scsi: qla2xxx: Add function call to qpair for door bell

Add call back to door bell for qpair. This help reduce access to
qla_hw_data structure, in order to reduce cach thrash.

Signed-off-by: Quinn Tran <quinn.tran@cavium.com>
Signed-off-by: Himanshu Madhani <himanshu.madhani@cavium.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

drivers/scsi/qla2xxx/qla_def.h

@@ -3272,6 +3272,9 @@ struct qla_qpair {
 	uint16_t vp_idx;		/* vport ID */
 	mempool_t *srb_mempool;
 
+	struct pci_dev  *pdev;
+	void (*reqq_start_iocbs)(struct qla_qpair *);
+
 	/* to do: New driver: move queues to here instead of pointers */
 	struct req_que *req;
 	struct rsp_que *rsp;

drivers/scsi/qla2xxx/qla_init.c

@@ -7653,6 +7653,9 @@ struct qla_qpair *qla2xxx_create_qpair(struct scsi_qla_host *vha, int qos,
 
 		qpair->msix->in_use = 1;
 		list_add_tail(&qpair->qp_list_elem, &vha->qp_list);
+		qpair->pdev = ha->pdev;
+		if (IS_QLA27XX(ha) || IS_QLA83XX(ha))
+			qpair->reqq_start_iocbs = qla_83xx_start_iocbs;
 
 		mutex_unlock(&ha->mq_lock);
 

drivers/scsi/qla2xxx/qla_inline.h

@@ -352,3 +352,18 @@ qla_qpair_to_hint(struct qla_tgt *tgt, struct qla_qpair *qpair)
 
 	return NULL;
 }
+
+static inline void
+qla_83xx_start_iocbs(struct qla_qpair *qpair)
+{
+	struct req_que *req = qpair->req;
+
+	req->ring_index++;
+	if (req->ring_index == req->length) {
+		req->ring_index = 0;
+		req->ring_ptr = req->ring;
+	} else
+		req->ring_ptr++;
+
+	WRT_REG_DWORD(req->req_q_in, req->ring_index);
+}

drivers/scsi/qla2xxx/qla_os.c

@@ -351,6 +351,28 @@ int qla2xxx_mqueuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd,
 	struct qla_qpair *qpair);
 
 /* -------------------------------------------------------------------------- */
+static void qla_init_base_qpair(struct scsi_qla_host *vha, struct req_que *req,
+    struct rsp_que *rsp)
+{
+	struct qla_hw_data *ha = vha->hw;
+	rsp->qpair = ha->base_qpair;
+	rsp->req = req;
+	ha->base_qpair->req = req;
+	ha->base_qpair->rsp = rsp;
+	ha->base_qpair->vha = vha;
+	ha->base_qpair->qp_lock_ptr = &ha->hardware_lock;
+	ha->base_qpair->use_shadow_reg = IS_SHADOW_REG_CAPABLE(ha) ? 1 : 0;
+	ha->base_qpair->msix = &ha->msix_entries[QLA_MSIX_RSP_Q];
+	INIT_LIST_HEAD(&ha->base_qpair->hints_list);
+	ha->base_qpair->enable_class_2 = ql2xenableclass2;
+	/* init qpair to this cpu. Will adjust at run time. */
+	qla_cpu_update(rsp->qpair, smp_processor_id());
+	ha->base_qpair->pdev = ha->pdev;
+
+	if (IS_QLA27XX(ha) || IS_QLA83XX(ha))
+		ha->base_qpair->reqq_start_iocbs = qla_83xx_start_iocbs;
+}
+
 static int qla2x00_alloc_queues(struct qla_hw_data *ha, struct req_que *req,
 				struct rsp_que *rsp)
 {
@@ -378,18 +400,7 @@ static int qla2x00_alloc_queues(struct qla_hw_data *ha, struct req_que *req,
 		goto fail_base_qpair;
 	}
 
-	rsp->qpair = ha->base_qpair;
-	rsp->req = req;
-	ha->base_qpair->req = req;
-	ha->base_qpair->rsp = rsp;
-	ha->base_qpair->vha = vha;
-	ha->base_qpair->qp_lock_ptr = &ha->hardware_lock;
-	ha->base_qpair->use_shadow_reg = IS_SHADOW_REG_CAPABLE(ha) ? 1 : 0;
-	/* init qpair to this cpu. Will adjust at run time. */
-	ha->base_qpair->msix = &ha->msix_entries[QLA_MSIX_RSP_Q];
-	INIT_LIST_HEAD(&ha->base_qpair->hints_list);
-	ha->base_qpair->enable_class_2 = ql2xenableclass2;
-	qla_cpu_update(rsp->qpair, smp_processor_id());
+	qla_init_base_qpair(vha, req, rsp);
 
 	if (ql2xmqsupport && ha->max_qpairs) {
 		ha->queue_pair_map = kcalloc(ha->max_qpairs, sizeof(struct qla_qpair *),

drivers/scsi/qla2xxx/qla_target.c

@@ -1728,7 +1728,10 @@ static void qlt_24xx_send_abts_resp(struct qla_qpair *qpair,
 
 	/* Memory Barrier */
 	wmb();
-	qla2x00_start_iocbs(vha, qpair->req);
+	if (qpair->reqq_start_iocbs)
+		qpair->reqq_start_iocbs(qpair);
+	else
+		qla2x00_start_iocbs(vha, qpair->req);
 }
 
 /*
@@ -2058,7 +2061,10 @@ static void qlt_24xx_send_task_mgmt_ctio(struct qla_qpair *qpair,
 
 	/* Memory Barrier */
 	wmb();
-	qla2x00_start_iocbs(ha, qpair->req);
+	if (qpair->reqq_start_iocbs)
+		qpair->reqq_start_iocbs(qpair);
+	else
+		qla2x00_start_iocbs(ha, qpair->req);
 }
 
 void qlt_free_mcmd(struct qla_tgt_mgmt_cmd *mcmd)
@@ -2071,12 +2077,13 @@ EXPORT_SYMBOL(qlt_free_mcmd);
  * ha->hardware_lock supposed to be held on entry. Might drop it, then
  * reacquire
  */
-void qlt_send_resp_ctio(scsi_qla_host_t *vha, struct qla_tgt_cmd *cmd,
+void qlt_send_resp_ctio(struct qla_qpair *qpair, struct qla_tgt_cmd *cmd,
     uint8_t scsi_status, uint8_t sense_key, uint8_t asc, uint8_t ascq)
 {
 	struct atio_from_isp *atio = &cmd->atio;
 	struct ctio7_to_24xx *ctio;
 	uint16_t temp;
+	struct scsi_qla_host *vha = cmd->vha;
 
 	ql_dbg(ql_dbg_tgt_dif, vha, 0x3066,
 	    "Sending response CTIO7 (vha=%p, atio=%p, scsi_status=%02x, "
@@ -2127,7 +2134,11 @@ void qlt_send_resp_ctio(scsi_qla_host_t *vha, struct qla_tgt_cmd *cmd,
 	/* Memory Barrier */
 	wmb();
 
-	qla2x00_start_iocbs(vha, vha->req);
+	if (qpair->reqq_start_iocbs)
+		qpair->reqq_start_iocbs(qpair);
+	else
+		qla2x00_start_iocbs(vha, qpair->req);
+
 out:
 	return;
 }
@@ -2205,7 +2216,7 @@ static int qlt_pci_map_calc_cnt(struct qla_tgt_prm *prm)
 	BUG_ON(cmd->sg_cnt == 0);
 
 	prm->sg = (struct scatterlist *)cmd->sg;
-	prm->seg_cnt = pci_map_sg(prm->tgt->ha->pdev, cmd->sg,
+	prm->seg_cnt = pci_map_sg(cmd->qpair->pdev, cmd->sg,
 	    cmd->sg_cnt, cmd->dma_data_direction);
 	if (unlikely(prm->seg_cnt == 0))
 		goto out_err;
@@ -2232,7 +2243,7 @@ static int qlt_pci_map_calc_cnt(struct qla_tgt_prm *prm)
 
 		if (cmd->prot_sg_cnt) {
 			prm->prot_sg      = cmd->prot_sg;
-			prm->prot_seg_cnt = pci_map_sg(prm->tgt->ha->pdev,
+			prm->prot_seg_cnt = pci_map_sg(cmd->qpair->pdev,
 				cmd->prot_sg, cmd->prot_sg_cnt,
 				cmd->dma_data_direction);
 			if (unlikely(prm->prot_seg_cnt == 0))
@@ -2260,21 +2271,24 @@ out_err:
 
 static void qlt_unmap_sg(struct scsi_qla_host *vha, struct qla_tgt_cmd *cmd)
 {
-	struct qla_hw_data *ha = vha->hw;
-
+	struct qla_hw_data *ha;
+	struct qla_qpair *qpair;
 	if (!cmd->sg_mapped)
 		return;
 
-	pci_unmap_sg(ha->pdev, cmd->sg, cmd->sg_cnt, cmd->dma_data_direction);
+	qpair = cmd->qpair;
+
+	pci_unmap_sg(qpair->pdev, cmd->sg, cmd->sg_cnt,
+	    cmd->dma_data_direction);
 	cmd->sg_mapped = 0;
 
 	if (cmd->prot_sg_cnt)
-		pci_unmap_sg(ha->pdev, cmd->prot_sg, cmd->prot_sg_cnt,
+		pci_unmap_sg(qpair->pdev, cmd->prot_sg, cmd->prot_sg_cnt,
 			cmd->dma_data_direction);
 
 	if (!cmd->ctx)
 		return;
-
+	ha = vha->hw;
 	if (cmd->ctx_dsd_alloced)
 		qla2x00_clean_dsd_pool(ha, cmd->ctx);
 
@@ -2324,7 +2338,6 @@ static inline void *qlt_get_req_pkt(struct req_que *req)
 /* ha->hardware_lock supposed to be held on entry */
 static inline uint32_t qlt_make_handle(struct qla_qpair *qpair)
 {
-	struct scsi_qla_host *vha = qpair->vha;
 	uint32_t h;
 	int index;
 	uint8_t found = 0;
@@ -2349,9 +2362,9 @@ static inline uint32_t qlt_make_handle(struct qla_qpair *qpair)
 	if (found) {
 		req->current_outstanding_cmd = h;
 	} else {
-		ql_dbg(ql_dbg_io, vha, 0x305b,
-			"qla_target(%d): Ran out of empty cmd slots\n",
-			vha->vp_idx);
+		ql_dbg(ql_dbg_io, qpair->vha, 0x305b,
+		    "qla_target(%d): Ran out of empty cmd slots\n",
+		    qpair->vha->vp_idx);
 		h = QLA_TGT_NULL_HANDLE;
 	}
 
@@ -3189,7 +3202,10 @@ int qlt_xmit_response(struct qla_tgt_cmd *cmd, int xmit_type,
 
 	/* Memory Barrier */
 	wmb();
-	qla2x00_start_iocbs(vha, qpair->req);
+	if (qpair->reqq_start_iocbs)
+		qpair->reqq_start_iocbs(qpair);
+	else
+		qla2x00_start_iocbs(vha, qpair->req);
 	spin_unlock_irqrestore(qpair->qp_lock_ptr, flags);
 
 	return 0;
@@ -3264,7 +3280,10 @@ int qlt_rdy_to_xfer(struct qla_tgt_cmd *cmd)
 
