Re: + kthread-add-a-missing-memory-barrier-to-kthread_stop.patch added to -mm tree

["Dmitry Adamushko" <dmitry.adamushko@gmail.com>]

On 23/02/2008, Linus Torvalds <torvalds@linux-foundation.org> wrote:
>
>  On Sat, 23 Feb 2008, Dmitry Adamushko wrote:
>  >
>  > it's not a LOAD that escapes *out* of the region. It's a MODIFY that gets *in*:
>
>
> Not with the smp_wmb(). That's the whole point.
>
>  Ie the patch I'm suggesting is sufficient is appended, and the point is
>  that any write before the critical region will be ordered wrt the critical
>  region because of the write barrier before the spinlock (which itself is a
>  write).

Yeah, good point!

(heh... I wouldn't dare to say this 'obvious thing' only to Anton
Blanchard who is "the only person who always 'have a point' by
definition" :-))

> This is also why I mentioned that if you have a really odd architecure
> that considers spinlocks to be "outside" the normal cache coherency
> domain, that would be broken, but I cannot think of a single valid case of
> that, and I consider it insane.

Yeah, some potential implementations come into my mind but, I guess,
they are as far away from real hardware as science-fiction from
science :-/

So how should we proceed with this issue?

let's use your patch and declare try_to_wake_up() a 'full' mb for the case:

MODIFY
try_to_wake_up
LOAD or MODIFY (that take place either after or inside try_to_wake_up())

so we'll fix (lots of) potentially problematic cases with a single shot.

and

LOAD
try_to_wake_up()
LOAD or MODIFY

is probably not that common so we don't care.


-- 
Best regards,
Dmitry Adamushko