Re: [PATCH 0/3] usb: USB Type-C Class and driver for UCSI

From: Heikki Krogerus
Date: Fri May 06 2016 - 04:29:24 EST


On Fri, May 06, 2016 at 01:05:05AM -0700, Guenter Roeck wrote:
> Felipe,
>
> On 05/05/2016 11:50 PM, Felipe Balbi wrote:
> >
> > Hi Guenter,
> >
> > Guenter Roeck <linux@xxxxxxxxxxxx> writes:
> > > On Tue, Feb 09, 2016 at 07:01:20PM +0200, Heikki Krogerus wrote:
> > > > Hi,
> > > >
> > > > The OS, or more precisely the user space, needs to be able to control
> > > > a few things regarding USB Type-C ports. The first thing that must be
> > > > allowed to be controlled is the data role. USB Type-C ports will
> > > > select the data role randomly with DRP ports. When USB PD is
> > > > supported, also independent (from data role) power role swapping can
> > > > be supported together with Alternate Mode control.
> > > >
> > > > I'm proposing with this set a Class for the Type-C connectors that
> > > > gives the user space control over those things on top of getting basic
> > > > details about the USB Type-C connectors and also partners. The details
> > > > include the capabilities of the port, the supported data and power
> > > > roles, supported accessories (audio and debug), supported Alternate
> > > > Modes, USB PD support and of course the type of the partner (USB, Alt
> > > > Mode, Accessory or Charger), and more or less the same details about
> > > > the partner.
> > > >
> > > > I'm not considering cables with this Class, and I have deliberately
> > > > left out some more technical details, like cable orientation, firstly
> > > > because I did not see much use for the user space from knowing that
> > > > an secondly because that kind of details are not always available for
> > > > example with UCSI.
> > > >
> > > > So the interface to the user space is kept as simple as I dared to
> > > > make it.
> > > >
> > > > NOTE: In case there is somebody wondering, this is not adding USB PD
> > > > support to Linux kernel. This is just about USB Type-C.
> > > >
> > >
> > > Hello Heikki,
> > >
> > > we have implemented a prototype TCPM (USB Type-C Protocol Manager)
> > > software on top of your patch set. It will support TCPCI as well
> > > as other USB-C controllers such as FUSB302. The plan is to use
> > > this software in systems where no separate controller is available.
> > >
> > > Is there any chance to advance this patch set ? It would be instrumental
> > > to get a unified interface to user space.
> >
> > A newer version of $subject is already in Greg's queue [1]
> >
> > [1] https://git.kernel.org/cgit/linux/kernel/git/gregkh/usb.git/commit/?h=usb-next&id=0c1849a8c7af652c92ad0265a7ca5934fd773c69
> >
> I am aware of that patch.
>
> Unfortunately, unlike the original submission, the new patch is not an
> infrastructure, it is just a driver supporting Intel's UCSI. Unlike the
> original series, it does not provide an infrastructure, and it does not
> support other implementations of USB Type-C port management systems.
>
> In our system, we'll have (at least) three such implementations:
>
> - TCPM and TCPC implemented in EC and/or microcontrollers.
> This is currently implemented and shipping with some Chromebooks.
> - TCPM implemented in Linux, interfacing to a standard TCPC, using TCPCI
> for TCPM-TCPC communication
> This will be needed for systems with no EC and a standard Type-C port
> controller.
> - TCPM implemented in Linux, interfacing to FUSB302.
> This will be needed for systems with no EC, utilizing a FUSB302
> port controller.
>
> All those fit nicely into the infrastructure provided by the original
> patch series, where UCSI was just one possible implementation of a
> USB Type-C port management system.
>
> The original patch series had the tremendous advantage of presenting a
> unified ABI to user space. With the new patch, this is no longer the case.
> All implementations would be completely separate and thus effectively
> guarantee ABI fragmentation (Fairchild's code supporting FUSB302 in Linux
> is a good example. The existing implementation of Type-C support in the
> Chromebooks mentioned above is another).
>
> I know there has been a lengthy discussion about the patch set, but I may
> have missed the conclusion. Is there some reason to _not_ advance it
> that I may have missed ?

No, we are still continuing with the class driver. We just descided to
split the UCIS into separate driver for now, just because we needed it
to be supported fast. But I did mention in the commit message of the
UCSI patch that the goal is to merge that into a Type-C framework once
it's awailable.


Thanks,

--
heikki