Re: [PATCH v2 01/11] mtd: add get/set of_node/flash_node helpers

[Brian Norris <computersforpeace@gmail.com>]

On Mon, Nov 02, 2015 at 12:27:58AM +0100, Boris Brezillon wrote:
> On Fri, 30 Oct 2015 20:33:20 -0700
> Brian Norris <computersforpeace@gmail.com> wrote:
> 
> > We are going to begin using the mtd->dev.of_node field for MTD device
> > nodes, so let's add helpers for it. Also, we'll be making some
> > conversions on spi_nor (and nand_chip eventually) too, so get that ready
> > with their own helpers.
> > 
> > Signed-off-by: Brian Norris <computersforpeace@gmail.com>
> > ---
> > v2: new
> > 
> >  include/linux/mtd/mtd.h     | 11 +++++++++++
> >  include/linux/mtd/nand.h    | 11 +++++++++++
> >  include/linux/mtd/spi-nor.h | 11 +++++++++++
> >  3 files changed, 33 insertions(+)
> > 
> > diff --git a/include/linux/mtd/mtd.h b/include/linux/mtd/mtd.h
> > index f17fa75809aa..cc84923011c0 100644
> > --- a/include/linux/mtd/mtd.h
> > +++ b/include/linux/mtd/mtd.h
> > @@ -254,6 +254,17 @@ struct mtd_info {
> >  	int usecount;
> >  };
> >  
> > +static inline void mtd_set_of_node(struct mtd_info *mtd,
> > +				   struct device_node *np)
> > +{
> > +	mtd->dev.of_node = np;
> 
> Maybe we should retain a reference to the device_node here (IOW,
> replace '= np' by '= of_node_get(np)'). Of course this implies calling
> of_node_put() when the mtd device is unregistered.

I'll admit I never really followed (nor verified) the OF get/put logic
that well. But I suppose that makes sense.

> > +}
> > +
> > +static inline struct device_node *mtd_get_of_node(struct mtd_info *mtd)
> > +{
> > +	return mtd->dev.of_node;
> 
> Not sure this is relevant to to the same here before returning the
> device_node because it's mostly used by the MTD drivers and those are
> the ones who called mtd_set_of_node() in the first place, so we can
> assume it's safe to return an non-retained reference to a device_node.

I think once the node has been "set", its consumers should stay in sync
with the lifetime of the MTD; so I think just that initial refcount is
sufficient.

IOW, if MTD drivers are still fiddling with an MTD's of_node after the
MTD is unregistered, then we've failed somewhere else.

Brian