Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 4 Sep 2023 10:04:29 +0200
To:     Geert Uytterhoeven <geert@linux-m68k.org>
Cc:     Javier Martinez Canillas <javierm@redhat.com>,
        linux-kernel@vger.kernel.org,
        Thomas Zimmermann <tzimmermann@suse.de>,
        Daniel Vetter <daniel@ffwll.ch>,
        David Airlie <airlied@gmail.com>,
        dri-devel@lists.freedesktop.org
Subject: Re: [RFC PATCH] drm/ssd130x: Allocate buffer in the CRTC's
 .atomic_check() callback
Message-ID: <yvxmbpbeuis7zjqyg6yrpdfyr3oa2xstcoeb2prqvznczzhj5k@7i37gxyyqfn3>
From:   Maxime Ripard <mripard@kernel.org>
References: <20230830062546.720679-1-javierm@redhat.com>
 <zitno3p7tbnld5auedkx5g4wey2csng4ncmtdhzinbuhblunyk@chnwsnsgq36v>
 <CAMuHMdWv_QSatDgihr8=2SXHhvp=icNxumZcZOPwT9Q_QiogNQ@mail.gmail.com>
 <4zfgmvfstyjfo5slggfmfuvnirrhrq773el52gkav2r6jxliub@7qjbyy7rkj3g>
 <CAMuHMdV_775mPbTgWmzCo4mKCd3kqL=vfVFrt2W=bR3uveNW_Q@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable
In-Reply-To: <CAMuHMdV_775mPbTgWmzCo4mKCd3kqL=vfVFrt2W=bR3uveNW_Q@mail.gmail.com>
Precedence: bulk

On Fri, Sep 01, 2023 at 02:08:11PM +0200, Geert Uytterhoeven wrote:
> Hi Maxime,
>=20
> On Fri, Sep 1, 2023 at 2:00=E2=80=AFPM Maxime Ripard <mripard@kernel.org>=
 wrote:
> > On Fri, Sep 01, 2023 at 10:36:17AM +0200, Geert Uytterhoeven wrote:
> > > On Fri, Sep 1, 2023 at 10:22=E2=80=AFAM Maxime Ripard <mripard@kernel=
=2Eorg> wrote:
> > > > On Wed, Aug 30, 2023 at 08:25:08AM +0200, Javier Martinez Canillas =
wrote:
> > > > > The commit 45b58669e532 ("drm/ssd130x: Allocate buffer in the pla=
ne's
> > > > > .atomic_check() callback") moved the allocation of the intermedia=
te and
> > > > > HW buffers from the encoder's .atomic_enable callback to primary =
plane's
> > > > > .atomic_check callback.
> > > > >
> > > > > This was suggested by Maxime Ripard because drivers aren't allowe=
d to fail
> > > > > after drm_atomic_helper_swap_state() has been called, and the enc=
oder's
> > > > > .atomic_enable happens after the new atomic state has been swappe=
d.
> > > > >
> > > > > But that change caused a performance regression in very slow plat=
forms,
> > > > > since now the allocation happens for every plane's atomic state c=
ommit.
> > > > > For example, Geert Uytterhoeven reports that is the case on a Vex=
RiscV
> > > > > softcore (RISC-V CPU implementation on an FPGA).
> > > >
> > > > I'd like to have numbers on that. It's a bit surprising to me that,
> > > > given how many objects we already allocate during a commit, two sma=
ll
> > > > additional allocations affect performances so dramatically, even on=
 a
> > > > slow platform.
> > >
> > > To be fair, I didn't benchmark that.  Perhaps it's just too slow due =
to
> > > all these other allocations (and whatever else happens).
> > >
> > > I just find it extremely silly to allocate a buffer over and over aga=
in,
> > > while we know that buffer is needed for each and every display update.
> >
> > Maybe it's silly, but I guess it depends on what you want to optimize
> > for. You won't know the size of that buffer before you're in
> > atomic_check. So it's a different trade-off than you would like, but I
> > wouldn't call it extremely silly.
>=20
> The size of ssd130x_plane_state.data_array[] is fixed, and depends
> on the actual display connected.

That one can be tied to the CRTC state if needed. It would only be
allocated on each modeset, so probably once for that kind of device.

> The size of ssd130x_plane_state.buffer[]  is also fixed, and depends
> on the plane's size (which is currently fixed to the display size).

Doesn't it depend on the format as well?

Maxime