Re: acl_permission_check: disgusting performance

[Linus Torvalds]

On Fri, May 13, 2011 at 9:16 AM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> Looks ok to me. And generates good code for acl_permission_check
> without CONFIG_USER_NS.
>
> I'll see how much that function drops on the kernel profiles..

Yup, looking good.

For my "kernel make with no changes" workload, it dropped from

    1.28%           make  [kernel.kallsyms]             [k] acl_permission_check

to

     0.88%           make  [kernel.kallsyms]             [k]
acl_permission_check

which is pretty much exactly the expected 30% drop from no longer
having that expensive load of user_ns.

Of course, that 30% improvement is just a 0.4% performance improvement
in the big picture, but hey, almost half a percentage point on a real
load from just one single function in the kernel is definitely worth
doing.

Do you want to carry this for 2.6.40, or should I just apply it?

                                Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/