[SCSI] lpfc 8.3.31: Correct handling of SLI4-port XRI resource-provisioning profile change

Signed-off-by: NAlex Iannicelli <alex.iannicelli@emulex.com>
Signed-off-by: NJames Smart <james.smart@emulex.com>
Signed-off-by: NJames Bottomley <JBottomley@Parallels.com>

drivers/scsi/lpfc/lpfc_crtn.h

@@ -460,6 +460,7 @@ int lpfc_hba_init_link_fc_topology(struct lpfc_hba *, uint32_t, uint32_t);
 int lpfc_issue_reg_vfi(struct lpfc_vport *);
 int lpfc_issue_unreg_vfi(struct lpfc_vport *);
 int lpfc_selective_reset(struct lpfc_hba *);
-int lpfc_sli4_read_config(struct lpfc_hba *phba);
-int lpfc_scsi_buf_update(struct lpfc_hba *phba);
-void lpfc_sli4_node_prep(struct lpfc_hba *phba);
+int lpfc_sli4_read_config(struct lpfc_hba *);
+void lpfc_sli4_node_prep(struct lpfc_hba *);
+int lpfc_sli4_xri_sgl_update(struct lpfc_hba *);
+void lpfc_free_sgl_list(struct lpfc_hba *, struct list_head *);

drivers/scsi/lpfc/lpfc_hw4.h

@@ -228,19 +228,15 @@ struct lpfc_sli4_flags {
 #define lpfc_idx_rsrc_rdy_MASK		0x00000001
 #define lpfc_idx_rsrc_rdy_WORD		word0
 #define LPFC_IDX_RSRC_RDY		1
-#define lpfc_xri_rsrc_rdy_SHIFT		1
-#define lpfc_xri_rsrc_rdy_MASK		0x00000001
-#define lpfc_xri_rsrc_rdy_WORD		word0
-#define LPFC_XRI_RSRC_RDY		1
-#define lpfc_rpi_rsrc_rdy_SHIFT		2
+#define lpfc_rpi_rsrc_rdy_SHIFT		1
 #define lpfc_rpi_rsrc_rdy_MASK		0x00000001
 #define lpfc_rpi_rsrc_rdy_WORD		word0
 #define LPFC_RPI_RSRC_RDY		1
-#define lpfc_vpi_rsrc_rdy_SHIFT		3
+#define lpfc_vpi_rsrc_rdy_SHIFT		2
 #define lpfc_vpi_rsrc_rdy_MASK		0x00000001
 #define lpfc_vpi_rsrc_rdy_WORD		word0
 #define LPFC_VPI_RSRC_RDY		1
-#define lpfc_vfi_rsrc_rdy_SHIFT		4
+#define lpfc_vfi_rsrc_rdy_SHIFT		3
 #define lpfc_vfi_rsrc_rdy_MASK		0x00000001
 #define lpfc_vfi_rsrc_rdy_WORD		word0
 #define LPFC_VFI_RSRC_RDY		1

drivers/scsi/lpfc/lpfc_init.c

@@ -64,8 +64,8 @@ static int lpfc_sli4_queue_verify(struct lpfc_hba *);
 static int lpfc_create_bootstrap_mbox(struct lpfc_hba *);
 static int lpfc_setup_endian_order(struct lpfc_hba *);
 static void lpfc_destroy_bootstrap_mbox(struct lpfc_hba *);
-static void lpfc_free_sgl_list(struct lpfc_hba *);
-static int lpfc_init_sgl_list(struct lpfc_hba *);
+static void lpfc_free_els_sgl_list(struct lpfc_hba *);
+static void lpfc_init_sgl_list(struct lpfc_hba *);
 static int lpfc_init_active_sgl_array(struct lpfc_hba *);
 static void lpfc_free_active_sgl(struct lpfc_hba *);
 static int lpfc_hba_down_post_s3(struct lpfc_hba *phba);
@@ -2766,36 +2766,6 @@ lpfc_offline(struct lpfc_hba *phba)
 	lpfc_destroy_vport_work_array(phba, vports);
 }
 