 	/* Memory Barrier */
 	wmb();
-	qla2x00_start_iocbs(vha, qpair->req);
+	if (qpair->reqq_start_iocbs)
+		qpair->reqq_start_iocbs(qpair);
+	else
+		qla2x00_start_iocbs(vha, qpair->req);
 	spin_unlock_irqrestore(qpair->qp_lock_ptr, flags);
 
 	return res;
@@ -3282,7 +3301,7 @@ EXPORT_SYMBOL(qlt_rdy_to_xfer);
  * it is assumed either hardware_lock or qpair lock is held.
  */
 static void
-qlt_handle_dif_error(struct scsi_qla_host *vha, struct qla_tgt_cmd *cmd,
+qlt_handle_dif_error(struct qla_qpair *qpair, struct qla_tgt_cmd *cmd,
 	struct ctio_crc_from_fw *sts)
 {
 	uint8_t		*ap = &sts->actual_dif[0];
@@ -3290,6 +3309,7 @@ qlt_handle_dif_error(struct scsi_qla_host *vha, struct qla_tgt_cmd *cmd,
 	uint64_t	lba = cmd->se_cmd.t_task_lba;
 	uint8_t scsi_status, sense_key, asc, ascq;
 	unsigned long flags;
+	struct scsi_qla_host *vha = cmd->vha;
 
 	cmd->trc_flags |= TRC_DIF_ERR;
 
@@ -3370,7 +3390,8 @@ out:
 		}
 		spin_unlock_irqrestore(&cmd->cmd_lock, flags);
 
-		qlt_send_resp_ctio(vha, cmd, scsi_status, sense_key, asc, ascq);
+		qlt_send_resp_ctio(qpair, cmd, scsi_status, sense_key, asc,
+		    ascq);
 		/* assume scsi status gets out on the wire.
 		 * Will not wait for completion.
 		 */
@@ -3525,7 +3546,10 @@ static int __qlt_send_term_exchange(struct qla_qpair *qpair,
 
 	/* Memory Barrier */
 	wmb();
-	qla2x00_start_iocbs(vha, qpair->req);
+	if (qpair->reqq_start_iocbs)
+		qpair->reqq_start_iocbs(qpair);
+	else
+		qla2x00_start_iocbs(vha, qpair->req);
 	return ret;
 }
 
@@ -3883,7 +3907,7 @@ static void qlt_do_ctio_completion(struct scsi_qla_host *vha,
 			    *((u64 *)&crc->actual_dif[0]),
 			    *((u64 *)&crc->expected_dif[0]));
 
-			qlt_handle_dif_error(vha, cmd, ctio);
+			qlt_handle_dif_error(qpair, cmd, ctio);
 			return;
 		}
 		default:
@@ -5121,7 +5145,10 @@ static int __qlt_send_busy(struct qla_qpair *qpair,
 	ctio24->u.status1.scsi_status = cpu_to_le16(status);
 	/* Memory Barrier */
 	wmb();
-	qla2x00_start_iocbs(vha, qpair->req);
+	if (qpair->reqq_start_iocbs)
+		qpair->reqq_start_iocbs(qpair);
+	else
+		qla2x00_start_iocbs(vha, qpair->req);
 	return 0;
 }
 

drivers/scsi/qla2xxx/qla_target.h

@@ -1097,7 +1097,7 @@ extern int qlt_free_qfull_cmds(struct qla_qpair *);
 extern void qlt_logo_completion_handler(fc_port_t *, int);
 extern void qlt_do_generation_tick(struct scsi_qla_host *, int *);
 
-void qlt_send_resp_ctio(scsi_qla_host_t *, struct qla_tgt_cmd *, uint8_t,
+void qlt_send_resp_ctio(struct qla_qpair *, struct qla_tgt_cmd *, uint8_t,
     uint8_t, uint8_t, uint8_t);
 extern void qlt_abort_cmd_on_host_reset(struct scsi_qla_host *,
     struct qla_tgt_cmd *);