On Sat, Mar 4, 2023 at 3:08 PM Linus Torvalds
[off-list ref] wrote:

Well, this particular patch at least boots for me for my normal
config. Not that I've run any extensive tests, but I'm writing this
email while running this patch, so ..

Hmm. I enabled the KUNIT tests, and used an odd CONFIG_NR_CPUS to test
this a bit more.

So in my situation, I have 64 threads, and so nr_cpu_ids is 64, and
CONFIG_NR_CPUS is 150.

Then one cpumask KUNIT test fails with

     # test_cpumask_weight: EXPECTATION FAILED at lib/cpumask_kunit.c:70
                  Expected ((unsigned int)150) == cpumask_weight(&mask_all), but
                      ((unsigned int)150) == 150 (0x96)
                      cpumask_weight(&mask_all) == 64 (0x40)
              &mask_all contains CPUs 0-63

but I think that's actually a KUNIT test bug.

The KUNIT test there is

        KUNIT_EXPECT_EQ_MSG(test, nr_cpumask_bits,
cpumask_weight(&mask_all), MASK_MSG(&mask_all));

and it should *not* expect the cpumask weight to be nr_cpumask_bits,
it should expect it to be nr_cpu_ids.

That only matters now that nr_cpumask_bits isn't the same as nr_cpu_ids./

Anyway, I still think that patch of mine is fine, and I think this
test failure only ends up being about the test, not the patch.

            Linus