-/**
- * lpfc_scsi_buf_update - Update the scsi_buffers that are already allocated.
- * @phba: pointer to lpfc hba data structure.
- *
- * This routine goes through all the scsi buffers in the system and updates the
- * Physical XRIs assigned to the SCSI buffer because these may change after any
- * firmware reset
- *
- * Return codes
- *   0 - successful (for now, it always returns 0)
- **/
-int
-lpfc_scsi_buf_update(struct lpfc_hba *phba)
-{
-	struct lpfc_scsi_buf *sb, *sb_next;
-
-	spin_lock_irq(&phba->hbalock);
-	spin_lock(&phba->scsi_buf_list_lock);
-	list_for_each_entry_safe(sb, sb_next, &phba->lpfc_scsi_buf_list, list) {
-		sb->cur_iocbq.sli4_xritag =
-			phba->sli4_hba.xri_ids[sb->cur_iocbq.sli4_lxritag];
-		set_bit(sb->cur_iocbq.sli4_lxritag, phba->sli4_hba.xri_bmask);
-		phba->sli4_hba.max_cfg_param.xri_used++;
-		phba->sli4_hba.xri_count++;
-	}
-	spin_unlock(&phba->scsi_buf_list_lock);
-	spin_unlock_irq(&phba->hbalock);
-	return 0;
-}
-
 /**
  * lpfc_scsi_free - Free all the SCSI buffers and IOCBs from driver lists
  * @phba: pointer to lpfc hba data structure.
@@ -2803,11 +2773,8 @@ lpfc_scsi_buf_update(struct lpfc_hba *phba)
  * This routine is to free all the SCSI buffers and IOCBs from the driver
  * list back to kernel. It is called from lpfc_pci_remove_one to free
  * the internal resources before the device is removed from the system.
- *
- * Return codes
- *   0 - successful (for now, it always returns 0)
  **/
-static int
+static void
 lpfc_scsi_free(struct lpfc_hba *phba)
 {
 	struct lpfc_scsi_buf *sb, *sb_next;
@@ -2833,7 +2800,178 @@ lpfc_scsi_free(struct lpfc_hba *phba)
 	}
 
