Date: Tue, 8 May 2007 12:14:04 -0700
From: Randy Dunlap <randy.dunlap@oracle.com>
To: Andrew Morton <akpm@linux-foundation.org>
Cc: Paul Sokolovsky <pmiscml@gmail.com>, linux-kernel@vger.kernel.org
Subject: [RFC/PATCH] doc: volatile considered evil
Message-Id: <20070508121404.17bd97a6.randy.dunlap@oracle.com>
In-Reply-To: <20070430235642.e576e917.akpm@linux-foundation.org>
References: <516386418.20070501080839@gmail.com>
	<20070430235642.e576e917.akpm@linux-foundation.org>
Organization: Oracle Linux Eng.
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 6168
Lines: 154

On Mon, 30 Apr 2007 23:56:42 -0700 Andrew Morton wrote:

> Oh my eyes.  What are these doing?
> 
> The volatiles are a worry - volatile is said to be basically-always-wrong
> in-kernel, although we've never managed to document why, and i386
> cheerfully uses it in readb() and friends.
> 
> Perhaps if you can describe presisely what's going on here, alternatives
> might be suggested.

[well, can be turned into a patch]

Here are some 'volatile' comments from Linus, extracted from
several emails in at least 2 threads.

If this is close to useful, we can add it to Documentation/.

===============================================================

***** "volatile" considered useless and evil:  Just Say NO! *****

Do not use the C-language "volatile" keyword
(extracted from lkml emails from Linus)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

[Comment about a patch:]

> Also made all the relevant mce_log fields volatile for further safety.

I refuse to apply this part.

If the memory barriers are right, then the "volatile" doesn't matter.

And if the memory barriers aren't right, then "volatile" doesn't help.

Using "volatile" in data structures is basically _always_ a bug.

The only acceptable uses for "volatile" are:

 - in _code_, i.e., for things like the definition of "readb()" etc, where we
   use it to force a particular access.
 - with inline asms
 - on "jiffies", for stupid legacy reasons

Basically, a volatile on a data structure can NEVER be right. If it makes
a difference, it's a sign of improper locking.

And the reason I refuse to apply that part of the patch is that anybody
who even _thinks_ that they make a difference is horribly and utterly
confused, and doesn't understand locking.

So please _never_ use them like this.


> mce_log is fully lockless - it deals with machine checks which act like NMIs.
> That is it's problem.
>
> In theory the memory barriers should be sufficient, the volatiles are
> just an additional safety net to make it clear to humans/compiler these memory
> areas can change any time.

If the memory barriers aren't sufficient, the volatiles are useless. If
the memory barriers _are_ sufficient, the volatiles are useless.

See? They're useless.

The only thing they do is

 - potentially make the compiler generate worse code for no reason (the
   "no reason" being that if there aren't any barriers in between, the
   compiler _should_ merge accesses)

 - make some people _believe_ that that compiler does something "right".

The first point doesn't much matter. The second point matters a LOT.

Anybody who thinks "volatile" matters is WRONG. As such, a "volatile" is
anti-documentation - it makes people think that something is true that is
NOT true.

In other words, volatile on data structures is _evil_, because it instills
the wrong kind of beliefs in people. "volatility" is not a data structure
issue. It's a matter of the _code_ working on the data structure.


"volatile" really _is_ misdesigned. The semantics of it are so unclear as
to be totally useless. The only thing "volatile" can ever do is generate
worse code, WITH NO UPSIDES.

Historically (and from the standpoint of the C standard), the definition
of "volatile" is that any access is "visible" in the machine, and it
really kind of makes sense for hardware accesses, except these days
hardware accesses have other rules that are _not_ covered by "volatile",
so you can't actually use them for that.

And for accesses that have some software rules (i.e., not IO devices etc),
the rules for "volatile" are too vague to be useful.

So if you actually have rules about how to access a particular piece of
memory, just make those rules _explicit_. Use the real rules. Not
volatile, because volatile will always do the wrong thing.

Also, more importantly, "volatile" is on the wrong _part_ of the whole
system. In C, it's "data" that is volatile, but that is insane. Data
isn't volatile - _accesses_ are volatile. So it may make sense to say
"make this particular _access_ be careful", but not "make all accesses to
this data use some random strategy".

So the only thing "volatile" is potentially useful for is:

 - actual accessor functions can use it in a _cast_ to make one particular
   access follow the rules of "don't cache this one dereference". That is
   useful as part of a _bigger_ set of rules about that access (i.e., it
   might be the internal implementation of a "readb()", for example).

 - for "random number generation" data locations, where you literally
   don't _have_ any rules except "it's a random number". The only really
   valid example of this is the "jiffy" timer tick.

Any other use of "volatile" is almost certainly a bug, or just useless.

Side note: it's also totally possible that a volatiles _hides_ a bug, i.e.,
removing the volatile ends up having bad effects, but that's because the
software itself isn't actually following the rules (or, more commonly, the
rules are broken, and somebody added "volatile" to hide the problem).

It's a bug if the volatile means that you don't follow the proper protocol
for accessing the data, and it's useless (and generally generates worse
code) if you already do.

So just say NO! to volatile except under the above circumstances.


[from another email thread:]

I suspect we should just face up to the fact that:

 (a) "volatile" on kernel data is basically always a bug, and you should
     use locking. "volatile" doesn't help anything at all with memory
     ordering and friends, so it's insane to think it "solves" anything on
     its own.
 (b) on "iomem" pointers it does make sense, but those need special
     accessor functions _anyway_, so things like test_bit() wouldn't work
     on them.
 (c) if you spin on a value [that's] changing, you should use "cpu_relax()" or
     "barrier()" anyway, which will force gcc to re-load any values from
     memory over the loop.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/