Date:	Thu, 5 Dec 2013 10:02:00 -0800
From:	"Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>
To:	Ingo Molnar <mingo@...nel.org>
Cc:	linux-kernel@...r.kernel.org, laijs@...fujitsu.com,
	dipankar@...ibm.com, akpm@...ux-foundation.org,
	mathieu.desnoyers@...icios.com, josh@...htriplett.org,
	niv@...ibm.com, tglx@...utronix.de, peterz@...radead.org,
	rostedt@...dmis.org, dhowells@...hat.com, edumazet@...gle.com,
	darren@...art.com, fweisbec@...il.com, sbw@....edu,
	Oleg Nesterov <oleg@...hat.com>,
	Jonathan Corbet <corbet@....net>,
	Rusty Russell <rusty@...tcorp.com.au>
Subject: Re: [PATCH tip/core/locking 4/4] Documentation/memory-barriers.txt:
 Document ACCESS_ONCE()

On Thu, Dec 05, 2013 at 10:33:34AM +0100, Ingo Molnar wrote:
> 
> * Paul E. McKenney <paulmck@...ux.vnet.ibm.com> wrote:
> 
> > + (*) The compiler is within its rights to reorder memory accesses unless
> > +     you tell it not to.  For example, consider the following interaction
> > +     between process-level code and an interrupt handler:
> > +
> > +	void process_level(void)
> > +	{
> > +		msg = get_message();
> > +		flag = true;
> > +	}
> > +
> > +	void interrupt_handler(void)
> > +	{
> > +		if (flag)
> > +			process_message(msg);
> > +	}
> > +
> > +     There is nothing to prevent the the compiler from transforming
> > +     process_level() to the following, in fact, this might well be a
> > +     win for single-threaded code:
> > +
> > +	void process_level(void)
> > +	{
> > +		flag = true;
> > +		msg = get_message();
> > +	}
> > +
> > +     If the interrupt occurs between these two statement, then
> > +     interrupt_handler() might be passed a garbled msg.  Use ACCESS_ONCE()
> > +     to prevent this as follows:
> > +
> > +	void process_level(void)
> > +	{
> > +		ACCESS_ONCE(msg) = get_message();
> > +		ACCESS_ONCE(flag) = true;
> > +	}
> > +
> > +	void interrupt_handler(void)
> > +	{
> > +		if (ACCESS_ONCE(flag))
> > +			process_message(ACCESS_ONCE(msg));
> > +	}
> 
> Technically, if the interrupt handler is the innermost context, the 
> ACCESS_ONCE() is not needed in the interrupt_handler() code.
> 
> Since for the vast majority of Linux code IRQ handlers are the most 
> atomic contexts (very few drivers deal with NMIs) I suspect we should 
> either remove that ACCESS_ONCE() from the example or add a comment 
> explaining that in many cases those are superfluous?

How about the following additional paragraph?

     Note that the ACCESS_ONCE() wrappers in interrupt_handler()
     are needed if this interrupt handler can itself be interrupted
     by something that also accesses 'flag' and 'msg', for example,
     a nested interrupt or an NMI.  Otherwise, ACCESS_ONCE() is not
     needed in interrupt_handler() other than for documentation purposes.

> > + (*) For aligned memory locations whose size allows them to be accessed
> > +     with a single memory-reference instruction, prevents "load tearing"
> > +     and "store tearing," in which a single large access is replaced by
> > +     multiple smaller accesses.  For example, given an architecture having
> > +     16-bit store instructions with 7-bit immediate fields, the compiler
> > +     might be tempted to use two 16-bit store-immediate instructions to
> > +     implement the following 32-bit store:
> > +
> > +	p = 0x00010002;
> > +
> > +     Please note that GCC really does use this sort of optimization,
> > +     which is not surprising given that it would likely take more
> > +     than two instructions to build the constant and then store it.
> > +     This optimization can therefore be a win in single-threaded code.
> > +     In fact, a recent bug (since fixed) caused GCC to incorrectly use
> > +     this optimization in a volatile store.  In the absence of such bugs,
> > +     use of ACCESS_ONCE() prevents store tearing:
> > +
> > +	ACCESS_ONCE(p) = 0x00010002;
> 
> I suspect the last sentence should read:
> 
> > +                                             In the absence of such bugs,
> > +     use of ACCESS_ONCE() prevents store tearing in this example:
> > +
> > +	ACCESS_ONCE(p) = 0x00010002;
> 
> Otherwise it could be read as a more generic statement (leaving out 
> 'load tearing')?

Good point, fixed.

Indeed, I don't have a good example for load tearing.  I do have some -bad-
examples, like the following:

	struct __attribute__((__packed__)) foo {
		short a;
		int b;
		short c;
	};
	struct foo foov;
	short aa;
	int bb;
	short cc;

	...

	aa = foov.a;
	bb = foov.b;
	cc = foov.c;

A clever compiler might choose to pack aa, bb, and cc in memory, then
implement the three assignments using two 32-bit loads and two 32-bit
stores, which would result in load tearing of foov.b.

Hmmm...  Maybe I should give this example anyway, just to show that
load tearing really could occur in practice...  If nothing else, it
should be a cautionary tale for those tempted to pack their structures.
And there are quite a number of packed structures in the Linux kernel.

Sold!  I have added this example, but using a pair of struct foo variables
in order to forestall maidenly protests from those who believe that no
production-quality compiler would ever misalign variable bb.  ;-)

							Thanx, Paul

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

Powered by blists - more mailing lists