[LTP] [PATCH] ksm: fix segfault on s390

[Li Wang]

Luiz Capitulino <luizcap@redhat.com> wrote:

> > I might be a bit too picky:). So I compared the two approaches on a
> > 2 CPUs, KVM, x86_64 system:
> >
> > Per-block checking cost time:
> >     real 0m5.862s
> >     user 0m1.098s
> >     sys 0m1.505s
> >
> > Per-byte checking cost time:
> >    real    0m6.819s
> >    user    0m2.498s
> >    sys     0m1.495s
> >
> >  From the data, block-by-block checking can reduce the total execution
> > time by about 14% and reduce CPU usage by more than 35%, especially
> > in user-space calculations. This number may not be large, but considering
> > that tests are frequently run in CI, I think it would be a good thing if we can
> > reduce 1 second each time :).
>
> Just to make sure I understand: you measured total test run-time, correct?
> How many times did you run it?
>
> In any case, I'm not sure a 1 second run-time (or even CPU utilization) matters
> that much. You're running test code, you shouldn't expect otherwise unless you
> hit a very bad case (say something taking several hours to complete).
>
> The trade off is more complex code with bugs that can hide for 10+ years and
> take developer time to debug. Also, higher memory utilization: 's' doubles
> memory utilization per child only to do that check.

Ture, that's why the problem not been find so many years!

>
> So, I suggest we stick to the simpler code. Or, get it merged now (since it's
> fixing a bug and possibly making the code _faster_) and then you can optimize
> on top later if you like.

Ok, sounds reasonable.

Reviewed-by: Li Wang <liwang@redhat.com>

@Cyril, @Petr, I vote to merge this one (as it is) before our May release.

-- 
Regards,
Li Wang