rhashtable concurrent writers issue, 4.2+

Hi,

After testing with commit 7def0f952eccdd0edb3c504f4dab35ee0d3aba1f ("lib: fix data race in rhashtable_rehash_one") on the current 4.4-rc4 kernel, I believe I am still seeing race conditions that seem to bite concurrent readers.

The reproducer is found in my stress-ng system stress tool:

git://kernel.ubuntu.com/cking/stress-ng.git

one needs libattr1-dev, libkeyutils-dev to build this.

run the concurrent procfs stressor on 2 CPUs:

./stress-ng --procfs 2

..sometimes this hangs w/o a kernel stack dump, but sometimes it oopses and one can see that we get a concurrent reader failure.

Colin