Re: [PATCH v2] introduce atomic_pointer to fix a race condition in cancelable mcs spinlocks

From: Linus Torvalds
Date: Mon Jun 02 2014 - 16:53:55 EST


On Mon, Jun 2, 2014 at 1:46 PM, Mikulas Patocka <mpatocka@xxxxxxxxxx> wrote:
>
> And what else do you want to do?
>
> Peter Zijlstra said "I've been using xchg() and cmpxchg() without such
> consideration for quite a while." - so it basically implies that the
> kernel is full of such races, mcs_spinlock is just the most visible one
> that crashes the kernel first.

.. so your whole argument is bogus, because it doesn't actually fix
anything else.

Now, something that *would* fix something else is (for example) to
just make "ACCESS_ONCE()" a rvalue so that you cannot use it for
assignments, and then trying to sort out what happens then. It's
possible that the "atomic_pointer_t" would be a part of the solution
to that "what happens then", but THERE IS NO WAY IN HELL we're adding
it for just one architecture and one use that doesn't warrant even
_existing_ on that architecture.

See what I'm saying?

You're not fixing the problem, you're fixing one unimportant detail
that isn't worth fixing that way.

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/