Re: [PATCH v6 1/2] dt-bindings: i2c: aspeed: support for AST2600-i2cv2

[Jeremy Kerr <jk@codeconstruct.com.au>]

Hi Ryan,

> --- a/Documentation/devicetree/bindings/i2c/aspeed,i2c.yaml
> +++ b/Documentation/devicetree/bindings/i2c/aspeed,i2c.yaml
> @@ -49,6 +49,25 @@ properties:
>      description:
>        states that there is another master active on this bus
>  
> +  aspeed,timeout:
> +    type: boolean
> +    description: I2C bus timeout enable for master/slave mode
> +
> +  aspeed,xfer-mode:
> +    description: |
> +      I2C bus transfer mode selection.
> +      - "byte": I2C bus byte transfer mode.
> +      - "buffered": I2C bus buffer register transfer mode.
> +      - "dma": I2C bus dma transfer mode (default)
> +    items:
> +      enum: [byte, buffered, dma]
> +    maxItems: 1
> +    $ref: /schemas/types.yaml#/definitions/non-unique-string-array

There are still unresolved questions about this xfer-mode property from
previous submissions of this binding. We don't yet have a justification
on why the mode configuration is needed in the device tree rather than
something that is specified in a driver implementation.

By now, I think we well understand what the modes are, and how a driver
implementation might configure them, but none of that has (so far)
provided sufficient rationale on why this belongs in the device tree.

The previous threads had a couple of pending discussions, following up on
those here:

A) You mentioned in [1] that the DMA controller is shared between all i3c
devices, does that have any consequence on which modes individual
devices might want to choose?

B) You implied in [2] that the different transfer modes might be related
to whether there are other masters present on the bus, but the logic
behind that is not clear.

C) In [3] you mentioned that there might be some DRAM savings by using a
particular mode.

and, most importantly:

D) unanswered from [4] and [5]: what are the hardware-specified reasons
why a DT author would chose one mode over another?

If you can write this out in some format like:

 - in hardware situation X, you should use DMA mode
 - in hardware situation Y, you should use byte mode
 - [...]

that might help us to understand where this configuration belongs, or
what a reasonable DT representation should look like, or even if
existing DT schema can already provide the information required to
decide.

Cheers,


Jeremy

[1]: https://lists.ozlabs.org/pipermail/linux-aspeed/2023-February/009876.html
[2]: https://lists.ozlabs.org/pipermail/linux-aspeed/2023-February/009892.html
[3]: https://lists.ozlabs.org/pipermail/linux-aspeed/2023-February/009880.html
[4]: https://lists.ozlabs.org/pipermail/linux-aspeed/2023-February/009871.html
[5]: https://lists.ozlabs.org/pipermail/linux-aspeed/2023-February/009884.html