Re: [PATCH 1/2] dt-bindings: mux: Document mux-states property

[Rob Herring <robh@kernel.org>]

On Tue, Nov 30, 2021 at 10:32 PM Aswath Govindraju <a-govindraju@ti.com> wrote:
>
> Hi Rob,
>
> On 01/12/21 2:18 am, Peter Rosin wrote:
> >
> >
> > On 2021-11-30 21:14, Rob Herring wrote:
> >> On Tue, Nov 30, 2021 at 05:48:46PM +0530, Aswath Govindraju wrote:
> >>> In some cases, it is required to provide the state to which the mux
> >>> controller has be set to, from the consumer device tree node. Document the
> >>> property mux-states that can be used for adding this support.
> >>
> >> I having a hard time understanding why you need this. One consumer
> >> configures a mux one way and another consumer another way? How do you
> >> arbitrate that? Please elaborate on what 'some cases' are and why it's
> >> required.
> >>
> >> Can't you just add a cell for the 'state' allowing for 1-2 cells
> >> instead of 0-1?
> >
> > A mux controller can control several muxes. That happens e.g. when the
> > same gpio lines are connected to several mux chips in parallel. When
> > you operate one mux, the other parallel muxes just follow along. If
> > these muxes are then used orthogonally, coordination is needed. The real
> > world case I had was I2C and an analog signal connected to an ADC that
> > went through parallel/dependent muxes like this. It is simply not
> > possible to freely mux the I2C bus and the analog signal, they are tied
> > together and dependent and must coordinate their accesses.
> >
> > The addition now is that Aswath wants a mux control client to "point
> > at" a single state instead of the whole mux control, and I see that as
> > a usable addition. It seems like a natural place to specify a single mux
> > state that some driver needs in some circumstance.
> >
> > But, since a mux control is inherently a shared resource (see above),
> > one consumer might need a specific state and some other consumer might
> > need the whole mux control and manage the states as e.g. the existing
> > i2c-mux-gpmux binding is doing. So, you need to be able to specify both
> > ways to point at muxes; either to a single mux state, or to the whole mux
> > control.
> >
> > While you could make the extra cell optional, that does not work for
> > the mux/adi,adg792a binding, since it is using the #mux-control-cells
> > property to determine which mode it should operate its three muxes in.
> > Either with one common/parallel mux control, or with three independent
> > mux controls.
> >
> > So, that binding is already in the 0-1 territory, and adding an optional
> > extra cell makes it 0-1-2 with no way to determine what is intended when
> > the cell count is 1 (three independent mux controls OR one mux control
> > and a state). I see no way to add the extra state to that binding, short
> > of adding an extra property somewhere for that driver, but I simply did
> > not want to go that path because it would get inconsistent when trying
> > to add that in a backwards compatible way. Or rather, that was my
> > conclusion.
> >
> > Suggestions welcome...
> >
>
>
> In addition to what Peter has mentioned, I would like to elaborate on my
> use case for adding this feature. I am trying to implement this feature
> in the TCAN104x transceiver driver, for selecting the mux state to route
> the signals from CAN controller to transceivers on the board. The state
> of the mux line to be set, can change based on the design and this is
> needs to be provided from the device tree. Hence, I am trying to add
> this support for providing the state to be set to the driver from the
> device tree node.

Okay, please add something along the lines of what Peter said for when
you use which binding.

> Also, one more question on regarding DT check errors, may I know what
> should be the order in which the patches need to be posted in order to
> not get the error? This is because mux-states would be a new property to
> be added in the TCAN104x bindings and I thought that it would need to be
> posted after the patch for the changes in mux-controller are merged.

Looks like a circular dependency. Assuming you ran dt_binding_check on
the series, just add a note about the dependency and I won't send the
report.

Rob