Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <HE1PR06MB4011EDD5F2686A05BC35F61CAC790@HE1PR06MB4011.eurprd06.prod.outlook.com>
 <20191106223408.2176-1-jonas@kwiboo.se> <HE1PR06MB4011FF930111A869E4645C8CAC790@HE1PR06MB4011.eurprd06.prod.outlook.com>
 <CAAFQd5CSWea=DNjySJxZmVi+2c5U4EKVPa1mf3vHh70+YrAQCA@mail.gmail.com>
 <7b92111b0c6443653de45f1eeec867645c127f32.camel@collabora.com>
 <CAAFQd5CZo5g2gtuvU+OuoRj18ZcCH8XEinGyAjRxqUXRfSQhgg@mail.gmail.com> <9606106c-5c49-cbc1-cb8f-0389ff8281bd@kwiboo.se>
In-Reply-To: <9606106c-5c49-cbc1-cb8f-0389ff8281bd@kwiboo.se>
From:   Tomasz Figa <tfiga@chromium.org>
Date:   Thu, 16 Jan 2020 12:56:02 +0900
Message-ID: <CAAFQd5DMVKH_f6nTohw_7LpLAYNWOv=fN77S_sLPPmYZwskosA@mail.gmail.com>
Subject: Re: [PATCH v3 2/5] media: hantro: Reduce H264 extra space for motion vectors
To:     Jonas Karlman <jonas@kwiboo.se>
Cc:     Ezequiel Garcia <ezequiel@collabora.com>,
        Mauro Carvalho Chehab <mchehab+samsung@kernel.org>,
        Hans Verkuil <hverkuil@xs4all.nl>,
        Boris Brezillon <boris.brezillon@collabora.com>,
        Philipp Zabel <p.zabel@pengutronix.de>,
        "linux-media@vger.kernel.org" <linux-media@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Jan 9, 2020 at 12:10 AM Jonas Karlman <jonas@kwiboo.se> wrote:
>
> On 2020-01-08 13:59, Tomasz Figa wrote:
> > On Tue, Dec 10, 2019 at 3:11 AM Ezequiel Garcia <ezequiel@collabora.com> wrote:
> >>
> >> On Wed, 2019-11-20 at 21:44 +0900, Tomasz Figa wrote:
> >>> Hi Jonas,
> >>>
> >>> On Thu, Nov 7, 2019 at 7:34 AM Jonas Karlman <jonas@kwiboo.se> wrote:
> >>>> A decoded 8-bit 4:2:0 frame need memory for up to 448 bytes per
> >>>> macroblock with additional 32 bytes on multi-core variants.
> >>>>
> >>>> Memory layout is as follow:
> >>>>
> >>>> +---------------------------+
> >>>>> Y-plane   256 bytes x MBs |
> >>>> +---------------------------+
> >>>>> UV-plane  128 bytes x MBs |
> >>>> +---------------------------+
> >>>>> MV buffer  64 bytes x MBs |
> >>>> +---------------------------+
> >>>>> MC sync          32 bytes |
> >>>> +---------------------------+
> >>>>
> >>>> Reduce the extra space allocated now that motion vector buffer offset no
> >>>> longer is based on the extra space.
> >>>>
> >>>> Only allocate extra space for 64 bytes x MBs of motion vector buffer
> >>>> and 32 bytes for multi-core sync.
> >>>>
> >>>> Fixes: a9471e25629b ("media: hantro: Add core bits to support H264 decoding")
> >>>> Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
> >>>> Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
> >>>> ---
> >>>> Changes in v3:
> >>>>   - add memory layout to code comment (Boris)
> >>>> Changes in v2:
> >>>>   - updated commit message
> >>>> ---
> >>>>  drivers/staging/media/hantro/hantro_v4l2.c | 20 ++++++++++++++++++--
> >>>>  1 file changed, 18 insertions(+), 2 deletions(-)
> >>>>
> >>>
> >>> Thanks for the patch!
> >>>
> >>> What platform did you test it on and how? Was it tested with IOMMU enabled?
> >>
> >> Hello Tomasz,
> >>
> >> Please note that this series has been picked-up and is merged
> >> in v5.5-rc1.
> >>
> >> IIRC, we tested these patches on RK3399 and RK3288 (that means
> >> with an IOMMU). I've just ran some more extensive tests on RK3288,
> >> on media/master; and I plan to test some more on RK3399 later this week.
> >>
> >> Do you have any specific concern in mind?
> >
> > I specifically want to know whether we're 100% sure that those sizes
> > are correct. The IOMMU still works on page granularity so it's
> > possible that the allocation could be just big enough by luck.
>
> One of my RK3288 TRM [1] contains the following:
>
> Direct mode motion vector write:
>   Its base addr is right after decode output picture data
>   Its length is mbwidth*mbheight*64
>
> Also both the mpp library and imx-vpu-hantro code both use mbwidth*mbheight*64.
> So I feel confident that the buffer size is correct.
>
> [1] Rockchip RK3288TRM V1.1 Part3-Graphic and multi-media.pdf

Fair enough. Thanks!

Best regards,
Tomasz