Subject: Re: libpthread
To: Lennart Augustsson <lennart@augustsson.net>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: tech-kern
Date: 06/20/2003 13:29:29
On Fri, Jun 20, 2003 at 12:49:12PM +0200, Lennart Augustsson wrote:
> Thor Lancelot Simon wrote:
>
> >Honestly, I do not think it was a good decision on core's part to allow
> >the pthread code to come in in a state in which it would render the
> >preexisting SMP functionality so unstable as to be useless in a multiuser
> >environment, certainly not if it was going to stay that way for *months*
> >on end.
> >
> Well, it's not only on SMP that pthreads are broken. They don't
> really work on a single processor either. Just read email with
> mozilla and crashes abound.
I don't know about "abound" (I've never seen one) but I do know that Eric
can make Mozilla crash the machine at will by repeatedly stopping and
continuing it using a "helper" program. It takes some time for it to occur,
but it's reproducible. The panic is in remrunqueue, I believe; we were
going to at least take a better peek at it when I was at his house a few
weeks ago but ran out of time, and now we're both 3,000 miles from our NetBSD
systems and their consoles... :-(
The SMP issue is really bad. I have a cluster of 30 nodes that are stuck
at -current as of just before the pthread merge because I can't run the risk
of malicious or juvenile users using the pthread system calls to crash the
machines.
Thor