Re: [PATCH v2 7/9] dt-bindings: spi: mtk-snfi: Add read latch latency property

[AngeloGioacchino Del Regno]

Il 07/12/22 03:00, Xiangsheng Hou (侯祥胜) ha scritto:
> Hi Angelo,
>
> On Tue, 2022-12-06 at 13:19 +0100, AngeloGioacchino Del Regno wrote:
> > > > > diff --git
> > > > > a/Documentation/devicetree/bindings/spi/mediatek,spi-
> > > > > mtk-snfi.yaml
> > > > > b/Documentation/devicetree/bindings/spi/mediatek,spi-
> > > > > mtk-snfi.yaml
> > > > > index bab23f1b11fd..6e6ff8d73fcd 100644
> > > > > --- a/Documentation/devicetree/bindings/spi/mediatek,spi-mtk-
> > > > > snfi.yaml
> > > > > +++ b/Documentation/devicetree/bindings/spi/mediatek,spi-mtk-
> > > > > snfi.yaml
> > > > > @@ -45,6 +45,13 @@ properties:
> > > > >         description: device-tree node of the accompanying ECC
> > > > > engine.
> > > > >         $ref: /schemas/types.yaml#/definitions/phandle
> > > > >     
> > > > > +  mediatek,rx-latch-latency:
> > > > > +    description: Rx delay to sample data with this value, the
> > > > > value
> > > > > +                 unit is clock cycle.
> > > >
> > > > Can't we use nanoseconds or microseconds as a unit here, instead
> > > > of
> > > > clock cycles?
> > >
> > > The clock cycle will be various with MediaTek SPI NAND controller
> > > which
> > > clock frequency can support 26/52/68/81/104MHz...
> > > It`s may be easy to configure and understand with clock cycle in
> > > unit.
> >
> > Yes, but whatever clock frequency we use, the target is to always
> > wait for
> > X nanoseconds, right?
> >
> > Waiting for 5 clock cycles at 104MHz is obviously not the same as
> > waiting
> > for the same 5 clock cycles at 26MHz: in that case, expressing the
> > value
> > in nanoseconds or microseconds would make that independent from the
> > controller's clock frequency as the calculation from `time` to
> > `cycles`
> > would be performed inside of the driver.
>
> There have two rx related timing properties in spi-peripheral-props.
> The rx-sample-delay-ns have been used in Mediatek snfi driver to adjust
> controller sample delay.
> However another spi-rx-delay-us is in microseconds. Take 52MHz for
> example, the clock cycle will be 19.23ns which lower than 1us. This may
> not easy to by one clock cycle.

I agree, but nothing prevents you from adding your own property for that.

I propose "mediatek,rx-latch-latency-ns" or "mediatek,rx-latency-ns", so that
we can specify the delay in nanoseconds: in that case, when we specify 19ns,
the driver will safely round that resulting in 52MHz == 19.23ns => 19ns valid.

Regards,
Angelo