nvme-fc: change ctlr state assignments during reset/reconnect
authorJames Smart <jsmart2021@gmail.com>
Wed, 25 Oct 2017 23:43:14 +0000 (16:43 -0700)
committerChristoph Hellwig <hch@lst.de>
Wed, 1 Nov 2017 15:28:23 +0000 (16:28 +0100)
Clean up some of the controller state checks and add the
RESETTING->RECONNECTING state transition.

Specifically:
- the movement of the RESETTING state change and schedule of reset_work
  to core doesn't work wiht nvme_fc_error_recovery setting state to
  RECONNECTING before attempting to reset.  Remove the state change as
  the reset request does it.
- In the rare cases where an error occurs right as we're transitioning
  to LIVE, defer the controller start actions.
- In error handling on teardown of associations while performing initial
  controller creation - avoid quiesce calls on the admin_q.  They are
  unneeded.
- Add the RESETTING->RECONNECTING transition in the reset handler.

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
drivers/nvme/host/fc.c

index 6a025a8d8c32c51958f54ab0c8a503ce0ebe6d48..e37c69f7921d37d6b71ff8585f7a50a9d10a2694 100644 (file)
@@ -1884,13 +1884,6 @@ nvme_fc_error_recovery(struct nvme_fc_ctrl *ctrl, char *errmsg)
        dev_warn(ctrl->ctrl.device,
                "NVME-FC{%d}: resetting controller\n", ctrl->cnum);
 
-       if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RECONNECTING)) {
-               dev_err(ctrl->ctrl.device,
-                       "NVME-FC{%d}: error_recovery: Couldn't change state "
-                       "to RECONNECTING\n", ctrl->cnum);
-               return;
-       }
-
        nvme_reset_ctrl(&ctrl->ctrl);
 }
 
@@ -2528,11 +2521,11 @@ nvme_fc_create_association(struct nvme_fc_ctrl *ctrl)
        }
 
        changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE);
-       WARN_ON_ONCE(!changed);
 
        ctrl->ctrl.nr_reconnects = 0;
 
-       nvme_start_ctrl(&ctrl->ctrl);
+       if (changed)
+               nvme_start_ctrl(&ctrl->ctrl);
 
        return 0;       /* Success */
 
@@ -2600,7 +2593,8 @@ nvme_fc_delete_association(struct nvme_fc_ctrl *ctrl)
         * use blk_mq_tagset_busy_itr() and the transport routine to
         * terminate the exchanges.
         */
-       blk_mq_quiesce_queue(ctrl->ctrl.admin_q);
+       if (ctrl->ctrl.state != NVME_CTRL_NEW)
+               blk_mq_quiesce_queue(ctrl->ctrl.admin_q);
        blk_mq_tagset_busy_iter(&ctrl->admin_tag_set,
                                nvme_fc_terminate_exchange, &ctrl->ctrl);
 
@@ -2649,12 +2643,8 @@ nvme_fc_delete_ctrl(struct nvme_ctrl *nctrl)
 static void
 nvme_fc_reconnect_or_delete(struct nvme_fc_ctrl *ctrl, int status)
 {
-       /* If we are resetting/deleting then do nothing */
-       if (ctrl->ctrl.state != NVME_CTRL_RECONNECTING) {
-               WARN_ON_ONCE(ctrl->ctrl.state == NVME_CTRL_NEW ||
-                       ctrl->ctrl.state == NVME_CTRL_LIVE);
+       if (ctrl->ctrl.state != NVME_CTRL_RECONNECTING)
                return;
-       }
 
        dev_info(ctrl->ctrl.device,
                "NVME-FC{%d}: reset: Reconnect attempt failed (%d)\n",
@@ -2683,9 +2673,17 @@ nvme_fc_reset_ctrl_work(struct work_struct *work)
        int ret;
 
        nvme_stop_ctrl(&ctrl->ctrl);
+
        /* will block will waiting for io to terminate */
        nvme_fc_delete_association(ctrl);
 
+       if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RECONNECTING)) {
+               dev_err(ctrl->ctrl.device,
+                       "NVME-FC{%d}: error_recovery: Couldn't change state "
+                       "to RECONNECTING\n", ctrl->cnum);
+               return;
+       }
+
        ret = nvme_fc_create_association(ctrl);
        if (ret)
                nvme_fc_reconnect_or_delete(ctrl, ret);