Accessing rpmsg_device in sysfs attribute functions

[Greg KH]

On Tue, Mar 24, 2020 at 01:05:48PM +0000, Pelle Windestam wrote:
> > From: Greg KH <greg at kroah.com>
> > Sent: Tuesday, 24 March 2020 13:31
> > To: Pelle Windestam <Pelle.Windestam at tagmaster.com>
> > Cc: kernelnewbies at kernelnewbies.org
> > Subject: Re: Accessing rpmsg_device in sysfs attribute functions
> > 
> > On Tue, Mar 24, 2020 at 12:02:15PM +0000, Pelle Windestam wrote:
> > > > On Tue, Mar 24, 2020 at 05:34:44AM +0000, Pelle Windestam wrote:
> > > > > Hi,
> > > > >
> > > > > I am trying to develop a simple driver for the rpmsg bus, in order
> > > > > to send
> > > > various commands from user space in Linux to a secondary CPU (A Cortex
> > M4).
> > > > I'm trying to keep things as simple as possible, so my idea was to
> > > > create a driver that just has a few attributes which can be set in
> > > > /sys which would trigger commands to be sent to the M4 CPU. I have
> > > > the communication between the CPU:s up and running, but where I'm
> > > > having trouble moving forward is how to access the "struct
> > > > rpmsg_device *" that I need in order to communicate with the
> > > > endpoint for the M4 CPU from the store/show function of the sysfs
> > > > attributes. What my driver does is to register a rpmsg_driver in the
> > > > init
> > > > function:
> > > > >
> > > > > register_rpmsg_driver(&pwm_rpmsg_driver);
> > > > >
> > > > > the device_driver member of my rpmsg_driver struct has its groups
> > > > > member
> > > > set to my driver attribute groups array:
> > > > > static struct rpmsg_driver pwm_rpmsg_driver = {
> > > > > 	.probe = pwm_rpmsg_probe,
> > > > > 	.remove = pwm_rpmsg_remove,
> > > > > 	.callback = pwm_rpmsg_cb,
> > > > > 	.id_table = pwm_rpmsg_device_id_table,
> > > > > 	.drv = {
> > > > > 		.groups = driver_pwm_groups,
> > > > > 		.name = "pwm_rpmsg",
> > > > > 	},
> > > > > };
> > > > >
> > > > > My issue is that that I am not sure how to access the struct
> > > > > "rpmsg_device
> > > > *" (i.e. from the probe() function) in the show/store functions for
> > > > the sysfs attributes, which have a "struct device_driver *" argument:
> > > >
> > > > That is because you have created a driver attribute, not a device
> > attribute.
> > > > Create device attributes and you should be fine, they bind to the
> > > > device your driver is passed.
> > >
> > > Thanks! Changing them to device attributes was a breeze. Now I am
> > > slightly confused about the "struct device *" argument to the
> > > store/show functions. I was under the impression that this would be
> > > the "struct device" in the struct rpmsg_device, (which would let me
> > > get the struct rpmsg_device using container_of()?), but it appears to
> > > be some completely other device (by looking at the pointer address). I
> > > have tried searching the kernel code for similar example, but I have
> > > not found anything so far. It feels like I am stumbling a bit in the
> > > dark here, looking for my rpmsg_device.
> > 
> > It's a bit hard to figure out what exactly you are doing here without a
> > pointer to the code itself :)
> > 
> > Are you sure you aren't pointing the platform device accidentally?
> > 
> > thanks,
> > 
> > greg k-h
> 
> I suppose anything is possible 😊 Here is the actual code if you want to have a look:
> https://gist.github.com/iceaway/9900a9c2dd221eb836c5acda49f5d688

Odd, that "should" work.  But the pwm core is strange, I suggest posting
it to the pwm driver mailing list and asking for help there, as there
might be some odd "wrapper" structures involved here as can be seen by
the pwm core sysfs file accesses.

good luck!

greg k-h