Re: [PATCH] sched/completion: completion_done() should serialize with complete()

From: Davidlohr Bueso <dave@stgolabs.net>
To: Oleg Nesterov <oleg@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	Peter Zijlstra <peterz@infradead.org>,
	linux-kernel@vger.kernel.org, waiman.long@hp.com,
	raghavendra.kt@linux.vnet.ibm.com
Subject: Re: [PATCH] sched/completion: completion_done() should serialize with complete()
Date: Fri, 13 Feb 2015 14:02:52 -0800	[thread overview]
Message-ID: <1423864972.2046.84.camel@stgolabs.net> (raw)
In-Reply-To: <1423864582.2046.82.camel@stgolabs.net>

On Fri, 2015-02-13 at 13:56 -0800, Davidlohr Bueso wrote:
> On Thu, 2015-02-12 at 20:59 +0100, Oleg Nesterov wrote:
> > Commit de30ec47302c "Remove unnecessary ->wait.lock serialization when
> > reading completion state" was not correct, without lock/unlock the code
> > like stop_machine_from_inactive_cpu()
> > 
> > 	while (!completion_done())
> > 		cpu_relax();
> > 
> > can return before complete() finishes its spin_unlock() which writes to
> > this memory. And spin_unlock_wait().
> 
> How about reverting the patch altogether?
> 
> This was never a problem nor have I ever seen a performance issues in
> completions that would merit these lockless checks. The commit changelog
> has *zero* information, so I don't know if this was ever a real issue.
> 

hmm I guess you're patch is more optimal tho, because we don't update
the lock, less cacheline bouncing issues etc.