Re: [RFC] of/platform: Create device link between simple-mfd and its children

[Nicolas Saenz Julienne]

Hi Uwe,
Sorry for the late reply, got distracted with other stuff.

On Mon, 2020-10-19 at 08:52 +0200, Uwe Kleine-König wrote:
> On Fri, Oct 16, 2020 at 05:26:56PM +0200, Nicolas Saenz Julienne wrote:
> > On Fri, 2020-10-16 at 09:38 -0500, Rob Herring wrote:
> > > On Thu, Oct 15, 2020 at 6:43 AM Nicolas Saenz Julienne
> > > <nsaenzjulienne@xxxxxxx> wrote:
> > > > 'simple-mfd' usage implies there might be some kind of resource sharing
> > > > between the parent device and its children.
> > > 
> > > It does? No! The reason behind simple-mfd was specifically because
> > > there was no parent driver or dependency on the parent. No doubt
> > > simple-mfd has been abused.
> > 
> > Fair enough, so we're doing things wrong. Just for the record, I'm looking at
> > RPi´s firmware interface:
> > 
> > 	firmware: firmware {
> > 		compatible = "raspberrypi,bcm2835-firmware", "simple-mfd";
> > 		#address-cells = <1>;
> > 		#size-cells = <1>;
> > 		mboxes = <&mailbox>;
> > 
> > 		firmware_clocks: clocks {
> > 			compatible = "raspberrypi,firmware-clocks";
> > 			#clock-cells = <1>;
> > 		};
> > 
> > 		reset: reset {
> > 			compatible = "raspberrypi,firmware-reset";
> > 			#reset-cells = <1>;
> > 		};
> > 		[...]
> > 	};
> > 
> > Note that "raspberrypi,bcm2835-firmware" has a driver, it's not just a
> > placeholder. Consumer drivers get a handle to RPi's firmware interface through
> > the supplier's API, rpi_firmware_get(). The handle to firmware becomes
> > meaningless if it is unbinded, which I want to protect myself against.
> > 
> > A simpler solution would be to manually create a device link between both
> > devices ("raspberrypi,bcm2835-firmware" and "raspberrypi,firmware-clocks" for
> > example) upon calling rpi_firmware_get(). But I wanted to try addressing the
> > problem in a generic way first.
> 
> IMHO rpi_firmware_get() should get a reference on the firmware device
> (and call try_module_get()) which prevents unbinding it.

Yes, it seems the way to go. Just one last question WRT this, since
'drv->remove(dev)' can't fail should I just block until the reference count
hits zero?

Regards,
Nicolas

Attachment: signature.asc
Description: This is a digitally signed message part