On Wed, Jul 7, 2021 at 5:23 PM André Almeida andrealmeid@riseup.net wrote:
Hi Daniel,
Às 03:48 de 07/07/21, Daniel Latypov escreveu:
On Mon, Jul 5, 2021 at 1:41 PM André Almeida andrealmeid@riseup.net wrote:
Hello,
We belong to two student groups, FLUSP [1] and LKCAMP [2], both of which are focused on sharing kernel and free software development knowledge and experience with fellow free software developers and newcomers.
As part of our efforts, we'll be organizing a KUnit hackathon in the next Saturday (July 10), where we intend to help newcomers convert existing runtime tests (the ones found at lib/) to KUnit and maybe create new ones. Depending on the number of attendees, a high volume of patches may be sent throughout the day. We will do our best to review all patches before they go to the kernel mailing lists hoping to avoid wasting your time with minor patching issues.
That sounds great!
:)
So we wanted to let you know of all this beforehand and give you the time to send any suggestions or comments on all this. For instance, we
Some random, unorganized thoughts:
- I (as a kernel newbie) found the ownership of tests under lib/ to be
a bit unclear.
Indeed, but we found that tests under lib/ are the easiest ones to convert, given that they deal with more "generic" programming aspects, like linked lists and sort, and doesn't requires subsystem specific knowledge.
Yeah, I think these are the best for a hackathon.
Plus they're a better fit for KUnit right now, since they're the more standalone bits of code. Writing tests for more subsystem-specific code can require a lot of groundwork, e.g. davidgow@google.com is trudging through what needs to be set in a superblock to write ext4 tests. Eventually, we'd ideally have some chunks of shared code to setup subsystem-specific structures for tests, but they don't exist yet.
I didn't articulate it well, but my concern was finding the "right" people to review the patches. It feels like an unfortunate risk that we'll have to take, having some patches potentially sit unreviewed for a while.
- Make sure to base on 5.14, there's some new features that could be of use
- QEMU support in kunit.py: it's no longer as big of a deal if we
can't get tests running under UML. Brendan's made it fairly easy to run tests in an x86 VM.
- Skipped tests support: can now mark a test skipped. Some of the
lib/ tests could benefit, e.g. test_hash.c
- another lib/ test conversion just landed in 5.14, [1], along with
kunit_kcalloc/kunit_malloc_array() helpers
- Documentation/dev-tools/kunit/running_tips.rst - has instructions
for how to generate code coverage reports using UML/kunit.py * This seems like it could be useful... [2]
Thanks for those info! I wasn't aware of those features.
- I feel that the hackathon will surface a number of feature requests
for KUnit. Some patches might be blocked on these features (like your uuid patch).
- Might be good to track these as they come up and note which
patches are blocked.
Good suggestion, I'll try to track this.
Thanks! Hopefully we can try and help get some of these resolved, but it might take a while.
- Should we make a decision beforehand wrt renaming test files and
kconfig options to match the style guide? In the most recent conversion [1], I left them as-is. The answer depends on the test owner, but we should probably be consistent with what we do in the v1's at least.
I believe that to left as-is is easier, and we will suggest people to do that.
- I assume test modules used in selftests (test_bitmap, test_printf,
etc.) should remain untouched for now ?
- KUnit does *not* return a non-zero exit code when you modprobe the
test, as the userspace part of those tests assumes.
- We could write a wrapper like kselftest/module.sh, but for KUnit
test modules, but this extra in-flight patch would become a dependency for maybe multiple other conversions.
- I'd like it if we had ^ eventually, but I don't know how others
feel about that.
Yeah, I don't think we would be able to deal with all of this in this hackathon, so we won't touch those tests for now.
- I had some small patches for converting test_min_heap.c and
test_sort.c that I should get around to sending out.
So this means that we shouldn't convert them, right?
Yeah, you can leave those ones to me. I'll try and send them out real soon so people can just check the mailing list to see what's in flight.
I'm not aware of anyone else working on converting the other tests right now, so everything else should be fair game.
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i... [2] I currently see Overall coverage rate: lines......: 16.4% (20034 of 122481 lines) functions..: 18.1% (2032 of 11205 functions) with the following kunitconfig CONFIG_KUNIT=y CONFIG_KUNIT_ALL_TESTS=y CONFIG_DEBUG_KERNEL=y CONFIG_DEBUG_INFO=y CONFIG_GCOV=y
This might be a neat way to motivate participants by showing #s going up. Note: I'm more interested in the # of covered functions than anything else. The %s are nice to look at, but not that useful since we're only compiling a small fraction of kernel code.
Cool! In the end of the hackathon we will compare the numbers to see how much we were able to do in the event.
may ask people to add a special tag to their patches so you may batch review them all at a later time if you wish.
Anyhow, we'd really appreciate having your opinion on this. Thanks!
[1] - https://flusp.ime.usp.br/ [2] - https://lkcamp.dev/
-- You received this message because you are subscribed to the Google Groups "KUnit Development" group. To unsubscribe from this group and stop receiving emails from it, send an email to kunit-dev+unsubscribe@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/kunit-dev/83ff3ed3-e004-29b2-b23c-a2ba4d5a....