Date: Wed, 1 Mar 2017 16:26:48 +0000
From: Chris Wilson <chris@chris-wilson.co.uk>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Fengguang Wu <fengguang.wu@intel.com>,
        Boqun Feng <boqun.feng@gmail.com>,
        Nicolai =?iso-8859-1?Q?H=E4hnle?= <Nicolai.Haehnle@amd.com>,
        Ingo Molnar <mingo@kernel.org>, linux-kernel@vger.kernel.org,
        LKP <lkp@01.org>
Subject: Re: [locking/ww_mutex] 2a0c112828 WARNING: CPU: 0 PID: 18 at
 kernel/locking/mutex.c:305 __ww_mutex_wakeup_for_backoff
Message-ID: <20170301162648.GF4740@nuc-i3427.alporthouse.com>
References: <20170227051409.zbqwtekoa3hvggta@wfg-t540p.sh.intel.com>
 <20170227102824.GV6500@twins.programming.kicks-ass.net>
 <20170227103543.GA6536@twins.programming.kicks-ass.net>
 <20170301150138.hdixnmafzfsox7nn@tardis.cn.ibm.com>
 <20170301154043.hcsbpgooc3kqt45j@wfg-t540p.sh.intel.com>
 <20170301155414.GN6515@twins.programming.kicks-ass.net>
 <20170301161148.GD4740@nuc-i3427.alporthouse.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20170301161148.GD4740@nuc-i3427.alporthouse.com>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1440
Lines: 31

On Wed, Mar 01, 2017 at 04:11:48PM +0000, Chris Wilson wrote:
> On Wed, Mar 01, 2017 at 04:54:14PM +0100, Peter Zijlstra wrote:
> > On Wed, Mar 01, 2017 at 11:40:43PM +0800, Fengguang Wu wrote:
> > > Thanks for the patch! I applied the patch on top of "locking/ww_mutex:
> > > Add kselftests for ww_mutex stress", and find no "bad unlock balance
> > > detected" but this warning. Attached is the new dmesg which is a bit
> > > large due to lots of repeated errors.
> > 
> > So with all the various patches it works for me.
> > 
> > I also have the following on top; which I did when I was looking through
> > this code trying to figure out wth was happening.
> > 
> > Chris, does this make sense to you?
> > 
> > It makes each loop a fully new 'instance', otherwise we'll never update
> > the ww_class->stamp and the threads will aways have the same order.
> 
> Sounds ok, I just thought the stamp order of the threads was
> immaterial - with each test doing a different sequence of locks and each
> being identical in behaviour, it would not matter which had priority,
> there would have be some shuffling no matter waht. However, for the
> purpose of testing, having each iteration be a new locking instance does
> make it behaviour more like a typical user.

Correcting myself, the workers didn't reorder the locks, so changing the
stamp does make the test more interesting.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre