[LTP] [PATCH 0/2] mem: ksm: race condition fixes

[Andrea Arcangeli]

Hello,

some false positives triggered the LTP ksm?? tests on some arch with
slower CPU. The volatile pages weren't zero during the checks, as they
should have been.

Volatile pages left around means there are orphaned rmap_items, that
can happen if the KSM scans aren't complete.

I believe the problem is with the waiting that wasn't waiting two full
passes, but it only waited until the values stopped changing after 10
seconds of wait time. Plus the ksm?? children were still running by
the time the wait time started.

The first patch takes care of waiting two scans (more efficiently,
with a polling of 1 sec). That change then exposes the problem with
the ksm?? children not being stopped before we start waiting, and the
second patch fixes that.

After applying these two fixes, we had hundred of successful runs on
the same system that showed volatile pages not zero, without any
further failures.

Try #2, first submit bounced because I thought I subscribed, but I was
still one click away.

Thanks,
Andrea

Andrea Arcangeli (2):
  mem: ksm: fix volatile page false positive
  mem: ksm: fixes for race conditions

 testcases/kernel/mem/lib/mem.c | 93 ++++++++++++++++++++++--------------------
 1 file changed, 49 insertions(+), 44 deletions(-)