Re: [PATCH 2/7] dt-bindings: phy: mediatek,tphy: add property to set pre-emphasis

[Chunfeng Yun]

On Wed, 2022-08-31 at 09:03 +0300, Krzysztof Kozlowski wrote:
> On 31/08/2022 06:00, Chunfeng Yun wrote:
> > On Tue, 2022-08-30 at 13:08 +0300, Krzysztof Kozlowski wrote:
> > > On 29/08/2022 05:37, Chunfeng Yun wrote:
> > > > On Fri, 2022-08-26 at 09:36 +0300, Krzysztof Kozlowski wrote:
> > > > > On 26/08/2022 08:36, Chunfeng Yun wrote:
> > > > > > On Tue, 2022-08-23 at 13:24 +0300, Krzysztof Kozlowski
> > > > > > wrote:
> > > > > > > On 22/08/2022 10:07, Chunfeng Yun wrote:
> > > > > > > > On Fri, 2022-08-19 at 15:15 +0300, Krzysztof Kozlowski
> > > > > > > > wrote:
> > > > > > > > > On 19/08/2022 12:13, Chunfeng Yun wrote:
> > > > > > > > > > Add a property to set usb2 phy's pre-emphasis.
> > > > > > > > > > 
> > > > > > > > > > Signed-off-by: Chunfeng Yun <
> > > > > > > > > > chunfeng.yun@xxxxxxxxxxxx>
> > > > > > > > > > ---
> > > > > > > > > >  Documentation/devicetree/bindings/phy/mediatek,tph
> > > > > > > > > > y.ya
> > > > > > > > > > ml |
> > > > > > > > > > 7
> > > > > > > > > > +++++++
> > > > > > > > > >  1 file changed, 7 insertions(+)
> > > > > > > > > > 
> > > > > > > > > > diff --git
> > > > > > > > > > a/Documentation/devicetree/bindings/phy/mediatek,tp
> > > > > > > > > > hy.y
> > > > > > > > > > aml
> > > > > > > > > > b/Documentation/devicetree/bindings/phy/mediatek,tp
> > > > > > > > > > hy.y
> > > > > > > > > > aml
> > > > > > > > > > index 848edfb1f677..aee2f3027371 100644
> > > > > > > > > > ---
> > > > > > > > > > a/Documentation/devicetree/bindings/phy/mediatek,tp
> > > > > > > > > > hy.y
> > > > > > > > > > aml
> > > > > > > > > > +++
> > > > > > > > > > b/Documentation/devicetree/bindings/phy/mediatek,tp
> > > > > > > > > > hy.y
> > > > > > > > > > aml
> > > > > > > > > > @@ -219,6 +219,13 @@ patternProperties:
> > > > > > > > > >          minimum: 1
> > > > > > > > > >          maximum: 15
> > > > > > > > > >  
> > > > > > > > > > +      mediatek,pre-emphasis:
> > > > > > > > > > +        description:
> > > > > > > > > > +          The selection of pre-emphasis (U2 phy)
> > > > > > > > > > +        $ref:
> > > > > > > > > > /schemas/types.yaml#/definitions/uint32
> > > > > > > > > > +        minimum: 1
> > > > > > > > > > +        maximum: 3
> > > > > > > > > 
> > > > > > > > > Instead of hard-coding register values in bindings,
> > > > > > > > > you
> > > > > > > > > should
> > > > > > > > > rather
> > > > > > > > > describe here feature/effect. If it is in units, use
> > > > > > > > > unit
> > > > > > > > > suffixes.
> > > > > > > > > If
> > > > > > > > > it is some choice, usually string enum is
> > > > > > > > > appropriate.
> > > > > > > > 
> > > > > > > > How about changing description as bellow:
> > > > > > > > 
> > > > > > > > "The level of pre-emphasis, increases one level, boosts
> > > > > > > > the
> > > > > > > > relative
> > > > > > > > amplitudes of signal's higher frequencies components
> > > > > > > > about
> > > > > > > > 4.16%
> > > > > > > > (U2
> > > > > > > > phy)"
> > > > > > > > 
> > > > > > > 
> > > > > > > Still the question is what is the unit. 4.16%?
> > > > > > 
> > > > > > No unit, it's a level value, like an index of array.
> > > > > > 
> > > > > 
> > > > > So a value from register/device programming? 
> > > > 
> > > > Yes
> > > > > Rather a regular units
> > > > > should be used if that's possible. If not, this should be
> > > > > clearly
> > > > > described here, not some magical number which you encode into
> > > > > DTS...
> > > > 
> > > > Ok, I'll add more descriptions.
> > > 
> > > Better use logical value, e.g.
> > > 
> > 
> > 
https://urldefense.com/v3/__https://github.com/devicetree-org/dt-schema/blob/main/dtschema/schemas/property-units.yaml*L38__;Iw!!CTRNKA9wMg0ARbw!1e-h0R_uwcaHKfKC9qYfaRWYeuWRq1sLCGy3yupNmkFyuW5s1nmRotL7Y0vFG9ETLLTA$
> > >  
> > 
> > Optional unit may be -percent or -bp, but the value 4.16% * X
> > (X=1,2,3...)is not an exact value, they are variable in a range and
> > dependent more factors.
> > So I think use level value is simple enough.
> 
> Then again explain exactly what are the levels. How you wrote it last
> time, -bp would do the trick.

There are many different methods of measuring pre-emphasis.
The way used in MediaTek USB2 PHY as below:

pre-emphasis level equation = Vpp/Vs -1;
Vpp: peak-peak voltage of differential signal;
Vs : static voltage of differential signal, normal voltage, e.g. 400mV
for u2 phy;

The pre-emphasis circuitry within t-phy can be dynamically programmed
to three different levels of pre-emphasis. The exact value of
pre-emphasis cannot be predetermined, because each device requires
a percentage of pre-emphasis that is dependent on the output signal
strength and transmission path characteristics.

Below shows three programmable pre-emphasis levels for a differential
drive signal of 400 mV. The amount of pre-emphasis changes according
to the transmission path parameters.

programmable level   typical pre-emphasis level
1                    4.16%
2                    8.30%
3                    12.40%

The reasons that why prefer to use programmable level in dt-binding as
following:
1. as you said, -bp may do the trick, but the main problem is that
   pre-emphasis level is variable on different SoC, and is also
   variable even on different pcb for the same SoC. e.g. for the
   programmable level 1, pre-emphasis level may be 6% on a platform,
   but for the programmable level 2, pre-emphasis level may be also
   6% on another platform;
   I think use pre-emphasis level in property, e.g. 4.16%, the
   deviation may be bigger than 40%, may cause confusion for users,
   due to we can't promise that the actual measurement is about 4.16%,
   it may be 2%, or 5% when measured;
2. the programmable / logical level is flexible when we support more
   and pre-emphasis level, ans it is easy for us to tune the level
   due to it's continuous value.
3. all other vendor properties that can't provide exact measurable
   value in this dt-binding make use of logic level, I want to
   keep them align;

Thanks a lot

> 
> Best regards,
> Krzysztof