[SCSI] lpfc 8.3.33: Misc changes to optimize critical path

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

2 files changed, 29 insertions, 14 deletions

diff --git a/drivers/scsi/lpfc/lpfc_scsi.c b/drivers/scsi/lpfc/lpfc_scsi.c
index bb9a2249b75d..f035f3c7ec09 100644
--- a/drivers/scsi/lpfc/lpfc_scsi.c
+++ b/drivers/scsi/lpfc/lpfc_scsi.c
@@ -3919,6 +3919,8 @@ lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
         struct lpfc_iocbq *piocbq = &(lpfc_cmd->cur_iocbq);
         int datadir = scsi_cmnd->sc_data_direction;
         char tag[2];
+        uint8_t *ptr;
+        bool sli4;
 
         if (!pnode || !NLP_CHK_NODE_ACT(pnode))
                 return;
@@ -3930,8 +3932,13 @@ lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
         int_to_scsilun(lpfc_cmd->pCmd->device->lun,
                         &lpfc_cmd->fcp_cmnd->fcp_lun);
 
-        memset(&fcp_cmnd->fcpCdb[0], 0, LPFC_FCP_CDB_LEN);
-        memcpy(&fcp_cmnd->fcpCdb[0], scsi_cmnd->cmnd, scsi_cmnd->cmd_len);
+        ptr = &fcp_cmnd->fcpCdb[0];
+        memcpy(ptr, scsi_cmnd->cmnd, scsi_cmnd->cmd_len);
+        if (scsi_cmnd->cmd_len < LPFC_FCP_CDB_LEN) {
+                ptr += scsi_cmnd->cmd_len;
+                memset(ptr, 0, (LPFC_FCP_CDB_LEN - scsi_cmnd->cmd_len));
+        }
+
         if (scsi_populate_tag_msg(scsi_cmnd, tag)) {
                 switch (tag[0]) {
                 case HEAD_OF_QUEUE_TAG:
@@ -3947,6 +3954,8 @@ lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
         } else
                 fcp_cmnd->fcpCntl1 = 0;
 
+        sli4 = (phba->sli_rev == LPFC_SLI_REV4);
+
         /*
          * There are three possibilities here - use scatter-gather segment, use
          * the single mapping, or neither.  Start the lpfc command prep by
@@ -3956,11 +3965,12 @@ lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
         if (scsi_sg_count(scsi_cmnd)) {
                 if (datadir == DMA_TO_DEVICE) {
                         iocb_cmd->ulpCommand = CMD_FCP_IWRITE64_CR;
-                        if (phba->sli_rev < LPFC_SLI_REV4) {
+                        if (sli4)
+                                iocb_cmd->ulpPU = PARM_READ_CHECK;
+                        else {
                                 iocb_cmd->un.fcpi.fcpi_parm = 0;
                                 iocb_cmd->ulpPU = 0;
-                        } else
-                                iocb_cmd->ulpPU = PARM_READ_CHECK;
+                        }
                         fcp_cmnd->fcpCntl3 = WRITE_DATA;
                         phba->fc4OutputRequests++;
                 } else {
@@ -3984,7 +3994,7 @@ lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
          * of the scsi_cmnd request_buffer
          */
         piocbq->iocb.ulpContext = pnode->nlp_rpi;
-        if (phba->sli_rev == LPFC_SLI_REV4)
+        if (sli4)
                 piocbq->iocb.ulpContext =
                   phba->sli4_hba.rpi_ids[pnode->nlp_rpi];
         if (pnode->nlp_fcp_info & NLP_FCP_2_DEVICE)
diff --git a/drivers/scsi/lpfc/lpfc_sli.c b/drivers/scsi/lpfc/lpfc_sli.c
index 51c7fc746de1..d7afd0fb4579 100644
--- a/drivers/scsi/lpfc/lpfc_sli.c
+++ b/drivers/scsi/lpfc/lpfc_sli.c
@@ -94,6 +94,7 @@ lpfc_sli4_wq_put(struct lpfc_queue *q, union lpfc_wqe *wqe)
         union lpfc_wqe *temp_wqe;
         struct lpfc_register doorbell;
         uint32_t host_index;
+        uint32_t idx;
 
         /* sanity check on queue memory */
         if (unlikely(!q))
@@ -101,7 +102,8 @@ lpfc_sli4_wq_put(struct lpfc_queue *q, union lpfc_wqe *wqe)
         temp_wqe = q->qe[q->host_index].wqe;
 
         /* If the host has not yet processed the next entry then we are done */
-        if (((q->host_index + 1) % q->entry_count) == q->hba_index) {
+        idx = ((q->host_index + 1) % q->entry_count);
+        if (idx == q->hba_index) {
                 q->WQ_overflow++;
                 return -ENOMEM;
         }
@@ -115,7 +117,8 @@ lpfc_sli4_wq_put(struct lpfc_queue *q, union lpfc_wqe *wqe)
 
         /* Update the host index before invoking device */
         host_index = q->host_index;
-        q->host_index = ((q->host_index + 1) % q->entry_count);
+
+        q->host_index = idx;
 
         /* Ring Doorbell */
         doorbell.word0 = 0;
@@ -123,7 +126,6 @@ lpfc_sli4_wq_put(struct lpfc_queue *q, union lpfc_wqe *wqe)
         bf_set(lpfc_wq_doorbell_index, &doorbell, host_index);
         bf_set(lpfc_wq_doorbell_id, &doorbell, q->queue_id);
         writel(doorbell.word0, q->phba->sli4_hba.WQDBregaddr);
-        readl(q->phba->sli4_hba.WQDBregaddr); /* Flush */
 
         return 0;
 }
@@ -197,7 +199,6 @@ lpfc_sli4_mq_put(struct lpfc_queue *q, struct lpfc_mqe *mqe)
         bf_set(lpfc_mq_doorbell_num_posted, &doorbell, 1);
         bf_set(lpfc_mq_doorbell_id, &doorbell, q->queue_id);
         writel(doorbell.word0, q->phba->sli4_hba.MQDBregaddr);
-        readl(q->phba->sli4_hba.MQDBregaddr); /* Flush */
         return 0;
 }
 
@@ -237,6 +238,7 @@ static struct lpfc_eqe *
 lpfc_sli4_eq_get(struct lpfc_queue *q)
 {
         struct lpfc_eqe *eqe;
+        uint32_t idx;
 
         /* sanity check on queue memory */
         if (unlikely(!q))
@@ -247,10 +249,11 @@ lpfc_sli4_eq_get(struct lpfc_queue *q)
         if (!bf_get_le32(lpfc_eqe_valid, eqe))
                 return NULL;
         /* If the host has not yet processed the next entry then we are done */
-        if (((q->hba_index + 1) % q->entry_count) == q->host_index)
+        idx = ((q->hba_index + 1) % q->entry_count);
+        if (idx == q->host_index)
                 return NULL;
 
-        q->hba_index = ((q->hba_index + 1) % q->entry_count);
+        q->hba_index = idx;
         return eqe;
 }
 
@@ -321,6 +324,7 @@ static struct lpfc_cqe *
 lpfc_sli4_cq_get(struct lpfc_queue *q)
 {
         struct lpfc_cqe *cqe;
+        uint32_t idx;
 
         /* sanity check on queue memory */
         if (unlikely(!q))
@@ -330,11 +334,12 @@ lpfc_sli4_cq_get(struct lpfc_queue *q)
         if (!bf_get_le32(lpfc_cqe_valid, q->qe[q->hba_index].cqe))
                 return NULL;
         /* If the host has not yet processed the next entry then we are done */
-        if (((q->hba_index + 1) % q->entry_count) == q->host_index)
+        idx = ((q->hba_index + 1) % q->entry_count);
+        if (idx == q->host_index)
                 return NULL;
 
         cqe = q->qe[q->hba_index].cqe;
-        q->hba_index = ((q->hba_index + 1) % q->entry_count);
+        q->hba_index = idx;
         return cqe;
 }