Re: 2.0.32 talkd's stuck in do_down -- weird!

Malcolm Beattie (mbeattie@sable.ox.ac.uk)
Tue, 10 Feb 1998 10:15:36 +0000 (GMT)


Chris Evans writes:
>
> Hi,
>
> Last night our load average jumped and stayed at around 3. top did not
> show any noticeable activity.
>
> Eventually we noticed 3x in.ntalkd processes. These were stuck in the "D"
> state and accounting for the load average.
>
> There were waiting in "__do_down" for some unknown semaphore I presume!!
>
> Eventually they were persuaded to go away when we killed an idle bash(!!).
> The bash involved would have spawned one of the ytalk's associated with
> the in.ntalkd processes.
>
> So..
>
> 1) Why was our load average contributed to?

Processes in kernel mode count as running in the load average. At least,
that holds for other Unices so I suppose it holds here.

> 2) What on earth could the in.ntalkd's possibly have been waiting on in
> uninterruptible sleep that freed up when the bash process exited? Some
> weird tty layer thing?

Very likely. talk daemons going uninterruptible have affected us ever
since Ultrix was around and in Digital UNIX too: I'd wondered if it
was only a DEC thing :-/. I suspect the bug lies in the way ttys are
closed: the only programs which are likely to try to write to ttys in
strange states are talkd and shutdown (oh, and write but that's not as
heavily used). As you'd expect, we've seen problems with shutdown
hanging during its "wall" warning phase in Digital UNIX too.

--Malcolm

-- 
Malcolm Beattie <mbeattie@sable.ox.ac.uk>
Unix Systems Programmer
Oxford University Computing Services
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu