On Fri, Apr 10, 2020 at 01:51:58AM +0200, Max Staudt wrote: > On 4/9/20 2:16 PM, Heikki Krogerus wrote: > > On Thu, Apr 09, 2020 at 01:37:35PM +0300, Andy Shevchenko wrote: > >> Heikki, am I correct? > > > > In this case it should be possible supply a handle to a software node > > with the board info. That should then later replace the fwnode and > > properties members once the existing code is converted: > > > > [... snip sample patch ...] > > > > I2C core would then need to take care of registering that swnode of > > course. > > Are you saying that the comment in property.c is correct, and > i2c_new_client_device() shall *not* call device_add_properties() ? > > I mean, the code works and stuff, except that the swnode that > device_add_properties() created won't be freed as far as I can see. They actually are freed when device_del() is finally called, which is pretty damn confusing. That is actually one of the reasons why we should avoid the old device_add/del_properties() API. > In other words, should the current properties code in i2c_new_client_device() > be replaced by something that creates a swnode, just like the i2c-icy driver > currently does manually when it instantiates the ltc2990 I2C client? Yes. The subsystem needs to take care of that, not the drivers. thanks, -- heikki
