On Fri, 25 Mar 2022, Hugh Dickins wrote:
> On Fri, 25 Mar 2022, Mike Galbraith wrote:
>
> > Greetings,
> >
> > $subject bisected in a kvm ala:
> >
> > leap153:/usr/local/ltp # cat testme
> > export PATH=$PATH:`pwd`/testcases/bin
> > memcg_stat_test.sh
> > leap153:/usr/local/ltp # . ./testme
> >
> > Usually leads to...
> > memcg_stat_test 3 TINFO: Test unevictable with MAP_LOCKED
> > memcg_stat_test 3 TINFO: Running memcg_process --mmap-lock1 -s 135168
> > memcg_stat_test 3 TINFO: Warming up pid: 3460
> > memcg_stat_test 3 TINFO: Process is still here after warm up: 3460
> > memcg_stat_test 3 TFAIL: unevictable is 122880, 135168 expected
> > ...but may lead to...
> > memcg_stat_test 4 TINFO: Test unevictable with mlock
> > memcg_stat_test 4 TINFO: Running memcg_process --mmap-lock2 -s 135168
> > memcg_stat_test 4 TINFO: Warming up pid: 4271
> > memcg_stat_test 4 TINFO: Process is still here after warm up: 4271
> > memcg_stat_test 4 TFAIL: unevictable is 122880, 135168 expected
> > ...or both. A wee bit flaky.
> >
> > I wanted to verify with a revert on top of 85c7000fda00, but while the
> > revert patch applied, the result didn't boot. Config is full distro.
>
> Thanks a lot for spotting that. I'll have no trouble reproducing it here,
> looking through my old LTP test results. I never noticed because I'm used
> to memcg_stat failing - but looking closer, that's been because I'm usually
> running with THP shmem_enabled "force", which causes memcg_stat_test 1 to
> fail with a bigger number than expected (understandably): memcg_stat_test
> 3 and 4 failures are new to mlock/munlock changes.
>
> It will (almost certainly) be a pagevec draining issue, to be fixed by a
> strategically placed lru_add_drain() or mlock_page_drain(). I did have
> more of those in for a while, before understanding and arriving at
> b74355078b65 ("mm/munlock: page migration needs mlock pagevec drained");
> and with that fix, hadn't noticed the need for more, so left them out
> until proven desirable.
>
> If it's as I expect, then it's worth doing: not just to pass an LTP test,
> but more generally a good thing. I'll play around in the next few days
> and post a patch once I'm satisfied.
>
> Regarding your bisection and revert of b67bf49ce7aa ("mm/munlock: delete
> FOLL_MLOCK and FOLL_POPULATE"). I'm glad to hear that you got a build
> error trying to revert that one commit: not a supported combination!
> Maybe not too far wrong, but I wouldn't trust it.
>
> But yes, I can see that the revert will bring in an lru_add_drain()
> per page, so that fits with my guess above.
Right, I was easily able to reproduce those failures; and happily the
patch I had earlier, but left out, indeed fixes them as expected:
follows now.
Thanks,
Hugh
