[PATCH] NVMe: Fix possible scheduling while atomic error

[Ming Lin]

On Mon, May 23, 2016 at 7:55 AM, Keith Busch <keith.busch at intel.com> wrote:
>
>> > It's moved out of that function holding the rcu read lock since
>> > blk_mq_stop_hw_queues potentially sleeps.
>>
>> blk_mq_stop_hw_queues is a loop around blk_mq_stop_hw_queue,
>> which itself does two calls to cancel_delayed_work and a set_bit.
>> None of them can block.

It's blk_mq_cancel_requeue_work() that potentially sleeps.