On 3/2/23 13:27, Stephen Boyd wrote:

Quoting Rob Herring (2023-03-02 09:32:09)

...

On Thu, Mar 2, 2023 at 2:14 AM David Gow davidgow@google.com wrote:

...

On Thu, 2 Mar 2023 at 09:38, Stephen Boyd sboyd@kernel.org wrote:

...

This patch series adds unit tests for the clk fixed rate basic type and the clk registration functions that use struct clk_parent_data. To get there, we add support for loading a DTB into the UML kernel that's running the unit tests along with probing platform drivers to bind to device nodes specified in DT.

With this series, we're able to exercise some of the code in the common clk framework that uses devicetree lookups to find parents and the fixed rate clk code that scans devicetree directly and creates clks. Please review.

Thanks Stephen -- this is really neat!

This works well here, and I love all of the tests for the KUnit/device-tree integration as well.

I'm still looking through the details of it (alas, I've mostly lived in x86-land, so my device-tree knowledge is, uh, spotty to say the least), but apart from possibly renaming some things or similarly minor tweaks, I've not got any real suggestions thus far.

I do wonder whether we'll want, on the KUnit side, to have some way of supporting KUnit device trees on non-UML architecctures (e.g., if we need to test something architecture-specific, or on a big-endian platform, etc), but I think that's a question for the future, rather than something that affects this series.

I'll say that's a requirement. We should be able to structure the tests to not interfere with the running system's DT. The DT unittest does that.

That could be another choice in the unit test choice menu. CONFIG_OF_KUNIT_NOT_UML that injects some built-in DTB overlay on an architecture that wants to run tests.

...

As a side topic, Is anyone looking at getting UML to work on arm64? It's surprising how much x86 stuff there is which is I guess one reason it hasn't happened.

I've no idea but it would be nice indeed.

...

...

Similarly, I wonder if there's something we could do with device tree overlays, in order to make it possible for tests to swap nodes in and out for testing.

Yes, that's how the DT unittest works. But it is pretty much one big overlay (ignoring the overlay tests). It could probably be more modular where it is apply overlay, test, remove overlay, repeat.

I didn't want to rely on the overlay code to inject DT nodes. Having tests written for the fake KUnit machine is simple. It closely matches how clk code probes the DTB and how nodes are created and populated on the platform bus as devices. CLK_OF_DECLARE() would need the overlay to be applied early too, which doesn't happen otherwise as far as I know.

But perhaps this design is too much of an end-to-end test and not a unit test? In the spirit of unit testing we shouldn't care about how the node is added to the live devicetree, just that there is a devicetree at all.

Supporting overlays to more easily test combinations sounds like a good idea. Probably some kunit_*() prefixed functions could be used to

In an imaginary world where overlay support was completed, then _maybe_.

To me, the most important environment to test is where the devictree data is populated in early boot from an FDT. This is the environment that drivers currently exist in.

Populating devicetree data via an overlay adds in the functioning of the overlay apply code (and how the rules behind that functioning may differ from devicetree data populated in early boot from an FDT).

In an ideal world where overlay support was completed, most or all of the devicetree tests that were performed against the devicetree data populated in early boot from an FDT would be repeated, but against comparable devicetree data populated via an overlay load. The tests with the overlay data may have to be aware of some differences in how an overlay load processes an FDT vs how the early boot processing of an FDT behaves. This extra testing would verify that the overlay environment behaves the same as the non-overlay environment (with some known exceptions due to overlay policies).

Overlay support is not complete:

https://elinux.org/Device_Tree_Reference#Mainline_Linux_Support

https://elinux.org/Frank%27s_Evolving_Overlay_Thoughts

-Frank

apply a test managed overlay and automatically remove it when the test is over would work. The clk registration tests could use this API to inject an overlay and then manually call the of_platform_populate() function to create the platform device(s). The overlay could be built in drivers/clk/ too and then probably some macroish function can find the blob and apply it.

Is there some way to delete the platform devices that we populate from the overlay? I'd like the tests to be hermetic.