isci: Implement waiting for suspend in the abort path.

In order to prevent a device from receiving an I/O request while still
in an RNC suspending or resuming state (and therefore failing that
I/O back to libsas with a reset required status) wait for the RNC state
change before proceding in the abort path.

Signed-off-by: Jeff Skirvin <jeffrey.d.skirvin@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>

5 files changed, 66 insertions, 22 deletions

diff --git a/drivers/scsi/isci/remote_device.c b/drivers/scsi/isci/remote_device.c
index 86aca11120f3..acc94a454a1f 100644
--- a/drivers/scsi/isci/remote_device.c
+++ b/drivers/scsi/isci/remote_device.c
@@ -137,6 +137,14 @@ static enum sci_status sci_remote_device_terminate_reqs_checkabort(
         return status;
 }
 
+static bool isci_compare_suspendcount(
+        struct isci_remote_device *idev,
+        u32 localcount)
+{
+        smp_rmb();
+        return localcount != idev->rnc.suspend_count;
+}
+
 enum sci_status isci_remote_device_terminate_requests(
         struct isci_host *ihost,
         struct isci_remote_device *idev,
@@ -144,31 +152,44 @@ enum sci_status isci_remote_device_terminate_requests(
 {
         enum sci_status status = SCI_SUCCESS;
         unsigned long flags;
+        u32 rnc_suspend_count;
 
         spin_lock_irqsave(&ihost->scic_lock, flags);
+
         if (isci_get_device(idev) == NULL) {
                 dev_dbg(&ihost->pdev->dev, "%s: failed isci_get_device(idev=%p)\n",
                         __func__, idev);
                 spin_unlock_irqrestore(&ihost->scic_lock, flags);
                 status = SCI_FAILURE;
         } else {
+                /* If already suspended, don't wait for another suspension. */
+                smp_rmb();
+                rnc_suspend_count
+                        = sci_remote_node_context_is_suspended(&idev->rnc)
+                                ? 0 : idev->rnc.suspend_count;
+
                 dev_dbg(&ihost->pdev->dev,
                         "%s: idev=%p, ireq=%p; started_request_count=%d, "
+                                "rnc_suspend_count=%d, rnc.suspend_count=%d"
                                 "about to wait\n",
-                        __func__, idev, ireq, idev->started_request_count);
+                        __func__, idev, ireq, idev->started_request_count,
+                        rnc_suspend_count, idev->rnc.suspend_count);
                 if (ireq) {
                         /* Terminate a specific TC. */
                         sci_remote_device_terminate_req(ihost, idev, 0, ireq);
                         spin_unlock_irqrestore(&ihost->scic_lock, flags);
-                        wait_event(ihost->eventq, !test_bit(IREQ_ACTIVE,
-                                                            &ireq->flags));
-
+                        wait_event(ihost->eventq,
+                                   (isci_compare_suspendcount(idev,
+                                                              rnc_suspend_count)
+                                    && !test_bit(IREQ_ACTIVE, &ireq->flags)));
                 } else {
                         /* Terminate all TCs. */
                         sci_remote_device_terminate_requests(idev);
                         spin_unlock_irqrestore(&ihost->scic_lock, flags);
                         wait_event(ihost->eventq,
-                                   idev->started_request_count == 0);
+                                   (isci_compare_suspendcount(idev,
+                                                              rnc_suspend_count)
+                                    && idev->started_request_count == 0));
                 }
                 dev_dbg(&ihost->pdev->dev, "%s: idev=%p, wait done\n",
                         __func__, idev);
@@ -1234,19 +1255,20 @@ enum sci_status sci_remote_device_resume(
         return status;
 }
 
-enum sci_status isci_remote_device_resume(
+enum sci_status isci_remote_device_resume_from_abort(
         struct isci_host *ihost,
-        struct isci_remote_device *idev,
-        scics_sds_remote_node_context_callback cb_fn,
-        void *cb_p)
+        struct isci_remote_device *idev)
 {
         unsigned long flags;
         enum sci_status status;
 
         spin_lock_irqsave(&ihost->scic_lock, flags);
-        status = sci_remote_device_resume(idev, cb_fn, cb_p);
+        /* Preserve any current resume callbacks, for instance from other
+         * resumptions.
+         */
+        status = sci_remote_device_resume(idev, idev->rnc.user_callback,
+                                          idev->rnc.user_cookie);
         spin_unlock_irqrestore(&ihost->scic_lock, flags);
-
         return status;
 }
 /**
diff --git a/drivers/scsi/isci/remote_device.h b/drivers/scsi/isci/remote_device.h
index ef563e5360a3..d1d18925fbfc 100644
--- a/drivers/scsi/isci/remote_device.h
+++ b/drivers/scsi/isci/remote_device.h
@@ -357,11 +357,9 @@ enum sci_status sci_remote_device_resume(
         scics_sds_remote_node_context_callback cb_fn,
         void *cb_p);
 
-enum sci_status isci_remote_device_resume(
+enum sci_status isci_remote_device_resume_from_abort(
         struct isci_host *ihost,
-        struct isci_remote_device *idev,
-        scics_sds_remote_node_context_callback cb_fn,
-        void *cb_p);
+        struct isci_remote_device *idev);
 
 enum sci_status isci_remote_device_reset(
         struct isci_host *ihost,
diff --git a/drivers/scsi/isci/remote_node_context.c b/drivers/scsi/isci/remote_node_context.c
index 48565de50016..77c8b5138b7e 100644
--- a/drivers/scsi/isci/remote_node_context.c
+++ b/drivers/scsi/isci/remote_node_context.c
@@ -90,6 +90,15 @@ bool sci_remote_node_context_is_ready(
         return false;
 }
 
