On Mon, Oct 29, 2012 at 09:05:53PM +0200, Felipe Balbi wrote:

Hi,

On Mon, Oct 29, 2012 at 10:06:33AM -0700, Dmitry Torokhov wrote:

[ big snip ]

quoted

quoted

quoted

+static int __devexit smsc_remove(struct platform_device *pdev)
+{

shouldn't you unregister the input device here ??

And that is why I do not like devm_* interface myself... But no, since
input device was allocated with devm_input_allocate_device() it does not
need to be explicitly freed or unregistered.

IMHO, that's a fragility on current devm implementation for input
devices, then.

devm_input_allocate_device() is *only* allocating the input device (at
least judging by the name). Looks like you should introduce
devm_input_register_device() ? What happens if I
devm_input_allocate_device() but don't go as far as
input_register_device() (some error happens in-between) ?

It will be freed automatically.

I'm sure you have some proper handling for it, but it's quite misleading
the way this was implemented.

Well, I could add devm_input_register_device(),
devm_input_unregister_device(), devm_input_free_device() and then add
checks to "normal" input_register_device(), input_unregister_device()
and input_free_device() to throw warnings and errors if they are used
with managed resources, and similarly add checks to devm_* API so that
they are not called with unmanaged devices and make a big mess out of
it.

_OR_ I could just add one new call devm_input_allocate_device() to
create managed input devices and make the rest of old API work with both
managed and unmanaged input devices. I think the latter is much better.

Thanks.

-- 
Dmitry