percpu: fix synchronization between chunk->map_extend_work and chunk destruction

Atomic allocations can trigger async map extensions which is serviced
by chunk->map_extend_work.  pcpu_balance_work which is responsible for
destroying idle chunks wasn't synchronizing properly against
chunk->map_extend_work and may end up freeing the chunk while the work
item is still in flight.

This patch fixes the bug by rolling async map extension operations
into pcpu_balance_work.

Change-Id: Id41817a5c5f2c9ed9b19022df0c0ff7daacf6a71
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-and-tested-by: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Reported-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Sasha Levin <sasha.levin@oracle.com>
Cc: stable@vger.kernel.org # v3.18+
Fixes: 9c824b6a ("percpu: make sure chunk->map array has available space")
Git-repo: http://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git


Git-commit: 4f996e234dad488e5d9ba0858bc1bae12eff82c3
Signed-off-by: Dennis Cagle <d-cagle@codeaurora.org>