Re: [RFC PATCHv2 1/4] drivers/otp: add initial support for OTP memory

[Arnd Bergmann <arnd@arndb.de>]

On Saturday 26 March 2011, Greg KH wrote:
> On Sat, Mar 26, 2011 at 12:25:03PM +0100, Arnd Bergmann wrote:
> > On Saturday 26 March 2011 00:28:47 Greg KH wrote:
> > 
> > > Yes, that is how it used to be, but then it turns out that both of them
> > > are really just "subsystems" as far as it all goes.  They are just ways
> > > that devices are bound to drivers in a logical manner.  We have patches
> > > floating around that get rid of both busses and classes to merge them
> > > together, which is the end goal here.  I know Kay has posted detailed
> > > reasons for why this all is on lkml in the past, and had working code
> > > about 5 years ago, it's just been slow going...
> > 
> > How will that work? I suppose we can't really change the directory
> > structure anyway, so to users it will still look like it does today,
> > even if the kernel just uses the same code for bus and class internally.
> 
> Yes.  But we can have symlinks instead of "real" /sys/class and
> /sys/bus directories like we do for /sys/block today.  All of the major
> tools that use sysfs work properly with a /sys/subsystem/ setup today
> thanks to Kay's efforts.

I see. The migration from /sys/bus to /sys/subsystem plus symlinks
makes sense, but I still think that having symlinks that make sense
would be better than having symlinks that are purely there for historical
reasons.

One major flaw I see with registering abstract subsystems as a bus is
that a bus connected to the concept of a device_driver, and there is
a list of drivers with their respective devices in /sys/bus/*/drivers,
which would always be empty in this case. Similarly, the
/sys/bus/*/devices/*/driver symlinks for these subsystems would point
to drivers on other buses, unlike the respective symlinks for real
buses.

	Arnd