Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wuarchive!udel!rochester!pt.cs.cmu.edu!o.gp.cs.cmu.edu!andrew.cmu.edu!<UNAUTHENTICATED>+
From: Richard.Draves@cs.cmu.edu
Newsgroups: comp.os.mach
Subject: Re: Interrupting a msg_receive from another thread
Message-ID: <4bxs0Cu00hsQ5UaWR0@cs.cmu.edu>
Date: 1 Apr 91 18:46:38 GMT
References: <283@trwacs.UUCP>
Organization: Carnegie Mellon, Pittsburgh, PA
Lines: 44
In-Reply-To: <283@trwacs.UUCP>

> Excerpts from netnews.comp.os.mach: 1-Apr-91 Interrupting a msg_receive
> .. Jeremy Epstein@trwacs.UU (1672)

> Environment: Running Mach 2.5 on a Sun 3, using the cthreads library.

> I've got a server with three threads.  Clients send this server
> receive rights on two ports, one for normal messages and one for
> critical messages.

> In the server, one thread (call it A) sits and builds up a port set (by
> collecting receive rights for the normal messages), and also selects one
> of the critical ports to be the focus port.  The second thread (thread B)
> reads messages from the port set consisting of all of the normal ports.
> The third thread (thread C) reads from the current focus port only.

> There are two problems here: First, the port set may increase (new ports
> may be added by thread A) while thread B is doing a msg_receive()).
> Second, thread A may choose a different focus port while thread C is
> doing a msg_receive().

> In the first case, I can overcome the problem by periodically waking
> up from the msg_receive (i.e., use a timeout) and select the port set
> again [or will the port_set_add() in thread A automatically cause the
> msg_receive in thread C to see messages which arrive on the new port?]
> In the worst case, I'd wait a big longer than I need before I receive
> a message from the new port.  This isn't too serious.

Thread B does not have to restart its msg_receive to see the effects of
port set membership changes.

> However, in the second case I could receive a message from a port which
> is no longer the focus port.  This turns out to be undesirable.

> The question [finally!]: Is there any reasonable way for thread A to
> wake thread C from its msg_receive()?

The easiest way to wake a thread from a msg_receive is to send the
thread a message.  You could also put the critical ports into a port
set, except for the focus port from which thread C receives.  To change
focus ports, thread A takes the new focus port out of the port set and
adds the old focus port to the port set.  This will wake C with an
error, because you can't receive directly from a port in a port set.

Rich