In #4826 (and earlier with --strict), we introduced warnings for applying unregistered marks, because otherwise something like:

@pytest.mark.sloww  # sic!
def test_slow():
    pass

would be a silent failure, with the test not being run with -m slow.

However, there is a similar issue when using marks: If you end up e.g. accidentally typing -m "not sloww", this won't actually skip the slow tests, but you might not actually notice that this mark doesn't exist. Similarly, if you do -m "mark1 or mark2_with_typo", it's tricky to recognize that less tests are run than what was actually intended.

IMHO, such usage should raise a warning (or even an error?).