Re: [PATCH v3 1/6] dt-bindings: PCI: Add binding for qps615

From: Dmitry Baryshkov
Date: Thu Nov 28 2024 - 09:09:19 EST


On Thu, Nov 28, 2024 at 06:54:32PM +0530, Manivannan Sadhasivam wrote:
> On Tue, Nov 26, 2024 at 07:58:16AM +0100, Krzysztof Kozlowski wrote:
> > On 26/11/2024 07:50, Krishna Chaitanya Chundru wrote:
> > >
> > >
> > > On 11/25/2024 1:10 PM, Krzysztof Kozlowski wrote:
> > >> On 24/11/2024 02:41, Krishna Chaitanya Chundru wrote:
> > >>>> ...
> > >>>>
> > >>>>> + qps615,axi-clk-freq-hz:
> > >>>>
> > >>>> That's a downstream code you send us.
> > >>>>
> > >>>> Anyway, why assigned clock rates do not work for you? You are
> > >>>> re-implementing legacy property now under different name :/
> > >>>>
> > >>>> The assigned clock rates comes in to the picture when we are
> > >>>> using clock
> > >>> framework to control the clocks. For this switch there are no
> > >>> clocks needs to be control, the moment we power on the switch
> > >>> clocks are enabled by default. This switch provides a mechanism to
> > >>> control the frequency using i2c. And switch supports only two
> > >>> frequencies i.e
> > >>
> > >>
> > >> frequency of what, since there are no clocks?
> > >>
> > > The axi clock frequency internal to the switch, host can't control
> > > the enablement of the clocks it can control only the frequency.
> > >
> > > we already had a discussion on this on v2[1], and we taught you agreed
> > > on this property.
> > >
> > > [1]
> > > https://lore.kernel.org/netdev/d1af1eac-f9bd-7a8e-586b-5c2a76445145@xxxxxxxxxxxxxx/T/#m3d5864c758f2e05fa15ba522aad6a37e3417bd9f
> > >
> >
> > This points something else. I diged v2 and found many unanswered
> > questions and unfinished discussion:
> >
>
> The conversation is here:
> https://lore.kernel.org/linux-arm-msm/20240823094028.7xul4eoiexey5xjm@thinkpad/
>
> But there was no explicit agreement on the usage of 'qps615,axi-clk-freq-hz'.
>
> If describing the PCI device's internal clock frequency is not applicable, then
> I'd recommend to change the clock rate in the driver itself based on the number
> of DSPs enabled (or based on other configuration).

Sounds like the best approach. Otherwise it's not obvious, what is the
criteria for selecting this or that clock value.

>
> - Mani
>
> --
> மணிவண்ணன் சதாசிவம்

--
With best wishes
Dmitry