Re: [PATCH 08/14] hrtimer: Allow hrtimer::function() to free the timer

From: Peter Zijlstra <peterz@infradead.org>
To: Oleg Nesterov <oleg@redhat.com>
Cc: Kirill Tkhai <ktkhai@odin.com>,
	umgwanakikbuti@gmail.com, mingo@elte.hu, ktkhai@parallels.com,
	rostedt@goodmis.org, tglx@linutronix.de, juri.lelli@gmail.com,
	pang.xunlei@linaro.org, wanpeng.li@linux.intel.com,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH 08/14] hrtimer: Allow hrtimer::function() to free the timer
Date: Thu, 11 Jun 2015 09:31:28 +0200	[thread overview]
Message-ID: <20150611073128.GQ19282@twins.programming.kicks-ass.net> (raw)
In-Reply-To: <20150610160444.GB3341@redhat.com>

On Wed, Jun 10, 2015 at 06:04:44PM +0200, Oleg Nesterov wrote:
> If we rely on the "locking" semantics of seqcount_t, this doesn't really
> differ from
> 
> 	spinlock_t LOCK;
> 	bool X = true, Y = false;
> 
> 	void read(void)
> 	{
> 		bool x, y;
> 
> 		spin_lock(LOCK);
> 		x = X; y = Y;
> 		spin_unlock(LOCK);
> 
> 		BUG_ON(!x && !y);
> 	}
> 
> 	void write(void)
> 	{
> 		Y = true;
> 
> 		spin_lock(LOCK);
> 		spin_unlock(LOCK);
> 
> 		X = false;
> 	}

So when I first saw that I went, wait what?

Because I looked at it like:

	Y = true;
	ACQUIRE
	RELEASE
	X = false;

And we both know that is not ordered at all.

But its the actual lock semantics that make it work, we cannot acquire
until the other completes, at which time the acquire matches its
release and things end up ordered again.

And its exactly that part we lost in our transformation C yesterday.

In any case, breakfast time, will go do the patches shortly.