ibmvscsis: Enable Logical Partition Migration Support

Changes to support a new mechanism from phyp to better synchronize the
logical partition migration (LPM) of the client partition.
This includes a new VIOCTL to register that we support this new
functionality, and 2 new Transport Event types, and finally another
new VIOCTL to let phyp know once we're ready for the Suspend.

Signed-off-by: Michael Cyr <mikecyr@us.ibm.com>
Signed-off-by: Bryant G. Ly <bryantly@linux.vnet.ibm.com>
Signed-off-by: Nicholas Bellinger <nab@linux-iscsi.org>

drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.c

@@ -155,6 +155,9 @@ static long ibmvscsis_unregister_command_q(struct scsi_info *vscsi)
 		qrc = h_free_crq(vscsi->dds.unit_id);
 		switch (qrc) {
 		case H_SUCCESS:
+			spin_lock_bh(&vscsi->intr_lock);
+			vscsi->flags &= ~PREP_FOR_SUSPEND_FLAGS;
+			spin_unlock_bh(&vscsi->intr_lock);
 			break;
 
 		case H_HARDWARE:
@@ -422,6 +425,9 @@ static void ibmvscsis_disconnect(struct work_struct *work)
 	new_state = vscsi->new_state;
 	vscsi->new_state = 0;
 
+	vscsi->flags |= DISCONNECT_SCHEDULED;
+	vscsi->flags &= ~SCHEDULE_DISCONNECT;
+
 	pr_debug("disconnect: flags 0x%x, state 0x%hx\n", vscsi->flags,
 		 vscsi->state);
 
@@ -802,6 +808,13 @@ static long ibmvscsis_establish_new_q(struct scsi_info *vscsi)
 	long rc = ADAPT_SUCCESS;
 	uint format;
 
+	rc = h_vioctl(vscsi->dds.unit_id, H_ENABLE_PREPARE_FOR_SUSPEND, 30000,
+		      0, 0, 0, 0);
+	if (rc == H_SUCCESS)
+		vscsi->flags |= PREP_FOR_SUSPEND_ENABLED;
+	else if (rc != H_NOT_FOUND)
+		pr_err("Error from Enable Prepare for Suspend: %ld\n", rc);
+
 	vscsi->flags &= PRESERVE_FLAG_FIELDS;
 	vscsi->rsp_q_timer.timer_pops = 0;
 	vscsi->debit = 0;
@@ -950,6 +963,63 @@ static void ibmvscsis_free_cmd_resources(struct scsi_info *vscsi,
 	}
 }
 
