Re: [PATCH] NFC: Driver for Inside Secure MicroRead NFC chip

[Arnd Bergmann <arnd-r2nGTMty4D4@public.gmane.org>]

On Tuesday 29 March 2011, Waldemar.Rymarkiewicz-++hxYGjEMp0AvxtiuMwx3w@public.gmane.org wrote:
> >Yes, NFC seems to be a good fit for a new socket family. 
> >Especially if we ever want to have a proper NFC p2p support 
> >from the kernel.
> >Sending HCI commands should be done through a dedicated 
> >netlink socket too.
> >
> >I am currently strting to work on such solution, and I hope to 
> >be able to come up with a basic prototype for it in a few weeks.
> 
> What about common drivers interface in this case.
> Should we go for common /dev/nfcX interface as well?

I fear there can only be one. A good implementation of a socket
interface would mean that there is no need for a character device.

The difference between the two is where you keep the common
NFC logic:

If you have a character device, it will be like a serial port
connecting to a modem. Any higher-level protocols live in the
user space and are limited to a single application then, which
is required to have appropriate priviledges to open the device.

In contrast, a socket implementation puts the protocol
stack into the kernel, which requires much more kernel code
but almost no user space library code, aside from perhaps
a small shim layer. It makes it possible to have multiple
applications and/or users concurrently use NFC to make connections
to separate endpoints. Since sockets have no implicit permission
handling, the kernel code then needs to implement a way to enforce
policy.

I still don't understand enough about NFC to judge which of
the two is better suited for the problem, but my feeling is
that a socket based implementation would be better if you expect
a lot of people to use it, while the main advantage of the
character device is its simplicity, so that would be preferred
if you only expect a very small set of possible applications
for this.

	Arnd