On Sun, 18 Dec 2005, James Bottomley wrote:
>
> Actually, I don't think you want might_sleep(): there are a few cases
> where we do an up() from under a spinlock, which will spuriously trigger
> this. I'd suggest WARN_ON(in_interrupt()) instead.
Ahh, good point. Yes.
However, if even the arm people aren't all that interested in this, maybe
it simply doesn't matter. A lot of other architectures either have
"decrement in memory" or can already use ll/sc for it.
(of course, on some architectures, ll/sc is really really slow, so they
might well prefer using a normal load and store instead).
Linus
