nvme-pci: move nvme_kill_queues to nvme_remove_dead_ctrl
authorJianchao Wang <jianchao.w.wang@oracle.com>
Wed, 20 Jun 2018 05:42:22 +0000 (13:42 +0800)
committerChristoph Hellwig <hch@lst.de>
Thu, 21 Jun 2018 14:59:42 +0000 (16:59 +0200)
commit9f9cafc14016f23f982d3ce18f9057923bd3037a
tree15b08d09c35beacdd80cd36c25f8d0151540bde9
parent02d62a8bc48e92171c46540722e2d52ce77d87af
nvme-pci: move nvme_kill_queues to nvme_remove_dead_ctrl

There is race between nvme_remove and nvme_reset_work that can
lead to io hang.

nvme_remove                    nvme_reset_work
                               -> nvme_remove_dead_ctrl
                                 -> nvme_dev_disable
                                   -> quiesce request_queue
                                 -> queue remove_work
-> cancel_work_sync reset_work
-> nvme_remove_namespaces
  -> splice ctrl->namespaces
                               nvme_remove_dead_ctrl_work
                               -> nvme_kill_queues
  -> nvme_ns_remove               do nothing
    -> blk_cleanup_queue
      -> blk_freeze_queue

Finally, the request_queue is quiesced state when wait freeze,
we will get io hang here. To fix it, move the nvme_kill_queues
from nvme_remove_dead_ctrl_work to nvme_remove_dead_ctrl.

Suggested-by: Keith Busch <keith.busch@linux.intel.com>
Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
drivers/nvme/host/pci.c