Re: [PATCH v6 3/7] kunit: Add kunit wrappers for (root) device creation

[Greg Kroah-Hartman]

On Mon, Mar 27, 2023 at 02:34:02PM +0300, Matti Vaittinen wrote:
> A few tests need to have a valid struct device. One such example is
> tests which want to be testing devm-managed interfaces.
> 
> Add kunit wrapper for root_device_[un]register(), which create a root
> device and also add a kunit managed clean-up routine for the device
> destruction upon test exit.

I really do not like this as a "root device" is a horrible hack and
should only be used if you have to hang other devices off of it and you
don't have a real device to tie those devices to.

Here you are abusing it and attempting to treat it as a real device,
which it is not at all, because:

> Special note: In some cases the device reference-count does not reach
> zero and devm-unwinding is not done if device is not sitting on a bus.
> The root_device_[un]register() are dealing with such devices and thus
> this interface may not be usable by all in its current form. More
> information can be found from:
> https://lore.kernel.org/dri-devel/20221117165311.vovrc7usy4efiytl@houat/

See, not a real device, doesn't follow normal "struct device" rules and
lifetimes, don't try to use it for a test as it will only cause problems
and you will be forced to work around that in a test.

Do the right thing here, create a fake bus and add devices to it.

Heck, I'll even write that code if you want it, what's the requirement,
something like:
	struct device *kunit_device_create(struct kunit *test, const char *name);
	void kunit_device_destroy(struct device *dev);
?

Why do you want a "match" function?  You don't provide documentation
here for it so I have no idea.

Anything else needed?

> The use of root-devices in the kunit helpers is intended to be an
> intermediate solution to allow tests which do not require device to sit
> on a bus avoid directly abusing the root_device_[un]register() while
> proper kunit device solution is being worked on. Related discussion can be
> found from:
> https://lore.kernel.org/lkml/CABVgOSmx3A4Vwos2_8xO-XQrQAw5gvY0nc5zLpLmcJ7FtA-dTQ@xxxxxxxxxxxxxx/

Again, no, please let's not get this wrong now and say "we will fix this
later" as that's not how kernel development should work...

thanks,

greg k-h