Re: [PATCH 1/2] dt-bindings: usb: Introduce GPIO-based SBU mux

From: Bjorn Andersson <andersson@kernel.org>
To: Prashant Malani <pmalani@chromium.org>
Cc: Bjorn Andersson <bjorn.andersson@linaro.org>,
	Stephen Boyd <swboyd@chromium.org>, Rob Herring <robh@kernel.org>,
	Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
	Heikki Krogerus <heikki.krogerus@linux.intel.com>,
	linux-usb@vger.kernel.org, devicetree@vger.kernel.org,
	linux-kernel@vger.kernel.org, linux-arm-msm@vger.kernel.org,
	Pin-yen Lin <treapking@chromium.org>
Subject: Re: [PATCH 1/2] dt-bindings: usb: Introduce GPIO-based SBU mux
Date: Thu, 25 Aug 2022 22:02:22 -0500	[thread overview]
Message-ID: <20220826030222.jmgjlsf7p3sx2lld@baldur> (raw)
In-Reply-To: <CACeCKaewr92JcouMgyiL5rKdHf_dDmoWoyoW-U2snCNBkNijQg@mail.gmail.com>

On Tue, Aug 23, 2022 at 11:54:53AM -0700, Prashant Malani wrote:
> On Fri, Aug 19, 2022 at 9:09 PM Bjorn Andersson
> <bjorn.andersson@linaro.org> wrote:
> >
> >
> > We're talking about the static configuration here, where you describe
> > which component are connected together. We can not dynamically switch
> > the Devicetree representation around to match what the Type-C controller
> > negotiates.
> 
> Yes, but we don't need to switch the device tree representation at all.
> The pin routing/connections from the connector (not the cable or the partner),
> to the muxing hardware (QMP phy or anx7625) remains fixed always
> The port driver tells what orientation the peripheral is connected with,
> and the muxing/orientation hardware routes the signals according to that.
> 
> >
> > But why do you need to express the relationship between these 2
> > components with > 1 link in the graph?
> >
> > > The graph is static, since the hardware line routing between components
> > > doesn't change (e.g SSTX1 from the Type-C port is always routed to Pin
> > > X1,X2 on the switch hardware), but that is what the switch is for.
> > > Note that in this case, the expectation is that
> > > the switch driver only registers 1 switch (it can figure out that all
> > > 4 endpoints
> > > go to the same Type-C port).
> > >
> >
> > Why do we need to express this with 4 endpoints and then implement code
> > to discover that the 4 endpoints points to the same remote? Why not just
> > describe the logical relationship between the two components in one
> > endpoint reference?
> 
> The issue I see is with the "supplier" side of that graph relationship
> (i.e the DRM bridge side).
> Since the bridge can be directly connected to a DP panel, the
> endpoints can (technically)
> represent a single DP lane. So, using 4 end-points for the
> usb-c-connector port@1 gives
> us something which is compatible with the bridge side endpoints too
> (regardless of what
> the bridge is connected to on the "output" side).
> Reading the discussion, I agree 4 lanes is over-specifying, and 2
> endpoints is probably
> enough (especially if we can use data-lanes on the bridge side
> to define the number of lanes if needed for DP panel connections).
> 

I'm sorry, but the part I don't understand is what you gain from
representing each physical line in your connection with a
remote-endpoint pair?

What I propose is that you tie the two pieces together with a single
reference. If you need to express the number of data-lanes we have
several places where this is described separately, using the
"data-lanes" property.

With this model, if you have a 1:1 connection you have a single
remote-endpoint pair, if you have a 1:N connection, then you would have
N remote-endpoint pairs.

Regards,
Bjorn