Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
MIME-Version: 1.0
References: <20220430002555.3881-1-vvghjk1234@gmail.com> <49b0d611-e116-c78d-cf14-6d5f96ae500e@suse.cz>
In-Reply-To: <49b0d611-e116-c78d-cf14-6d5f96ae500e@suse.cz>
From:   Wonhyuk Yang <vvghjk1234@gmail.com>
Date:   Mon, 9 May 2022 18:42:40 +0900
Message-ID: <CAEcHRTo3keiv0Xjr_7CG-kiD+Y6ENbdf9fKZzW9zgRd9OrV03g@mail.gmail.com>
Subject: Re: [Patch v3] mm/slub: Remove repeated action in calculate_order()
To:     Vlastimil Babka <vbabka@suse.cz>
Cc:     Christoph Lameter <cl@linux.com>,
        Pekka Enberg <penberg@kernel.org>,
        David Rientjes <rientjes@google.com>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Hyeonggon Yoo <42.hyeyoo@gmail.com>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, May 2, 2022 at 7:00 PM Vlastimil Babka <vbabka@suse.cz> wrote:
>
> On 4/30/22 02:25, Wonhyuk Yang wrote:
> > To calculate order, calc_slab_order() is called repeatly changing the
> > fract_leftover. Thus, the branch which is not dependent on
> > fract_leftover is executed repeatly. So make it run only once.
> >
> > Plus, when min_object reached to 1, we set fract_leftover to 1. In
> > this case, we can calculate order by max(slub_min_order,
> > get_order(size)) instead of calling calc_slab_order().
> >
> > No functional impact expected.
> >
> > Signed-off-by: Wonhyuk Yang <vvghjk1234@gmail.com>
> > Reviewed-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>
> > ---
> >
> >  mm/slub.c | 18 +++++++-----------
> >  1 file changed, 7 insertions(+), 11 deletions(-)
> >
> > diff --git a/mm/slub.c b/mm/slub.c
> > index ed5c2c03a47a..1fe4d62b72b8 100644
> > --- a/mm/slub.c
> > +++ b/mm/slub.c
> > @@ -3795,9 +3795,6 @@ static inline unsigned int calc_slab_order(unsigned int size,
> >       unsigned int min_order = slub_min_order;
> >       unsigned int order;
> >
> > -     if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE)
> > -             return get_order(size * MAX_OBJS_PER_PAGE) - 1;
> > -
> >       for (order = max(min_order, (unsigned int)get_order(min_objects * size));
> >                       order <= max_order; order++) {
> >
> > @@ -3820,6 +3817,11 @@ static inline int calculate_order(unsigned int size)
> >       unsigned int max_objects;
> >       unsigned int nr_cpus;
> >
> > +     if (unlikely(order_objects(slub_min_order, size) > MAX_OBJS_PER_PAGE)) {
> > +             order = get_order(size * MAX_OBJS_PER_PAGE) - 1;
> > +             goto out;
> > +     }
>
> Hm interestingly, both before and after your patch, MAX_OBJS_PER_PAGE might
> be theoretically overflowed not by slub_min_order, but then with higher
> orders. Seems to be prevented only as a side-effect of fragmentation close
> to none, thus higher orders not attempted. Would be maybe less confusing to
> check that explicitly. Even if that's wasteful, but this is not really perf
> critical code.

Yes, I agree that checking the overflow of object number explicitly is
better even if
it is almost impossible. But it checked repeatedly by calling
calc_slab_order(). It
seems to me that is unnecessary doesn't it?

>
> > +
> >       /*
> >        * Attempt to find best configuration for a slab. This
> >        * works by first attempting to generate a layout with
> > @@ -3865,14 +3867,8 @@ static inline int calculate_order(unsigned int size)
> >        * We were unable to place multiple objects in a slab. Now
> >        * lets see if we can place a single object there.
> >        */
> > -     order = calc_slab_order(size, 1, slub_max_order, 1);
> > -     if (order <= slub_max_order)
> > -             return order;
> > -
> > -     /*
> > -      * Doh this slab cannot be placed using slub_max_order.
> > -      */
> > -     order = calc_slab_order(size, 1, MAX_ORDER, 1);
> > +     order = max_t(unsigned int, slub_min_order, get_order(size));
>
> If we failed to assign order above, then AFAICS it means even slub_min_order
> will not give us more than 1 object per slub. Thus it doesn't make sense to
> use it in a max() formula, and we can just se get_order(), no?

That's sounds reasonable. When it reached here, we don't need to keep the
slub_min_order.

>
> > +out:
> >       if (order < MAX_ORDER)
> >               return order;
> >       return -ENOSYS;
>