Re: [rfc] SLOB memory ordering issue

[Linus Torvalds]

On Thu, 16 Oct 2008, Nick Piggin wrote:
>
> Actually, there are surprisingly huge number of them. What I would be
> most comfortable doing, if I was making a kernel to run my life support
> system on an SMP powerpc box, would be to spend zero time on all the
> drivers and whacky things with ctors and just add smp_wmb() after them
> if they are not _totally_ obvious.

WHY?

THIS HAS NOTHING TO DO WITH CONSTRUCTORS!

If the driver is using locking, there is no memory ordering issues 
what-so-ever.

And if the driver isn't using locking, IT IS BROKEN.

It's that simple. Why do you keep bringing up non-issues?

What matters is not constructors. Never has been. Constructors are 
actually very rare, it's much more common to do

	ptr = kmalloc(..)
	.. initialize it by hand ..

and why do you think constructors are somehow different? They're not.

What matter is how you look things up on the other CPU's. If you don't use 
locking, you use some lockless thing, and then you need to be careful 
about memory ordering.

And quite frankly, if you're a driver, and you're trying to do lockless 
algorithms, you're just being crazy. You're going to have much worse bugs, 
and again, whether you use constructors or pink elephants is going to be 
totally irrelevant. 

So why do you bring up these totally pointless things? Why do you bring up 
drivers? Why do you bring up constructors? Why, why, why?

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