Re: [PATCHv11 2/3] usb: USB Type-C connector class

From: Greg KH <gregkh@linuxfoundation.org>
To: Heikki Krogerus <heikki.krogerus@linux.intel.com>
Cc: Guenter Roeck <linux@roeck-us.net>,
	Badhri Jagan Sridharan <badhri@google.com>,
	Oliver Neukum <oneukum@suse.com>,
	linux-kernel@vger.kernel.org, linux-usb@vger.kernel.org
Subject: Re: [PATCHv11 2/3] usb: USB Type-C connector class
Date: Mon, 21 Nov 2016 15:45:06 +0100	[thread overview]
Message-ID: <20161121144506.GA28410@kroah.com> (raw)
In-Reply-To: <20161121131103.GA18501@kuha.fi.intel.com>

On Mon, Nov 21, 2016 at 03:11:03PM +0200, Heikki Krogerus wrote:
> Hi Greg,
> 
> On Mon, Nov 21, 2016 at 11:35:28AM +0100, Greg KH wrote:
> > > +static void typec_partner_release(struct device *dev)
> > > +{
> > > +	struct typec_port *port = to_typec_port(dev->parent);
> > > +
> > > +	typec_unregister_altmodes(dev);
> > > +	port->partner = NULL;
> > > +}
> > 
> > This doesn't feel right, why are you both exporting
> > typec_unregister_altmodes() and also calling it from release callbacks?
> > That implies there is two way to clean stuff up, so what is a driver
> > writer to use?  Please simplify and force it to be one way or the other.
> 
> OK.
> 
> > Also typec_unregister_altmodes() doesn't free memory, which release is
> > supposed to be doing, which also implies that the reference counting
> > logic is all wrong here.  Please fix, like I asked you to previously.
> 
> There is nothing wrong with the reference counting, and nothing has
> been allocated so there is nothing to free.

The device structure itself that this release call is for needs to
be freed, right?  If not, something is really wrong...

> Please note that the partner device is meant to just represent the
> partner in user space and not to be actually used for anything. And
> please also note that there can only be one partner for a port at a
> time.

Ok, but these are still reference counted devices, you need to handle
that properly.

> We could allocate an extra structure for the partner when
> typec_connect() is called, but we would do that just for the sake of
> having something to free in the release hook. It would not be useful
> for anything. It would not help us increase/decrease the reference
> count of the device, and the port driver would still have to provide
> details about the partner capabilities the moment it tells us the
> partner was connected.

Again, free the device for which this release function is being called
for, that is why it is there.

thanks,

greg k-h