Re: [PATCH v2 3/3] arm64: dts: rockchip: Add Hantro encoder node to rk356x

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Donnerstag, 12. Mai 2022 16:16:52 CEST Ezequiel Garcia wrote:
> On Tue, May 10, 2022 at 12:28 PM Nicolas Frattaroli
> <frattaroli.nicolas@xxxxxxxxx> wrote:
> >
> > Hi Ezequiel,
> >
> > On Montag, 9. Mai 2022 16:17:03 CEST Ezequiel Garcia wrote:
> > > Hi Nicolas,
> > >
> > > On Sun, May 8, 2022 at 5:26 PM Nicolas Frattaroli
> > > <frattaroli.nicolas@xxxxxxxxx> wrote:
> > > >
> > > > The RK3566 and RK3568 come with a dedicated Hantro instance solely for
> > > > encoding. This patch adds a node for this to the device tree, along with
> > > > a node for its MMU.
> > > >
> > > > Signed-off-by: Nicolas Frattaroli <frattaroli.nicolas@xxxxxxxxx>
> > > > ---
> > > >  arch/arm64/boot/dts/rockchip/rk356x.dtsi | 21 +++++++++++++++++++++
> > > >  1 file changed, 21 insertions(+)
> > > >
> > > > diff --git a/arch/arm64/boot/dts/rockchip/rk356x.dtsi b/arch/arm64/boot/dts/rockchip/rk356x.dtsi
> > > > index 7cdef800cb3c..2e3c9e1887e3 100644
> > > > --- a/arch/arm64/boot/dts/rockchip/rk356x.dtsi
> > > > +++ b/arch/arm64/boot/dts/rockchip/rk356x.dtsi
> > > > @@ -508,6 +508,27 @@ gpu: gpu@fde60000 {
> > > >                 status = "disabled";
> > > >         };
> > > >
> > > > +       vepu: video-codec@fdee0000 {
> > > > +               compatible = "rockchip,rk3568-vepu";
> > > > +               reg = <0x0 0xfdee0000 0x0 0x800>;
> > > > +               interrupts = <GIC_SPI 64 IRQ_TYPE_LEVEL_HIGH>;
> > > > +               interrupt-names = "vepu";
> > >
> > > It this block "encoder only" and if so, maybe we should remove the
> > > "interrupt-names" [1]?
> > >
> > > The driver is able to handle it. See:
> > >
> > > https://elixir.bootlin.com/linux/latest/source/drivers/staging/media/hantro/hantro_drv.c#L962
> > >
> > > You might have to adjust the dt-bindings for this.
> > >
> > > [1] https://lore.kernel.org/linux-media/20210324151715.GA3070006@xxxxxxxxxxxxxxxxxx/
> >
> > What the Linux driver can handle should not matter to the device tree;
> > device trees are independent of drivers and kernels.
> >
> 
> I guess my message wasn't clear, no need to lecture me on Device
> Trees, although I appreciate
> your friendly reminder of what a Device Tree is.
> 
> Having said that, the binding is designed to support both decoders and encoders
> for instance:
> 
>         vpu: video-codec@ff9a0000 {
>                 compatible = "rockchip,rk3288-vpu";
>                 reg = <0x0 0xff9a0000 0x0 0x800>;
>                 interrupts = <GIC_SPI 9 IRQ_TYPE_LEVEL_HIGH>,
>                              <GIC_SPI 10 IRQ_TYPE_LEVEL_HIGH>;
>                 interrupt-names = "vepu", "vdpu";
>                 clocks = <&cru ACLK_VCODEC>, <&cru HCLK_VCODEC>;
>                 clock-names = "aclk", "hclk";
>                 iommus = <&vpu_mmu>;
>                 power-domains = <&power RK3288_PD_VIDEO>;
>         };
> 
> Hence the question is why do you splitted the encoder to its own node?

It has its own IOMMU and is in a different power domain than the decoder.
I think I have mentioned this multiple times before, including in the
cover letter.

