Re: [PATCH 0/3] 32-bit Meson: add the canvas module

[Martin Blumenstingl]

Hi Maxime,

On Mon, May 20, 2019 at 9:21 PM Maxime Jourdan <maxi.jourdan@xxxxxxxxxx> wrote:
>
> Hey Martin, so sorry for forgetting about this.
>
> On Thu, Apr 18, 2019 at 9:50 PM Martin Blumenstingl
> <martin.blumenstingl@xxxxxxxxxxxxxx> wrote:
> >
> > Hi Maxime,
> >
> > On Sat, Apr 13, 2019 at 8:54 PM Maxime Jourdan <maxi.jourdan@xxxxxxxxxx> wrote:
> > >
> > > Hi Martin,
> > > On Sat, Apr 13, 2019 at 7:15 PM Martin Blumenstingl
> > > <martin.blumenstingl@xxxxxxxxxxxxxx> wrote:
> > > >
> > > > This adds the canvas module on Meson8, Meson8b and Meson8m2. The canvas
> > > > IP is used by the video decoder hardware as well as the VPU (video
> > > > output) hardware.
> > > >
> > > > Neither the VPU nor the video decoder driver support the 32-bit SoCs
> > > > yet. However, we can still add the canvas module to have it available
> > > > once these drivers gain support for the older SoCs.
> > > >
> > > > I have tested this on my Meson8m2 board by hacking the VPU driver to
> > > > not re-initialize the VPU (and to use the configuration set by u-boot).
> > > > With that hack I could get some image out of the CVBS connector. No
> > > > changes to the canvas driver were required.
> > > >
> > > > Due to lack of hardware I could not test Meson8, but I'm following (as
> > > > always) what the Amlogic 3.10 vendor kernel uses.
> > > > Meson8b is also not tested because u-boot of my EC-100 doesn't have
> > > > video output enabled (so I couldn't use the same hack I used on my
> > > > Meson8m2 board).
> > > >
> > > > This series meant to be applied on top of "Meson8b: add support for the
> > > > RTC on EC-100 and Odroid-C1" from [0]
> > > >
> > > >
> > >
> > > The series looks good to me, however I wonder if we should maybe add a
> > > new compatible ?
> > >
> > > The canvas IP before the GX* generation does not handle what Amlogic
> > > calls "endianness", the field that allows doing some byte-switching to
> > > get proper NV12/NV21. So the following defines are unusable:
> > >
> > > #define MESON_CANVAS_ENDIAN_SWAP16 0x1
> > > #define MESON_CANVAS_ENDIAN_SWAP32 0x3
> > > #define MESON_CANVAS_ENDIAN_SWAP64 0x7
> > > #define MESON_CANVAS_ENDIAN_SWAP128 0xf
> > I didn't know about this - thank you for pointing this out.
> >
> > your suggestions to add new compatible strings is a good idea for that case.
> > Amlogic uses different defines for Meson8 and Meson8m2 in their vendor
> > kernel and they keep Meson8b different.
> > I will add three new compatibles, one for each SoC (Meson8, Meson8b,
> > Meson8m2) just to be on the safe side if we discover differences in
> > the canvas IP on these SoCs.
> >
> > what do you think?
> >
>
> Sure thing. Keep an eye out for any hints regarding the amount of
> canvases as well, I *think* I remember some old SoCs having only 192
> but I haven't been able to find it again.
Meson6 and older are limited to 192, Meson8 and newer already support
256. source: [0]

> > > It wouldn't change much functionally, but we could have e.g a warning
> > > if a m8 canvas user tries to set endianness even though it does
> > > nothing.
> > this is a good idea, that will make it easier to spot why something
> > doesn't work.
> > we can also return -EINVAL, like you already do for the case where the
> > canvas ID is already used.
> >
>
> Yes, returning an error is a good idea.
OK, I'll send an updated series later.


Martin


[0] https://github.com/endlessm/linux-meson/blob/5cb4882cdda584878a29132aeb9a90497a121df9/drivers/amlogic/canvas/canvas.c#L41