Re: [PATCH 1/2] drivers: base: swnode: link devices to software nodes

[Dmitry Torokhov <dmitry.torokhov@xxxxxxxxx>]

On Tue, Jul 30, 2019 at 04:49:50PM +0200, Rafael J. Wysocki wrote:
> On 7/30/2019 1:52 PM, Heikki Krogerus wrote:
> > On Mon, Jul 29, 2019 at 03:15:32PM +0200, Dmitry Torokhov wrote:
> > > On Mon, Jul 29, 2019 at 03:07:15PM +0300, Heikki Krogerus wrote:
> > > > On Sat, Jul 13, 2019 at 12:52:58AM -0700, Dmitry Torokhov wrote:
> > > > > It is helpful to know what device, if any, a software node is tied to, so
> > > > > let's store a pointer to the device in software node structure. Note that
> > > > > children software nodes will inherit their parent's device pointer, so we
> > > > > do not have to traverse hierarchy to see what device the [sub]tree belongs
> > > > > to.
> > > > > 
> > > > > We will be using the device pointer to locate GPIO lookup tables for
> > > > > devices with static properties.
> > > > > 
> > > > > Signed-off-by: Dmitry Torokhov <dmitry.torokhov@xxxxxxxxx>
> > > > > ---
> > > > >   drivers/base/property.c  |  1 +
> > > > >   drivers/base/swnode.c    | 35 ++++++++++++++++++++++++++++++++++-
> > > > >   include/linux/property.h |  5 +++++
> > > > >   3 files changed, 40 insertions(+), 1 deletion(-)
> > > > > 
> > > > > diff --git a/drivers/base/property.c b/drivers/base/property.c
> > > > > index 348b37e64944..3bc93d4b35c4 100644
> > > > > --- a/drivers/base/property.c
> > > > > +++ b/drivers/base/property.c
> > > > > @@ -527,6 +527,7 @@ int device_add_properties(struct device *dev,
> > > > >   	if (IS_ERR(fwnode))
> > > > >   		return PTR_ERR(fwnode);
> > > > > +	software_node_link_device(fwnode, dev);
> > > > >   	set_secondary_fwnode(dev, fwnode);
> > > > >   	return 0;
> > > > >   }
> > > > > diff --git a/drivers/base/swnode.c b/drivers/base/swnode.c
> > > > > index 7fc5a18e02ad..fd12eea539b6 100644
> > > > > --- a/drivers/base/swnode.c
> > > > > +++ b/drivers/base/swnode.c
> > > > > @@ -24,6 +24,9 @@ struct software_node {
> > > > >   	/* properties */
> > > > >   	const struct property_entry *properties;
> > > > > +
> > > > > +	/* device this node is associated with */
> > > > > +	struct device *dev;
> > > > >   };
> > > > Let's not do that! The nodes can be, and in many cases are, associated
> > > > with multiple devices.
> > > They do? Where? I see that set of properties can be shared between
> > > several devices, but when we instantiate SW node we create a new
> > > instance for device. This is also how ACPI and OF properties work; they
> > > not shared between devices (or, rather, the kernel creates distinct _and
> > > single_ devices for instance of ACPI or OF node). I do not think we want
> > > swnodes work differently from the other firmware nodes.
> > Having multiple devices linked to a single node is quite normal. Most
> > multifunctional devices will share a single node. The USB port devices
> > will share their node (if they have one) with any device that is
> > attached to them. Etc.
> > 
> > If you want to check how this works with ACPI, then find
> > "physical_node" named files from sysfs. The ACPI node folders in sysfs
> > have symlinks named "physical_node<n>" for every device they are bind
> > to. The first one is named just "physical_node", the second
> > "physical_node1", the third "physical_node2", and so on.
> > 
> > > > Every device is already linked with the software node kobject, so
> > > > isn't it possible to simply walk trough those links in order to check
> > > > the devices associated with the node?
> > > No, we need to know the exact instance of a device, not a set of
> > > devices, because even though some properties can be shared, others can
> > > not. For example, even if 2 exactly same touch controllers in the system
> > > have "reset-gpios" property, they won't be the same GPIO for the both of
> > > them.
> > I don't think I completely understand the use case you had in mind for
> > this API, but since you planned to use it with the GPIO lookup tables,
> > I'm going to assume it's not needed after all. Let's replace those
> > with the references instead like I proposed in my reply to the 2/2
> > patch.
> > 
> > Linking a single device with a node like that is in any case not
> > acceptable nor possible.
> > 
> I think I need to withdraw my ACK here at this point.

OK, fair enough, I'll see if I can make the references that Heikki
mentioned work for me.

Thanks.

-- 
Dmitry