Re: [PATCH v3, 03/10] dt-bindings: media: mtk-vcodec: Adds encoder cores dt-bindings for mt8195

[Rob Herring <robh@kernel.org>]

On Wed, Mar 30, 2022 at 04:48:47PM +0800, Irui Wang wrote:
> Dear Rob,
> 
> A sample encoder hardware block diagram attached.
> On Tue, 2022-03-29 at 08:09 -0500, Rob Herring wrote:
> > On Tue, Mar 29, 2022 at 09:26:37AM +0800, Irui Wang wrote:
> > > Dear Rob,
> > > 
> > > Many thanks for your attention.
> > > 
> > > On Mon, 2022-03-28 at 08:48 -0500, Rob Herring wrote:
> > > > On Sat, Mar 26, 2022 at 10:00:55AM +0800, Irui Wang wrote:
> > > > > Dear Rob,
> > > > > 
> > > > > Thanks for your review and comments.
> > > > > 
> > > > > On Fri, 2022-03-25 at 15:57 -0500, Rob Herring wrote:
> > > > > > On Thu, Mar 17, 2022 at 04:22:23PM +0800, Irui Wang wrote:
> > > > > > > Adds encoder cores dt-bindings for mt8195.
> > 
> > [...]
> > 
> > > > > > > +      mediatek,core-id:
> > > > > > > +        $ref: /schemas/types.yaml#/definitions/uint32
> > > > > > > +        description: |
> > > > > > > +          Current encoder core id.
> > > > > > 
> > > > > > What is this for and what does its value correspond to in the
> > > > > > h/w.
> > > > > > We 
> > > > > > generally don't do made up indices in DT.
> > > > > 
> > > > > It's for encoder core id, core@1a020000 must be core-0, 
> > > > > core@1b020000
> > > > > must be core-1, we add this property in each child node, so we
> > > > > can 
> > > > > get core-id in drivers. If it can't ref "uint32" types yaml,
> > > > > would 
> > > > > you mind giving some more suggestions ?
> > > > 
> > > > I still don't understand why it is needed. What is 'core-0'?
> > > > 
> > > > Is there some functional difference between the cores? If so,
> > > > describe 
> > > > that difference.
> > > > 
> > > > Rob
> > > 
> > > They are two different pieces of hardware, it's our encoder
> > > hardware
> > > design. There are two encoder hardware cores inside MT8195, named
> > > core0
> > > and core1(we can rename it, but core id should be declared),
> > > for core0, its module base address is 0x1A02_0000, uses IOMMU
> > > "vdo0_iommu" and power domain "POWER_DOMAIN_VENC",
> > > for core1, its module base address is 0x1B02_0000, uses IOMMU
> > > "vpp_iommu" and power domain "POWER_DOMAIN_VENC_CORE1".
> > > So the two encoder cores have their own base, IRQ, clock, power,
> > > etc.
> > > Each core can encode independently, moreover, they can work
> > > together
> > > for higher performance. 
> > > We will describe more details in YAML about it if it's OK for you.
> > 
> > All the resources you list are in the child nodes, so you don't need
> > 0 
> > and 1 numbering for those. 
> > 
> > Looking at the driver patches, the only thing I see distinguishing 
> > core numbers is this:
> > 
> > "frame#0 uses core#0, frame#1 uses core#1, frame#2 uses core#0...,
> > 
> > Lock the device and enable the clock by used core, for sequence
> > header encoding, it always uses core#0."
> > 
> > Is this a requirement in the h/w or just what the driver picked?
> > IOW, 
> > could frame#0 use core#1?
> No, it's a requirement in the h/w, driver trigger core start encoding
> is in order.
> About the encoder hardware block diagram, please check below:
> --------------------------------------------------------------
> Input Buffer: 0     1     2     3     4     5     6
>               |     |     |     |     |     |     |
>               v     |     v     |     v     |     v
>           +-------+ | +-------+ | +-------+ | +-------+
>           | core0 | | | core0 | | | core0 | | | core0 |
>           +-------+ | +-------+ | +-------+ | +-------+
>               |     |     |     |     |     |     |
>               |     v     |     v     |     v     |
>               | +-------+ | +-------+ | +-------+ |
>               | | core1 | | | core1 | | | core1 | |
>               | +-------+ | +-------+ | +-------+ |
>               |     |     |     |     |     |     |
>               v     v     v     v     v     v     v    <parent>
> --------------------------------------------------------------
>                         core || index                  <child>
>                              \/
>        +-----------------------------------------------+
>        |                  core0/core1                  |
>        |          enable/disable power/clk/irq         |
>        +-----------------------------------------------+
> --------------------------------------------------------------
> As above , there are parent and child devices, child mean each venc
> core, the child device controls the information of each core
> independent which inlcude power/clk/irq.
> When start encoding, input buffer 0 will be encoded by core0, and input
> buffer 1 can be encoded by core1 even if buffer 0 has not been encoded
> done yet, after buffer 0 encoded done, buffer 2 will be encoded by
> core0, and buffer 1 encoded done by core1. These two encoder cores will
> encode each input in this overlapping manner.
> 
> We need manage each child device in parent device by core-id property.
> And we also need record current encoding input buffer, encode done
> output buffers and which one core is in used through core-id, because
> the two cores are encoding at the same time under one parent driver.

Okay, please summarize the h/w requirement in the property description.

You might put the above diagram in the top description too.

Rob