On Tue, 30 Jul 2024 at 23:39, Shuah Khan <skhan@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On 7/29/24 02:15, Muhammad Usama Anjum wrote:
> > On 7/27/24 11:10 PM, Yury Norov wrote:
> >> On Fri, Jul 26, 2024 at 01:26:48PM -0600, Shuah Khan wrote:
> >>> On 7/26/24 05:06, Muhammad Usama Anjum wrote:
> >>>> In this series, test_bitmap is being converted to kunit test. Multiple
> >>>> patches will make the review process smooth.
> >>>>
> >>>> - Patch-1: Convert the tests in lib/test_bitmap.c to kunit
> >>>> - Patch-2: Rename the lib/test_bitmap.c to lib/bitmap_kunit.c and other
> >>>> configuration options
> >>>> - Patch-3: Remove the bitmap.sh selftest
> >>>>
> >>>> Muhammad Usama Anjum (3):
> >>>> bitmap: convert test_bitmap to KUnit test
> >>>> bitmap: Rename module
> >>>> selftests: lib: remove test_bitmap
> >>>>
> >>>> MAINTAINERS | 2 +-
> >>>> lib/Kconfig.debug | 15 +-
> >>>> lib/Makefile | 2 +-
> >>>> lib/{test_bitmap.c => bitmap_kunit.c} | 624 ++++++++++++--------------
> >>>> tools/testing/selftests/lib/Makefile | 2 +-
> >>>> tools/testing/selftests/lib/bitmap.sh | 3 -
> >>>> tools/testing/selftests/lib/config | 1 -
> >>>> 7 files changed, 295 insertions(+), 354 deletions(-)
> >>>> rename lib/{test_bitmap.c => bitmap_kunit.c} (70%)
> >>>> delete mode 100755 tools/testing/selftests/lib/bitmap.sh
> >>>>
> >>>
> >>> Can you tell me how this conversion helps?
> >>>
> >>> It is removing the ability to run bitmap tests during boot.
> >>> It doesn't make sense to blindly convert all test under lib
> >>> to kunit - Nack on this change or any change that takes away
> >>> the ability to run tests and makes them dependent on kunit.
> >>
> >> Hi Muhammad,
> >>
> >> In addition to Shuah's and John's reasoning. This patch wipes the
> >> test history (git blame will point on you for most of the test),
> > When files are renamed, their history isn't lost. We just need to use
> > --follow option with git log to get complete history[1].
> >
> >> breaks boot-time testing support, messes with config names and
> >> usability, and drops kselftest support for ... exactly, what?
> > AFAIU the kselftest wasn't detected the test results that's why I started
> > thinking on which could be best way to detect if any failure happens in
> > this test. Triggering the test from kselftest doesn't grantee the test it
> > would pass every time until we check results. For this kind of in-kernel
> > testing, kunit is best suites. Please find earlier discussion [2].
> >
>
> KUnit isn't idea for cases where people would want to check a subsystem
> on a running kernel - KUnit covers some use-cases and kselftest covers
> others.
Thanks, Shuah. That clarifies a lot. So the use-case we're concerned about is:
- We have an existing running kernel, which has CONFIG_KUNIT=n.
- Something breaks, and we want to check (for example) test_bitmap.
- We can modprobe test_bitmap on the current kernel.
That's definitely something KUnit can't do at the moment (though the
workaround of always building with CONFIG_KUNIT=m is working for Red
Hat and Android, and seems the closest equivalent). Maybe we need to
work out a way for KUnit tests to handle this case more smoothly.
Because I don't think there's anything special about this test which
makes this use-case more likely than any of the other KUnit tests we
have.
> What happens if we are debugging a problem that requires us to debug on
> a running system? Please don't go converting kselftest into kunit without
> understanding how these are intended to be used.
I don't think it's clear that this _is_ how the test is intended to be
used: the config help text (and earlier messages in this thread) were
all about running tests at boot, which is distinct from "debugging a
problem on a running system". I think this is where I've been most
confused. We've been happy to assume the "testing at runtime" case is
against a kernel with test configs enabled, too.
When it comes to test_bitmap specifically, I can't find any difference
(save possibly for a couple of performance test functions) between it
and any of the other tests we've converted to or written with KUnit in
the past. I'd imagine (and correct me if I'm wrong) that it's pretty
rare for there to be an issue which will be caught by test_bitmap on a
running kernel, but not on a fresh boot / in a kernel with KUnit
enabled. So the "debug a running system" situation seems to me only
decrease coverage of some very rare situations. (Of course, that's
ignoring that running the module against the current kernel might be
most convenient, but that's a separate issue from coverage.)
> Yes kselftest results need to be looked at. Write a parser which can
> be improved. What you are doing is reducing the coverage and talking
> away the ability to debug and test on running system.
>
> Fix what needs to be fixed instead of deleting tests.
>
> >>
> >> KUNIT engine here doesn't improve on readability, neither shorten
> >> the test length, to my taste.
> >>
> >> If you'd like to contribute to bitmaps testing - I'm all for that.
> >> This is the very core and performance-sensitive piece of kernel,
> >> and any extra-coverage is always welcome.
> >>
>
> +1 on this. Add new tests and look at the reports.
>
> >> But I think the best way would be either adding new cases to the
> >> existing test, or writing a new test, KUNIT-based, if you like.
> >>
> >
>
> +1
>
> As I mentioned in my earlier message, I am going to nack all patches
> that convert existing selftests to kunit such as this one.
I should point out that we've already been porting a lot of tests in
lib/ to KUnit for years, sometimes with the help of the test author,
sometimes by encouraging it as a task for new contributors. And while
most of those are not actually selftests (though we may have ported
one or two), they differ from test_bitmap only in that there's a
selftest wrapper script. The authors and users of those tests have
thus far found this a worthwhile trade -- I've not heard any
complaints -- so I'd rather we leave this as a "pause" on new
conversions while we sort this out, rather than do something drastic
like revert existing conversion.
Personally, I'd still prefer these sorts of conversions to go ahead --
I think the benefits outweightthe costs -- but I can totally get
behind doing it on a case-per-case basis. I'd be disappointed if we
NACKed changes like this in the case where the authors and users of
that individual test would like it to be converted, though.
Regardless, I think there's more discussion to be had here, and I'll
look into what changes we can make either to KUnit or to the testing
documentation to make sure we're all on the same page.
-- David
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature
