-----Original Message-----
From: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Sent: Wednesday, May 26, 2021 5:16 PM
To: Jun Li <redacted>
Cc: robh+dt@kernel.org; shawnguo@kernel.org; gregkh@linuxfoundation.org;
linux@roeck-us.net; linux-usb@vger.kernel.org; dl-linux-imx
[off-list ref]; devicetree@vger.kernel.org;
linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH 3/4] usb: typec: add typec orientation switch support
via mux controller

On Tue, May 25, 2021 at 11:46:18AM +0000, Jun Li wrote:

quoted

Hi Heikki,

quoted

-----Original Message-----
From: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Sent: Friday, May 21, 2021 4:38 PM
To: Jun Li <redacted>
Cc: robh+dt@kernel.org; shawnguo@kernel.org;
gregkh@linuxfoundation.org; linux@roeck-us.net;
linux-usb@vger.kernel.org; dl-linux-imx [off-list ref];
devicetree@vger.kernel.org; linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH 3/4] usb: typec: add typec orientation switch
support via mux controller

Hi,

On Thu, May 20, 2021 at 03:33:36PM +0300, Heikki Krogerus wrote:

quoted

Why not just do that inside fwnode_typec_switch_get() and handle
the whole thing in drivers/usb/typec/mux.c (or in its own file if
you prefer)?

You'll just need to register a "wrapper" Type-C switch object for
the OF mux controller, but that should not be a problem. That way
you don't need to export any new functions, touch this file or
anything else.

I wrote a bit of code just to see how that would look. I'm attaching
you the hack I made. I guess something like that would not be too bad.
A wrapper is probable always a bit clumsy, but I'm not sure that in
this case it's a huge problem. Of course if there are any better
ideas, let's here them :-)

Thanks for your patch, I am pasting the patch as below.

seems we need consider more than that.

diff --git a/drivers/usb/typec/Makefile b/drivers/usb/typec/Makefile
index a0adb8947a301..d85231b2fe10b 100644
--- a/drivers/usb/typec/Makefile
+++ b/drivers/usb/typec/Makefile

@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: GPL-2.0
 obj-$(CONFIG_TYPEC)		+= typec.o
 typec-y				:= class.o mux.o bus.o port-mapper.o
+typec-$(MULTIPLEXER)		+= of_mux.o
 obj-$(CONFIG_TYPEC)		+= altmodes/
 obj-$(CONFIG_TYPEC_TCPM)	+= tcpm/
 obj-$(CONFIG_TYPEC_UCSI)	+= ucsi/

diff --git a/drivers/usb/typec/mux.c b/drivers/usb/typec/mux.c index

9da22ae3006c9..282622276d97b 100644

--- a/drivers/usb/typec/mux.c
+++ b/drivers/usb/typec/mux.c

@@ -63,6 +63,9 @@ struct typec_switch *fwnode_typec_switch_get(struct

fwnode_handle *fwnode)

 	sw = fwnode_connection_find_match(fwnode, "orientation-switch", NULL,
 					  typec_switch_match);
+	if (!sw)
+		sw = of_switch_register(fwnode);
+

How to support multiple typec_switch_get() for of_mux case?
the second call to fwnode_typec_switch_get() will get the switch via
fwnode_connection_find_match()? This means we still need a property
"orientation-switch" for mux controller node, this seems not the
expected way for a mux consumer, even with this property, we will get
a EPROBE_DEFER for the first call.

If we use mux_control_get() for multiple caller/consumer, then we need
some other device as input.

typec_switch object looks to me is a provider, if we create and
maintain it in consumer side, I have not come out a better way:-).

Sorry, but can we rewind a bit: Why can't you just register the orientation
switch from your mux driver and be done with it? You should then be able
to use OF graph, and no special bindings should be needed, no?

So we still need a special property for OF graph per discussion on another
thread(use device type other than device name for match), and this has
to be a mux controller core binding for possible different mux chips
(GPIO/MMIO...), register a typec switch if this property exist, but this
is the user specific thing from mux controller point view, I feed this
is again against DT binding's expectation.

If you want to reuse a mux-controller driver, then you do need to modify
it (but only a little), and what ever mux-controller specific bindings there
are, you will not use those when the mux supplies the orientation switching
function, instead you'll use the OF graph for that. But surely that is not
a problem?

The mux-controller framework expects the "consumers" of the muxes to
understand the final function that the mux is used for. The Type-C "mux"
framework (I should not even talk about muxes with those) works the other
way around. 

Fully agree.

The driver for the component that supplies the orientation switch
function must understand that it is handling that function, and there is
a good reason for doing it that way with the USB Type-C switches. 

I understand yes if the switch is only part function of the driver.
  

The
orientation switch for example quite simply is _not_ always a mux. In fact,
it's seems to be rarely a mux these days. With USB4 for example the orientation
is handled almost always by the first on-board retimer.

If the mux is only part function of a new driver, use the tyepc
"mux" framework and create new binding for the new driver is fine.

But if the typec switch control need a dedicated driver to handle,
on DT platforms, now mux-controller is the only proposed way to go
from binding point view. I am not sure if my case is a normal HW
design, but I guess I should not the only user of this kind of
situation.

There are actually also some technical reasons why Hans failed to get the
mux-controller thing to work, which is the original reason why we introduced
the dedicated framework for the Type-C "muxes" (I really should stop talking
about muxes), but I don't remember what was the reason.

I checked the patches Hans did, that was mainly to address non-DT
platform, I don't see a clear reason why it can't fit DT platform,
maybe I missed something.

+Hans, It would be great if you can comment on this, thanks.

In any case, to summarise: the orientation switch is a function. A mux is
a device that can supply that function, and if it does, then the driver for
it really needs to register the dedicated orientation switch.

Understand your point, if register the dedicated orientation switch is a must,
I feel using general mux control can't make much sense.

Thanks
Li Jun 

thanks,

--
heikki