+/**
+ * ibmvscsis_ready_for_suspend() - Helper function to call VIOCTL
+ * @vscsi:	Pointer to our adapter structure
+ * @idle:	Indicates whether we were called from adapter_idle.  This
+ *		is important to know if we need to do a disconnect, since if
+ *		we're called from adapter_idle, we're still processing the
+ *		current disconnect, so we can't just call post_disconnect.
+ *
+ * This function is called when the adapter is idle when phyp has sent
+ * us a Prepare for Suspend Transport Event.
+ *
+ * EXECUTION ENVIRONMENT:
+ *	Process or interrupt environment called with interrupt lock held
+ */
+static long ibmvscsis_ready_for_suspend(struct scsi_info *vscsi, bool idle)
+{
+	long rc = 0;
+	struct viosrp_crq *crq;
+
+	/* See if there is a Resume event in the queue */
+	crq = vscsi->cmd_q.base_addr + vscsi->cmd_q.index;
+
+	pr_debug("ready_suspend: flags 0x%x, state 0x%hx crq_valid:%x\n",
+		 vscsi->flags, vscsi->state, (int)crq->valid);
+
+	if (!(vscsi->flags & PREP_FOR_SUSPEND_ABORTED) && !(crq->valid)) {
+		rc = h_vioctl(vscsi->dds.unit_id, H_READY_FOR_SUSPEND, 0, 0, 0,
+			      0, 0);
+		if (rc) {
+			pr_err("Ready for Suspend Vioctl failed: %ld\n", rc);
+			rc = 0;
+		}
+	} else if (((vscsi->flags & PREP_FOR_SUSPEND_OVERWRITE) &&
+		    (vscsi->flags & PREP_FOR_SUSPEND_ABORTED)) ||
+		   ((crq->valid) && ((crq->valid != VALID_TRANS_EVENT) ||
+				     (crq->format != RESUME_FROM_SUSP)))) {
+		if (idle) {
+			vscsi->state = ERR_DISCONNECT_RECONNECT;
+			ibmvscsis_reset_queue(vscsi);
+			rc = -1;
+		} else if (vscsi->state == CONNECTED) {
+			ibmvscsis_post_disconnect(vscsi,
+						  ERR_DISCONNECT_RECONNECT, 0);
+		}
+
+		vscsi->flags &= ~PREP_FOR_SUSPEND_OVERWRITE;
+
+		if ((crq->valid) && ((crq->valid != VALID_TRANS_EVENT) ||
+				     (crq->format != RESUME_FROM_SUSP)))
+			pr_err("Invalid element in CRQ after Prepare for Suspend");
+	}
+
+	vscsi->flags &= ~(PREP_FOR_SUSPEND_PENDING | PREP_FOR_SUSPEND_ABORTED);
+
+	return rc;
+}
+
 /**
  * ibmvscsis_trans_event() - Handle a Transport Event
  * @vscsi:	Pointer to our adapter structure
@@ -974,18 +1044,8 @@ static long ibmvscsis_trans_event(struct scsi_info *vscsi,
 	case PARTNER_FAILED:
 	case PARTNER_DEREGISTER:
 		ibmvscsis_delete_client_info(vscsi, true);
-		break;
+		if (crq->format == MIGRATED)
-
+			vscsi->flags &= ~PREP_FOR_SUSPEND_OVERWRITE;
-	default:
-		rc = ERROR;
-		dev_err(&vscsi->dev, "trans_event: invalid format %d\n",
-			(uint)crq->format);
-		ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT,
-					  RESPONSE_Q_DOWN);
-		break;
-	}
-
-	if (rc == ADAPT_SUCCESS) {
 		switch (vscsi->state) {
 		case NO_QUEUE:
 		case ERR_DISCONNECTED:
@@ -1034,6 +1094,60 @@ static long ibmvscsis_trans_event(struct scsi_info *vscsi,
 			vscsi->flags |= (RESPONSE_Q_DOWN | TRANS_EVENT);
 			break;
 		}
+		break;
+
+	case PREPARE_FOR_SUSPEND:
+		pr_debug("Prep for Suspend, crq status = 0x%x\n",
+			 (int)crq->status);
+		switch (vscsi->state) {
+		case ERR_DISCONNECTED:
+		case WAIT_CONNECTION:
+		case CONNECTED:
+			ibmvscsis_ready_for_suspend(vscsi, false);
+			break;
+		case SRP_PROCESSING:
+			vscsi->resume_state = vscsi->state;
+			vscsi->flags |= PREP_FOR_SUSPEND_PENDING;
+			if (crq->status == CRQ_ENTRY_OVERWRITTEN)
+				vscsi->flags |= PREP_FOR_SUSPEND_OVERWRITE;
+			ibmvscsis_post_disconnect(vscsi, WAIT_IDLE, 0);
+			break;
+		case NO_QUEUE:
+		case UNDEFINED:
+		case UNCONFIGURING:
+		case WAIT_ENABLED:
+		case ERR_DISCONNECT:
+		case ERR_DISCONNECT_RECONNECT:
+		case WAIT_IDLE:
+			pr_err("Invalid state for Prepare for Suspend Trans Event: 0x%x\n",
+			       vscsi->state);
+			break;
+		}
+		break;
+
+	case RESUME_FROM_SUSP:
+		pr_debug("Resume from Suspend, crq status = 0x%x\n",
+			 (int)crq->status);
+		if (vscsi->flags & PREP_FOR_SUSPEND_PENDING) {
+			vscsi->flags |= PREP_FOR_SUSPEND_ABORTED;
+		} else {
+			if ((crq->status == CRQ_ENTRY_OVERWRITTEN) ||
+			    (vscsi->flags & PREP_FOR_SUSPEND_OVERWRITE)) {
+				ibmvscsis_post_disconnect(vscsi,
+							  ERR_DISCONNECT_RECONNECT,
+							  0);
+				vscsi->flags &= ~PREP_FOR_SUSPEND_OVERWRITE;
+			}
+		}
+		break;
+
+	default:
+		rc = ERROR;
+		dev_err(&vscsi->dev, "trans_event: invalid format %d\n",
+			(uint)crq->format);
+		ibmvscsis_post_disconnect(vscsi, ERR_DISCONNECT,
+					  RESPONSE_Q_DOWN);
+		break;
 	}
 
 	rc = vscsi->flags & SCHEDULE_DISCONNECT;
@@ -1201,6 +1315,7 @@ static struct ibmvscsis_cmd *ibmvscsis_get_free_cmd(struct scsi_info *vscsi)
 static void ibmvscsis_adapter_idle(struct scsi_info *vscsi)
 {
 	int free_qs = false;
+	long rc = 0;
 
