. I usually
> > > don't see any evidence of the SIGUSR1 signal being received by either
> > > of the listener threads. Occasionally the first thread unblocks as
> > > expected but I have never seen both threads unblock. If I change my
> > > code so that there is only a single listener thread, then everything
> > > works correctly every time. Changing the msgrcv() call to a sleep
> > > call also allows unblocking to occur correctly. Note that this
> > > simplified code does not include removal of the created queues (since
> > > it never manages to exit anyway) so I manually remove these after each
> > > run using the ipcrm command.
> > > 
> > > Example output:
> > > 
> > > Running thread 2 with queue Id = 1441792
> > > Running thread 1 with queue Id = 1179649
> > > About to kill thread 1
> > > 
> > > Is there an issue with the Cygwin implementation of signal handling,
> > > or is there a problem with my code? Any help would be much
> > > appreciated.
> > 
> > It's an issue with the Cygwin code, probably.  The msgrcv code ultimately
> > hangs in a blocking ReadFile call on a named pipe, and this call is
> > non-interruptible.  Changing that requires a bigger change in Cygwin
> > which will take time.
> 
> ...or not.  I'm not sure I can fix that for the upcoming 1.7.34 release,
> but it seems cygserver is using the wrong signal handle.  At one point
> the signal handling got changed a lot to improve per-thread signalling,
> and, as far as I can see, the cygserver code still assumes a global
> per-process signal handle, which is the old version from before the
> change.
> 
> Just to let you know I'm looking into that.  Stay tuned.
> 
> Oh, btw., there's a bug in your testcase:
> 
>   int id = msgget(key, IPC_CREAT|IPC_EXCL);
> 
> This call is missing the permission bits, so the permissions are set
> to 000.  The permission bits should be added:
> 
>   int id = msgget(key, IPC_CREAT|IPC_EXCL|S_IWUSR|S_IRUSR);

Well, it was not as complicated as I anticipated.  I applied a patch
to cygserver to pull it into the wonderful modern times of per-thread
signalling.  I created and uploaded a new developer snapshot to
https://cygwin.com/snapshots/ with this change.

Please give it a try and report back.


Thanks,
Corinna

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

Attachment: pgpOJQ01lt_kq.pgp
Description: PGP signature