+bool sci_remote_node_context_is_suspended(struct sci_remote_node_context *sci_rnc)
+{
+        u32 current_state = sci_rnc->sm.current_state_id;
+
+        if (current_state == SCI_RNC_TX_RX_SUSPENDED)
+                return true;
+        return false;
+}
+
 static union scu_remote_node_context *sci_rnc_by_id(struct isci_host *ihost, u16 id)
 {
         if (id < ihost->remote_node_entries &&
@@ -339,6 +348,13 @@ static void sci_remote_node_context_tx_rx_suspended_state_enter(struct sci_base_
         struct sci_remote_node_context *rnc = container_of(sm, typeof(*rnc), sm);
         struct isci_remote_device *idev = rnc_to_dev(rnc);
         struct isci_host *ihost = idev->owning_port->owning_controller;
+        u32 new_count = rnc->suspend_count + 1;
+
+        if (new_count == 0)
+                rnc->suspend_count = 1;
+        else
+                rnc->suspend_count = new_count;
+        smp_wmb();
 
         /* Terminate outstanding requests pending abort. */
         sci_remote_device_abort_requests_pending_abort(idev);
@@ -634,6 +650,11 @@ enum sci_status sci_remote_node_context_resume(struct sci_remote_node_context *s
         enum scis_sds_remote_node_context_states state;
 
         state = sci_rnc->sm.current_state_id;
+        dev_dbg(scirdev_to_dev(rnc_to_dev(sci_rnc)),
+                 "%s: state %s, cb_fn = %p, cb_p = %p; dest_state = %d\n",
+                 __func__, rnc_state_name(state), cb_fn, cb_p,
+                 sci_rnc->destination_state);
+
         switch (state) {
         case SCI_RNC_INITIAL:
                 if (sci_rnc->remote_node_index == SCIC_SDS_REMOTE_NODE_CONTEXT_INVALID_INDEX)
diff --git a/drivers/scsi/isci/remote_node_context.h b/drivers/scsi/isci/remote_node_context.h
index 9eee304fdf9a..c61c02e095ad 100644
--- a/drivers/scsi/isci/remote_node_context.h
+++ b/drivers/scsi/isci/remote_node_context.h
@@ -170,6 +170,7 @@ struct sci_remote_node_context {
          */
         u32 suspend_type;
         enum sci_remote_node_suspension_reasons suspend_reason;
+        u32 suspend_count;
 
         /**
          * This field is true if the remote node context is resuming from its current
@@ -203,6 +204,8 @@ void sci_remote_node_context_construct(struct sci_remote_node_context *rnc,
 bool sci_remote_node_context_is_ready(
         struct sci_remote_node_context *sci_rnc);
 
+bool sci_remote_node_context_is_suspended(struct sci_remote_node_context *sci_rnc);
+
 enum sci_status sci_remote_node_context_event_handler(struct sci_remote_node_context *sci_rnc,
                                                            u32 event_code);
 enum sci_status sci_remote_node_context_destruct(struct sci_remote_node_context *sci_rnc,
diff --git a/drivers/scsi/isci/task.c b/drivers/scsi/isci/task.c
index 5d738fd5f882..c1c6dd0473ae 100644
--- a/drivers/scsi/isci/task.c
+++ b/drivers/scsi/isci/task.c
@@ -317,11 +317,11 @@ static int isci_task_execute_tmf(struct isci_host *ihost,
                 spin_unlock_irqrestore(&ihost->scic_lock, flags);
                 goto err_tci;
         }
-        /* The RNC must be unsuspended before the TMF can get a response. */
-        sci_remote_device_resume(idev, NULL, NULL);
-
         spin_unlock_irqrestore(&ihost->scic_lock, flags);
 
+        /* The RNC must be unsuspended before the TMF can get a response. */
+        isci_remote_device_resume_from_abort(ihost, idev);
+
         /* Wait for the TMF to complete, or a timeout. */
         timeleft = wait_for_completion_timeout(&completion,
                                                msecs_to_jiffies(timeout_ms));
@@ -554,11 +554,11 @@ int isci_task_abort_task(struct sas_task *task)
             sas_protocol_ata(task->task_proto) ||
             test_bit(IREQ_COMPLETE_IN_TARGET, &old_request->flags)) {
 
-                /* No task to send, so explicitly resume the device here */
-                sci_remote_device_resume(idev, NULL, NULL);
-
                 spin_unlock_irqrestore(&ihost->scic_lock, flags);
 
+                /* No task to send, so explicitly resume the device here */
+                isci_remote_device_resume_from_abort(ihost, idev);
+
                 dev_warn(&ihost->pdev->dev,
                          "%s: %s request"
                                  " or complete_in_target (%d), thus no TMF\n",
@@ -757,7 +757,7 @@ static int isci_reset_device(struct isci_host *ihost,
                 reset_stat = sas_phy_reset(phy, !dev_is_sata(dev));
 
         /* Explicitly resume the RNC here, since there was no task sent. */
-        isci_remote_device_resume(ihost, idev, NULL, NULL);
+        isci_remote_device_resume_from_abort(ihost, idev);
 
         dev_dbg(&ihost->pdev->dev, "%s: idev %p complete, reset_stat=%d.\n",
                 __func__, idev, reset_stat);