 	pr_debug("adapter_idle: flags 0x%x, state 0x%hx\n", vscsi->flags,
 		 vscsi->state);
@@ -1240,7 +1355,14 @@ static void ibmvscsis_adapter_idle(struct scsi_info *vscsi)
 		vscsi->rsp_q_timer.timer_pops = 0;
 		vscsi->debit = 0;
 		vscsi->credit = 0;
-		if (vscsi->flags & TRANS_EVENT) {
+		if (vscsi->flags & PREP_FOR_SUSPEND_PENDING) {
+			vscsi->state = vscsi->resume_state;
+			vscsi->resume_state = 0;
+			rc = ibmvscsis_ready_for_suspend(vscsi, true);
+			vscsi->flags &= ~DISCONNECT_SCHEDULED;
+			if (rc)
+				break;
+		} else if (vscsi->flags & TRANS_EVENT) {
 			vscsi->state = WAIT_CONNECTION;
 			vscsi->flags &= PRESERVE_FLAG_FIELDS;
 		} else {

drivers/scsi/ibmvscsi_tgt/ibmvscsi_tgt.h

@@ -262,6 +262,14 @@ struct scsi_info {
 #define DISCONNECT_SCHEDULED          0x00800
 	/* remove function is sleeping */
 #define CFG_SLEEPING                  0x01000
+	/* Register for Prepare for Suspend Transport Events */
+#define PREP_FOR_SUSPEND_ENABLED      0x02000
+	/* Prepare for Suspend event sent */
+#define PREP_FOR_SUSPEND_PENDING      0x04000
+	/* Resume from Suspend event sent */
+#define PREP_FOR_SUSPEND_ABORTED      0x08000
+	/* Prepare for Suspend event overwrote another CRQ entry */
+#define PREP_FOR_SUSPEND_OVERWRITE    0x10000
 	u32 flags;
 	/* adapter lock */
 	spinlock_t intr_lock;
@@ -272,6 +280,7 @@ struct scsi_info {
 	/* used in crq, to tag what iu the response is for */
 	u64  empty_iu_tag;
 	uint new_state;
+	uint resume_state;
 	/* control block for the response queue timer */
 	struct timer_cb rsp_q_timer;
 	/* keep last client to enable proper accounting */
@@ -324,8 +333,13 @@ struct scsi_info {
 #define TARGET_STOP(VSCSI) (long)(((VSCSI)->state & DONT_PROCESS_STATE) | \
 				  ((VSCSI)->flags & BLOCK))
 
+#define PREP_FOR_SUSPEND_FLAGS  (PREP_FOR_SUSPEND_ENABLED | \
+				 PREP_FOR_SUSPEND_PENDING | \
+				 PREP_FOR_SUSPEND_ABORTED | \
+				 PREP_FOR_SUSPEND_OVERWRITE)
+
 /* flag bit that are not reset during disconnect */
-#define PRESERVE_FLAG_FIELDS 0
+#define PRESERVE_FLAG_FIELDS (PREP_FOR_SUSPEND_FLAGS)
 
 #define vio_iu(IUE) ((union viosrp_iu *)((IUE)->sbuf->buf))
 
@@ -333,8 +347,15 @@ struct scsi_info {
 #define WRITE_CMD(cdb)	(((cdb)[0] & 0x1F) == 0xA)
 
 #ifndef H_GET_PARTNER_INFO
-#define H_GET_PARTNER_INFO      0x0000000000000008LL
+#define H_GET_PARTNER_INFO              0x0000000000000008LL
 #endif
+#ifndef H_ENABLE_PREPARE_FOR_SUSPEND
+#define H_ENABLE_PREPARE_FOR_SUSPEND    0x000000000000001DLL
+#endif
+#ifndef H_READY_FOR_SUSPEND
+#define H_READY_FOR_SUSPEND             0x000000000000001ELL
+#endif
+
 
 #define h_copy_rdma(l, sa, sb, da, db) \
 		plpar_hcall_norets(H_COPY_RDMA, l, sa, sb, da, db)

drivers/scsi/ibmvscsi_tgt/libsrp.h

@@ -30,10 +30,13 @@ enum srp_trans_event {
 	UNUSED_FORMAT = 0,
 	PARTNER_FAILED = 1,
 	PARTNER_DEREGISTER = 2,
-	MIGRATED = 6
+	MIGRATED = 6,
+	PREPARE_FOR_SUSPEND = 9,
+	RESUME_FROM_SUSP = 0xA
 };
 
 enum srp_status {
+	CRQ_ENTRY_OVERWRITTEN = 0x20,
 	HEADER_DESCRIPTOR = 0xF1,
 	PING = 0xF5,
 	PING_RESPONSE = 0xF6