Date: Thu, 21 Sep 2006 09:16:24 +0200
From: Ingo Molnar <mingo@elte.hu>
To: Bill Huey <billh@gnuppy.monkey.org>
Cc: linux-kernel@vger.kernel.org, Thomas Gleixner <tglx@linutronix.de>,
       John Stultz <johnstul@us.ibm.com>,
       "Paul E. McKenney" <paulmck@us.ibm.com>,
       Dipankar Sarma <dipankar@in.ibm.com>,
       Arjan van de Ven <arjan@infradead.org>,
       Esben Nielsen <simlo@phys.au.dk>
Subject: Re: [PATCH] move put_task_struct() reaping into a thread [Re: 2.6.18-rt1]
Message-ID: <20060921071624.GA25281@elte.hu>
References: <20060920141907.GA30765@elte.hu> <20060921065624.GA9841@gnuppy.monkey.org> <20060921065402.GA22089@elte.hu> <20060921071838.GA10337@gnuppy.monkey.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20060921071838.GA10337@gnuppy.monkey.org>
User-Agent: Mutt/1.4.2.1i
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2122
Lines: 45


* Bill Huey <billh@gnuppy.monkey.org> wrote:

> > > This patch moves put_task_struct() reaping into a thread instead 
> > > of an RCU callback function [...]
> > 
> > had some time to think about it since yesterday: RCU reaping is done 
> > in softirqs (check out the softirq-rcu threads on your -rt box), 
> > that's why i removed the delayed-task-drop code to begin with. Now i 
> > dont doubt
> 
> It's correct from the standpoint of it being reaped in another thread, 
> so it fixed those crashes. But I pushed it down into another thread at 
> the request of Esben and his private discussion with Paul McKenney, 
> since a summary from Esben felt that call_rcu() was somehow less than 
> ideal to do that.

but it _is_ already being reaped in another thread: softirq-rcu. 
Splitting that up any further will only fragment the context-switching 
and increases cache footprint - it wont (or rather, shouldnt) have any 
functional effect. (As a sidenote, i'm considering the unification of 
all 'same default priority' softirq threads into a single thread per 
CPU, to further reduce this cost of 'spreadout'.)

> > that you saw crashes under 2.6.17 - but did you manage to figure out 
> > what the reason is for those crashes, and do those reasons really 
> > necessiate the pushing of task-reapdown into yet another set of 
> > kernel threads?
> 
> Unfortunately no. I even used Robert's .config on my machine. I added 
> a disk controller and networking device driver just to boot into his 
> configuration and I still couldn't replicated any of his kjournald 
> problems at all. If I had his hardware I'd have a better way of 
> replicating those problems and pound it out.

ok, then i guess what we have left is to wait and see whether it still 
triggers with the current 2.6.18-rt codebase - maybe it triggers for 
someone in a scenario that is easier to debug.

	Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/