On Sun, Sep 22, 2024 at 11:13 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote: > > On 9/21/24 23:16, Hyeonggon Yoo wrote: > > On Sun, Sep 22, 2024 at 6:25 AM Vlastimil Babka <vbabka@xxxxxxx> wrote: > >> > >> On 9/21/24 23:08, Guenter Roeck wrote: > >>> On 9/21/24 13:40, Vlastimil Babka wrote: > >>>> +CC kunit folks > >>>> > >>>> On 9/20/24 15:35, Guenter Roeck wrote: > >>>>> Hi, > >>>> > >>>> Hi, > >>>> > >>>>> On Wed, Aug 07, 2024 at 12:31:20PM +0200, Vlastimil Babka wrote: > >>>>>> Add a test that will create cache, allocate one object, kfree_rcu() it > >>>>>> and attempt to destroy it. As long as the usage of kvfree_rcu_barrier() > >>>>>> in kmem_cache_destroy() works correctly, there should be no warnings in > >>>>>> dmesg and the test should pass. > >>>>>> > >>>>>> Additionally add a test_leak_destroy() test that leaks an object on > >>>>>> purpose and verifies that kmem_cache_destroy() catches it. > >>>>>> > >>>>>> Signed-off-by: Vlastimil Babka <vbabka@xxxxxxx> > >>>>> > >>>>> This test case, when run, triggers a warning traceback. > >>>>> > >>>>> kmem_cache_destroy TestSlub_kfree_rcu: Slab cache still has objects when called from test_leak_destroy+0x70/0x11c > >>>>> WARNING: CPU: 0 PID: 715 at mm/slab_common.c:511 kmem_cache_destroy+0x1dc/0x1e4 > >>>> > >>>> Yes that should be suppressed like the other slub_kunit tests do. I have > >>>> assumed it's not that urgent because for example the KASAN kunit tests all > >>>> produce tons of warnings and thus assumed it's in some way acceptable for > >>>> kunit tests to do. > >>>> > >>> > >>> I have all tests which generate warning backtraces disabled. Trying to identify > >>> which warnings are noise and which warnings are on purpose doesn't scale, > >>> so it is all or nothing for me. I tried earlier to introduce a patch series > >>> which would enable selective backtrace suppression, but that died the death > >>> of architecture maintainers not caring and people demanding it to be perfect > >>> (meaning it only addressed WARNING: backtraces and not BUG: backtraces, > >>> and apparently that wasn't good enough). > >> > >> Ah, didn't know, too bad. > >> > >>> If the backtrace is intentional (and I think you are saying that it is), > >>> I'll simply disable the test. That may be a bit counter-productive, but > >>> there is really no alternative for me. > >> > >> It's intentional in the sense that the test intentionally triggers a > >> condition that normally produces a warning. Many if the slub kunit test do > >> that, but are able to suppress printing the warning when it happens in the > >> kunit context. I forgot to do that for the new test initially as the warning > >> there happens from a different path that those that already have the kunit > >> suppression, but we'll implement that suppression there too ASAP. > > > > We might also need to address the concern of the commit > > 7302e91f39a ("mm/slab_common: use WARN() if cache still has objects on > > destroy"), > > the concern that some users prefer WARN() over pr_err() to catch > > errors on testing systems > > which relies on WARN() format, and to respect panic_on_warn. > > > > So we might need to call WARN() instead of pr_err() if there are errors in > > slub error handling code in general, except when running kunit tests? > > > > If people _want_ to see WARNING backtraces generated on purpose, so be it. > For me it means that _real_ WARNING backtraces disappear in the noise. > Manually maintaining a list of expected warning backtraces is too maintenance > expensive for me, so I simply disable all kunit tests which generate > backtraces on purpose. That is just me, though. Other testbeds may have > more resources available and may be perfectly happy with the associated > maintenance cost. > > In this specific case, I now have disabled slub kunit tests, and, as > mentioned before, from my perspective there is no need to change the > code just to accommodate my needs. I'll do the same with all other new > unit tests which generate backtraces in the future, without bothering > anyone. > > Sorry for the noise. I don't think this was a noise :) IMO some people want to see WARNING during testing to catch errors, but not for the slub_kunit test case. I think a proper approach here would be suppressing warnings while running slub_kunit test cases, but print WARNING when it is not running slub_kunit test cases. That would require some work changing the slub error reporting logic to print WARNING on certain errors. Any opinions, Vlastimil? Thanks, Hyeonggon