Re: >Re: [RFC] should VM_BUG_ON(cond) really evaluate cond

[Linus Torvalds <torvalds@linux-foundation.org>]

On Fri, Oct 28, 2011 at 7:47 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
> So we need a similar idea to remove the volatile from :
>
> static __always_inline int constant_test_bit(unsigned int nr, const volatile unsigned long *addr)
> {
> return ((1UL << (nr % BITS_PER_LONG)) &
>         (addr[nr / BITS_PER_LONG])) != 0;

Ho humm. Yeah, I think "test_bit()" falls under the same logic: we do
not want to combine multiple test-bits into one (because we know there
are people looping on it), and we do want to guarantee "access at most
once" semantics, but as with "atomic_read()" we should be able to
optimize away a totally unused test-bit.

The same "empty asm" trick would work there, I think, rather than
using "volatile" (well, the function declaration would still have
"volatile" because it's legal to use a volatile data type, but we'd
cast it away and use the asm trick for the access instead) .

Maybe we should make it a generic helper macro ("ACCESS_AT_MOST_ONCE()")?

Comments? I think I'm open to tested patches..

                      Linus