Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 26 Oct 2004 17:07:09 -0400
From:      John Baldwin <jhb@FreeBSD.org>
To:        Stephan Uphoff <ups@tree.com>
Cc:        freebsd-current@FreeBSD.org
Subject:   Re: number of CPUs and IPI panic
Message-ID:  <200410261707.09293.jhb@FreeBSD.org>
In-Reply-To: <1098811105.20390.16988.camel@palm.tree.com>
References:  <20041020221659.33824.qmail@web21124.mail.yahoo.com> <200410211705.45147.jhb@FreeBSD.org> <1098811105.20390.16988.camel@palm.tree.com>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tuesday 26 October 2004 01:18 pm, Stephan Uphoff wrote:
> On Thu, 2004-10-21 at 17:05, John Baldwin wrote:
> > On Wednesday 20 October 2004 09:05 pm, Stephan Uphoff wrote:
> > > Can you try the attached patch?
> > > ( Hope it still applies since it is a few weeks old)
> > >
> > > If this does not work I have a few more questions next week.
> >
> > Unfortunately it can enable interrupts with sched_lock held which can
> > result in deadlocks due to ithread preemption (IIRC, maybe not as badly
> > with the newer preemption code.)
>
> The code only enables interrupts for TLB shootdowns.
> In this case it is guaranteed that it holds the smp_ipi_mtx that as a
> side effect protects it against preemption and holds no other spin
> mutex.
> I don't see the problem.

I must have missed the TLB shootdown detail then when I looked over it.  We 
send some other IPI's such as IPI_AST while holding sched_lock and if we 
enabled interrupts during that we could have problems.

> I am thinking about implementing the bitmap based stuff be talked about
> a few weeks ago to avoid having more than one queued IPI of the same
> type.

-- 

John Baldwin <jhb@FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve!"  -  http://www.FreeBSD.org/



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200410261707.09293.jhb>