[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAHmME9ri2YHN9OEen2sw2TimNZO5Swca41qumxu_LmbqUztEKA@mail.gmail.com>
Date: Mon, 26 Sep 2022 23:33:14 +0200
From: "Jason A. Donenfeld" <Jason@...c4.com>
To: Kees Cook <keescook@...omium.org>
Cc: Andy Shevchenko <andriy.shevchenko@...ux.intel.com>,
linux-kernel@...r.kernel.org,
Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [PATCH v3 2/2] minmax: clamp more efficiently by avoiding extra comparison
On Mon, Sep 26, 2022 at 8:30 PM Kees Cook <keescook@...omium.org> wrote:
>
> On Mon, Sep 26, 2022 at 03:34:35PM +0200, Jason A. Donenfeld wrote:
> > [...]
> > In this case, we actually gain a branch, unfortunately, because the
> > compiler's replacement axioms no longer as cleanly apply.
> >
> > So all and all, this change is a bit of a mixed bag.
>
> I'm on the fence -- I think the new macro is a more correct way to
> describe the operation, though on the other hand, the old way provides a
> simple way to compose the bounds checks.
>
> I suspect we should probably optimize for _performance_, not code size,
> so if the new branch is actually visible via cycle counts in "perf"
> output, probably we shouldn't use this patch, and instead add a comment
> about why it is defined the way it is.
I *want* the better algorithm to yield better performance, because
that's a much less confusing world. But it seems like we have grounds
for suspecting that might not be the case. So until I come up with
some real measurements, I agree we should hold off on this 2/2.
Jason
Powered by blists - more mailing lists