From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
        id S1760221AbcHaVr5 (ORCPT <rfc822;w@1wt.eu>);
        Wed, 31 Aug 2016 17:47:57 -0400
Received: from gate.crashing.org ([63.228.1.57]:59746 "EHLO gate.crashing.org"
        rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
        id S1752273AbcHaVru (ORCPT <rfc822;linux-kernel@vger.kernel.org>);
        Wed, 31 Aug 2016 17:47:50 -0400
Message-ID: <1472680030.2388.95.camel@kernel.crashing.org>
Subject: Re: [RFC][PATCH] Fix a race between rwsem and the scheduler
From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Oleg Nesterov <oleg@redhat.com>, Balbir Singh <bsingharora@gmail.com>,
        LKML <linux-kernel@vger.kernel.org>,
        Nicholas Piggin <nicholas.piggin@gmail.com>
Date: Thu, 01 Sep 2016 07:47:10 +1000
In-Reply-To: <20160831133104.GD10153@twins.programming.kicks-ass.net>
References: <4050f2ce-1aee-d2aa-39e3-36e995b56252@gmail.com>
         <20160830121937.GQ10138@twins.programming.kicks-ass.net>
         <20160830130426.GA17795@redhat.com> <20160830141321.GB2794@worktop>
         <20160830165746.GA29218@redhat.com>
         <20160830183416.GV10138@twins.programming.kicks-ass.net>
         <1472592498.2388.40.camel@kernel.crashing.org>
         <20160831133104.GD10153@twins.programming.kicks-ass.net>
Content-Type: text/plain; charset="UTF-8"
X-Mailer: Evolution 3.20.5 (3.20.5-1.fc24) 
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Wed, 2016-08-31 at 15:31 +0200, Peter Zijlstra wrote:
> On Wed, Aug 31, 2016 at 07:28:18AM +1000, Benjamin Herrenschmidt
> wrote:
> 
> > 
> > On powerpc we have a sync deep in _switch to achieve that.
> 
> OK, for giggles, could you (or Balbir) check what happens if you take
> that sync out?
> 
> There should be enough serialization in the generic code to cover the
> case that code mentions.
> 
> ARM64 has a stronger barrier in its context switch code, but that's
> because they need to sync against external agents (like their TLB and
> cache) and no amount of generic locking is going to cover that.

The problem is no amount of testing can tell you it works for sure :-)

I would be nervous not having a real full sync in _switch. All we have
along the scheduler path is lwsync's and our isync based load construct
for spin_lock, I'm not sure what other assumptions we have around that
sync in there...

Cheers,
Ben.