Re: [linus:master] [fsnotify] a94204f4d4: stress-ng.timerfd.ops_per_sec 7.0% improvement

[Amir Goldstein]

On Thu, Feb 6, 2025 at 9:45 PM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, 5 Feb 2025 at 00:09, kernel test robot <oliver.sang@xxxxxxxxx> wrote:
> >
> > kernel test robot noticed a 7.0% improvement of stress-ng.timerfd.ops_per_sec on:
>
> I have no idea what the heck that benchmark does, but I am happy to
> see that this whole patch series did actually end up improving on the
> whole fsnotify cost that I was ranting about in the original
> submission.
>
> Obviously this load doesn't actually do the new pre-event hooks, so it
> seems to be all just about how fsnotify_file_area_perm() (or some
> other fsnotify hook in the normal read() path) now isn't the
> unconditional pig that it used to be.

Right.
We got this win by slightly modifying semantics of FAN_ACCESS_PERM
events, but I cannot imagine that anyone will care (famous last words).

It's all about the fact that
>
> I don't see how that would be worth 7%, but I'll happily take it.
>

Yeh, I was happy to see that as well :)

Thanks,
Amir.