Re: [PATCH v3 3/5] i3c: master: add i3c_for_each_dev helper

[Boris Brezillon]

On Fri, 21 Feb 2020 17:44:28 +0100
Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx> wrote:

> On Fri, 21 Feb 2020 13:59:11 +0100
> Boris Brezillon <boris.brezillon@xxxxxxxxxxxxx> wrote:
> 
> > On Fri, 21 Feb 2020 12:52:29 +0100
> > Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> >   
> > > On Fri, Feb 21, 2020 at 11:47:22AM +0000, Vitor Soares wrote:    
> > > > Hi Greg,
> > > > 
> > > > From: Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx>
> > > > Date: Wed, Feb 19, 2020 at 07:35:48
> > > >       
> > > > > On Wed, Feb 19, 2020 at 01:20:41AM +0100, Vitor Soares wrote:      
> > > > > > Introduce i3c_for_each_dev(), an i3c device iterator for use by i3cdev.
> > > > > > 
> > > > > > Signed-off-by: Vitor Soares <vitor.soares@xxxxxxxxxxxx>
> > > > > > ---
> > > > > >  drivers/i3c/internals.h |  1 +
> > > > > >  drivers/i3c/master.c    | 12 ++++++++++++
> > > > > >  2 files changed, 13 insertions(+)
> > > > > > 
> > > > > > diff --git a/drivers/i3c/internals.h b/drivers/i3c/internals.h
> > > > > > index bc062e8..a6deedf 100644
> > > > > > --- a/drivers/i3c/internals.h
> > > > > > +++ b/drivers/i3c/internals.h
> > > > > > @@ -24,4 +24,5 @@ int i3c_dev_enable_ibi_locked(struct i3c_dev_desc *dev);
> > > > > >  int i3c_dev_request_ibi_locked(struct i3c_dev_desc *dev,
> > > > > >  			       const struct i3c_ibi_setup *req);
> > > > > >  void i3c_dev_free_ibi_locked(struct i3c_dev_desc *dev);
> > > > > > +int i3c_for_each_dev(void *data, int (*fn)(struct device *, void *));
> > > > > >  #endif /* I3C_INTERNAL_H */
> > > > > > diff --git a/drivers/i3c/master.c b/drivers/i3c/master.c
> > > > > > index 21c4372..8e22da2 100644
> > > > > > --- a/drivers/i3c/master.c
> > > > > > +++ b/drivers/i3c/master.c
> > > > > > @@ -2640,6 +2640,18 @@ void i3c_dev_free_ibi_locked(struct i3c_dev_desc *dev)
> > > > > >  	dev->ibi = NULL;
> > > > > >  }
> > > > > >  
> > > > > > +int i3c_for_each_dev(void *data, int (*fn)(struct device *, void *))
> > > > > > +{
> > > > > > +	int res;
> > > > > > +
> > > > > > +	mutex_lock(&i3c_core_lock);
> > > > > > +	res = bus_for_each_dev(&i3c_bus_type, NULL, data, fn);
> > > > > > +	mutex_unlock(&i3c_core_lock);      
> > > > > 
> > > > > Ick, why the lock?  Are you _sure_ you need that?  The core should
> > > > > handle any list locking issues here, right?      
> > > > 
> > > > I want to make sure that no new devices (eg: Hot-Join capable device) are 
> > > > added during this iteration and after this call, each new device will 
> > > > release a bus notification.
> > > >       
> > > > > 
> > > > > I don't see bus-specific-locks around other subsystem functions that do
> > > > > this (like usb_for_each_dev).      
> > > > 
> > > > I based in I2C use case.      
> > > 
> > > Check to see if this is really needed, for some reason I doubt it...    
> > 
> > Can we please try the spidev approach before fixing those problems. None
> > of that would be needed if we declare the i3cdev driver as a regular
> > i3c_device_driver and let user space bind devices it wants to expose
> > through the sysfs interface. As I said earlier, we even have all the
> > pieces we need to automate that using a udev rule, and the resulting
> > patchset would be 'less invasive'/simpler for pretty much the same
> > result.  
> 
> So, I went ahead and implemented it the way I suggest. The diffstat is
> not representative here (though it's still in favor of this new version)
> since I also changed the way we expose/handle SDR transfers. What's
> most important IMO is the fact that
> 
> * we no longer need to access the internal I3C API
> * we no longer need to care about transitions between i3cdev and
>   other drivers (the core guarantees that a device is always bound to at
>   most one driver)
> * the registration/unregistration procedure is simplified

Looks like I was wrong, there's no way to disable auto-binding and keep
manual binding possible (they both use the bus->match() hook to know if
a dev can be attached to a driver, and the type of binding is not
specified here). There are of course other options to do this:

* add a i3c_driver->match() hook and expose driver-specific sysfs knobs
  to explicitly add devices to attach to this driver
* make the i3cdev module register one driver per-device, each time with
  its own i3c_device_id table and add sysfs knobs to trigger this
  registration

but none of these option sounds reasonable to me, so let's pursue with
your approach.