Assuming you do not believe me, feel free to check the TRM, of which I
am sure you also have a copy: page 475 of Part 1 shows the VPU being in
PD_VPU while the JPEG encoder is in PD_RGA. Pages 478 and 479 of Part 2,
Section 10.5, shows that the JPEG encoder (VEPU121)'s base is not the
same as the Hantro decoder (VDPU121)'s base, and their IOMMUs which are
based relative to their base offset are therefore also not at the same
address. If you think the TRM must be wrong then, consider the fact that
I have actually run this patch set, presumably being the only person to
do so, and found that it works, so no, the addresses and power domains
are correct.

I do not see any way in which it would make sense to put this into the
same node as the decoder. It would not even be possible to do this in
your bindings, as they specify a maxItems for power-domains and iommus
of 1. Even if I modified them the driver wouldn't know which PD and
IOMMU belongs to decoder and encoder.

I think if we put this encoder in the same node as the decoder, we
might as well take this to its natural conclusion and put the entire
device tree into a single very large node. It's not the same hardware,
it cannot be modelled as being the same hardware, just because the
bindings lets people model some separate hardware as the same hardware
doesn't mean this applies to this hardware.

Long story short, why did I split the encoder to its own node? The
answer is that I didn't. I simply refused to combine it into a node
that it has nothing to do with.
 
> If we have good reasons to have separated Device Tree nodes,
> then having interrupt-names = "vepu" for its only interrupt line
> doesn't make sense.

How does it not make sense? The bindings allow for a vdpu only
interrupt-names, which in my understanding makes the same amount
of sense.

Regards,
Nicolas Frattaroli

> 
> > What does matter though is to be consistent in the bindings.
> > interrupt-names is a required property even if there's only a vdpu
> > interrupt. I modelled my vepu-only binding after this case.
> >
> 
> The current binding models the idea of decoder and encoder
> being the same device. This has never been really really accurate,
> as the encoder and decoders have always been more or less independent.
> 
> The reason for having them on a single device are mostly historical,
> some old devices shared some resource. I don't think this is the case anymore,
> but the binding was still modeled to support that.
> 
> Hopefully this makes sense!
> Thanks,
> Ezequiel
> 
> 
> > If robh thinks there is no value to having the interrupt show up
> > as anything other than "default" in /proc/interrupts, then I respectfully
> > disagree with that opinion and point out that this should have been brought
> > up when the vdpu-only case in the bindings was made to require
> > interrupt-names also.
> >
> > Changing the binding now that there theoretically could be drivers out
> > in the wild (though I doubt it) that do require interrupt-names, because
> > the binding told them that this is okay to do, seems unwise to me.
> >
> > Regards,
> > Nicolas Frattaroli
> >
> > >
> > > Thanks,
> > > Ezequiel
> > >
> > > > +               clocks = <&cru ACLK_JENC>, <&cru HCLK_JENC>;
> > > > +               clock-names = "aclk", "hclk";
> > > > +               iommus = <&vepu_mmu>;
> > > > +               power-domains = <&power RK3568_PD_RGA>;
> > > > +       };
> > > > +
> > > > +       vepu_mmu: iommu@fdee0800 {
> > > > +               compatible = "rockchip,rk3568-iommu";
> > > > +               reg = <0x0 0xfdee0800 0x0 0x40>;
> > > > +               interrupts = <GIC_SPI 63 IRQ_TYPE_LEVEL_HIGH>;
> > > > +               clocks = <&cru ACLK_JENC>, <&cru HCLK_JENC>;
> > > > +               clock-names = "aclk", "iface";
> > > > +               power-domains = <&power RK3568_PD_RGA>;
> > > > +               #iommu-cells = <0>;
> > > > +       };
> > > > +
> > > >         sdmmc2: mmc@fe000000 {
> > > >                 compatible = "rockchip,rk3568-dw-mshc", "rockchip,rk3288-dw-mshc";
> > > >                 reg = <0x0 0xfe000000 0x0 0x4000>;
> > > > --
> > > > 2.36.0
> > > >
> > > >
> > > > _______________________________________________
> > > > Linux-rockchip mailing list
> > > > Linux-rockchip@xxxxxxxxxxxxxxxxxxx
> > > > http://lists.infradead.org/mailman/listinfo/linux-rockchip
> > >
> >
> >
> >
> >
> 







[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux