Re: + atomic-improve-atomic_inc_unless_negative-atomic_dec_unless_positive .patch added to -mm tree

[Oleg Nesterov <oleg@redhat.com>]

On 03/21, Eric Dumazet wrote:
>
> On Thu, 2013-03-21 at 19:30 +0100, Oleg Nesterov wrote:
>
> > To me, it would be better to kill atomic_inc_not_zero_hint() or unify
> > unify it with atomic_inc_not_zero(). But this is another story.
>
> git is your friend.
>
> I suggest you read 3f9d35b9514 changelog before killing it, thanks.

Thanks Eric for your friendly suggestion.

But I didn't mean we should kill this optimization. Yes, I am wondering
if we can avoid inc_not_zero_hint _or_ unify with add_unless. But let me
repeat, this is another story.


Perhaps I misread your previous email... I understood it as if you think
the patch I sent is wrong. No?

If you meant that get_write_access() can predict the current value of
i_writecount... how? And even if we could, why we cant/shouldnt try to
optimize the generic atomic_inc_unless_negative()?

So what did you actually mean?

Oleg.