[bug report] nvme-tcp: add NVMe over TCP host driver

[Dan Carpenter]

Hello Sagi Grimberg,

The patch 3f2304f8c6d6: "nvme-tcp: add NVMe over TCP host driver"
from Dec 3, 2018, leads to the following
Smatch static checker warning:

	drivers/nvme/host/multipath.c:101 nvme_kick_requeue_lists()
	warn: sleeping in atomic context

drivers/nvme/host/multipath.c
    97 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
    98 {
    99		struct nvme_ns *ns;
    100 
--> 101 	down_read(&ctrl->namespaces_rwsem);
    102 	list_for_each_entry(ns, &ctrl->namespaces, list) {
    103 		if (ns->head->disk)
    104 			kblockd_schedule_work(&ns->head->requeue_work);
    105 	}
    106 	up_read(&ctrl->namespaces_rwsem);
    107 }

This is a new Smatch warning I'm working on and it's sort of
overwhelming because it has generates too many warnings and they're
complicated to analyze and report.  I'm trying to send these
automatically generated call trees to see if it will help.

nvme_fc_unregister_remoteport() <- disables preempt
-> nvme_fc_ctrl_connectivity_loss()
   -> nvme_reset_ctrl()
nvme_fc_ctrl_connectivity_loss() <duplicate>
nvme_fc_unregister_remoteport() <- disables preempt <duplicate>
nvme_fc_exit_module() <- disables preempt
-> nvme_fc_cleanup_for_unload()
   -> nvme_fc_delete_controllers() <- disables preempt
      -> nvme_delete_ctrl()
nvme_tcp_state_change() <- disables preempt
-> nvme_tcp_error_recovery()
         -> nvme_change_ctrl_state()
            -> nvme_kick_requeue_lists()


I looked at the last call tree and it seems like potentially a real bug.

nvme_tcp_state_change() <- disables preempt
-> nvme_tcp_error_recovery()
         -> nvme_change_ctrl_state()
            -> nvme_kick_requeue_lists()

regards,
dan carpenter