 	spin_unlock_irq(&phba->hbalock);
+}
+
+/**
+ * lpfc_sli4_xri_sgl_update - update xri-sgl sizing and mapping
+ * @phba: pointer to lpfc hba data structure.
+ *
+ * This routine first calculates the sizes of the current els and allocated
+ * scsi sgl lists, and then goes through all sgls to updates the physical
+ * XRIs assigned due to port function reset. During port initialization, the
+ * current els and allocated scsi sgl lists are 0s.
+ *
+ * Return codes
+ *   0 - successful (for now, it always returns 0)
+ **/
+int
+lpfc_sli4_xri_sgl_update(struct lpfc_hba *phba)
+{
+	struct lpfc_sglq *sglq_entry = NULL, *sglq_entry_next = NULL;
+	struct lpfc_scsi_buf *psb = NULL, *psb_next = NULL;
+	uint16_t i, lxri, xri_cnt, els_xri_cnt, scsi_xri_cnt;
+	LIST_HEAD(els_sgl_list);
+	LIST_HEAD(scsi_sgl_list);
+	int rc;
+
+	/*
+	 * update on pci function's els xri-sgl list
+	 */
+	els_xri_cnt = lpfc_sli4_get_els_iocb_cnt(phba);
+	if (els_xri_cnt > phba->sli4_hba.els_xri_cnt) {
+		/* els xri-sgl expanded */
+		xri_cnt = els_xri_cnt - phba->sli4_hba.els_xri_cnt;
+		lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
+				"3157 ELS xri-sgl count increased from "
+				"%d to %d\n", phba->sli4_hba.els_xri_cnt,
+				els_xri_cnt);
+		/* allocate the additional els sgls */
+		for (i = 0; i < xri_cnt; i++) {
+			sglq_entry = kzalloc(sizeof(struct lpfc_sglq),
+					     GFP_KERNEL);
+			if (sglq_entry == NULL) {
+				lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
+						"2562 Failure to allocate an "
+						"ELS sgl entry:%d\n", i);
+				rc = -ENOMEM;
+				goto out_free_mem;
+			}
+			sglq_entry->buff_type = GEN_BUFF_TYPE;
+			sglq_entry->virt = lpfc_mbuf_alloc(phba, 0,
+							   &sglq_entry->phys);
+			if (sglq_entry->virt == NULL) {
+				kfree(sglq_entry);
+				lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
+						"2563 Failure to allocate an "
+						"ELS mbuf:%d\n", i);
+				rc = -ENOMEM;
+				goto out_free_mem;
+			}
+			sglq_entry->sgl = sglq_entry->virt;
+			memset(sglq_entry->sgl, 0, LPFC_BPL_SIZE);
+			sglq_entry->state = SGL_FREED;
+			list_add_tail(&sglq_entry->list, &els_sgl_list);
+		}
+		spin_lock(&phba->hbalock);
+		list_splice_init(&els_sgl_list, &phba->sli4_hba.lpfc_sgl_list);
+		spin_unlock(&phba->hbalock);
+	} else if (els_xri_cnt < phba->sli4_hba.els_xri_cnt) {
+		/* els xri-sgl shrinked */
+		xri_cnt = phba->sli4_hba.els_xri_cnt - els_xri_cnt;
+		lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
+				"3158 ELS xri-sgl count decreased from "
+				"%d to %d\n", phba->sli4_hba.els_xri_cnt,
+				els_xri_cnt);
+		spin_lock_irq(&phba->hbalock);
+		list_splice_init(&phba->sli4_hba.lpfc_sgl_list, &els_sgl_list);
+		spin_unlock_irq(&phba->hbalock);
+		/* release extra els sgls from list */
+		for (i = 0; i < xri_cnt; i++) {
+			list_remove_head(&els_sgl_list,
+					 sglq_entry, struct lpfc_sglq, list);
+			if (sglq_entry) {
+				lpfc_mbuf_free(phba, sglq_entry->virt,
+					       sglq_entry->phys);
+				kfree(sglq_entry);
+			}
+		}
+		spin_lock_irq(&phba->hbalock);
+		list_splice_init(&els_sgl_list, &phba->sli4_hba.lpfc_sgl_list);
+		spin_unlock_irq(&phba->hbalock);
+	} else
+		lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
+				"3163 ELS xri-sgl count unchanged: %d\n",
+				els_xri_cnt);
+	phba->sli4_hba.els_xri_cnt = els_xri_cnt;
+
+	/* update xris to els sgls on the list */
+	sglq_entry = NULL;
+	sglq_entry_next = NULL;
+	list_for_each_entry_safe(sglq_entry, sglq_entry_next,
+				 &phba->sli4_hba.lpfc_sgl_list, list) {
+		lxri = lpfc_sli4_next_xritag(phba);
+		if (lxri == NO_XRI) {
+			lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
+					"2400 Failed to allocate xri for "
+					"ELS sgl\n");
+			rc = -ENOMEM;
+			goto out_free_mem;
+		}
+		sglq_entry->sli4_lxritag = lxri;
+		sglq_entry->sli4_xritag = phba->sli4_hba.xri_ids[lxri];
+	}
+
+	/*
+	 * update on pci function's allocated scsi xri-sgl list
+	 */
+	phba->total_scsi_bufs = 0;
+
+	/* maximum number of xris available for scsi buffers */
+	phba->sli4_hba.scsi_xri_max = phba->sli4_hba.max_cfg_param.max_xri -
+				      els_xri_cnt;
+
+	lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
+			"2401 Current allocated SCSI xri-sgl count:%d, "
+			"maximum  SCSI xri count:%d\n",
+			phba->sli4_hba.scsi_xri_cnt,
+			phba->sli4_hba.scsi_xri_max);
+
+	spin_lock_irq(&phba->scsi_buf_list_lock);
+	list_splice_init(&phba->lpfc_scsi_buf_list, &scsi_sgl_list);
+	spin_unlock_irq(&phba->scsi_buf_list_lock);
+
+	if (phba->sli4_hba.scsi_xri_cnt > phba->sli4_hba.scsi_xri_max) {
+		/* max scsi xri shrinked below the allocated scsi buffers */
+		scsi_xri_cnt = phba->sli4_hba.scsi_xri_cnt -
+					phba->sli4_hba.scsi_xri_max;
+		/* release the extra allocated scsi buffers */
+		for (i = 0; i < scsi_xri_cnt; i++) {
+			list_remove_head(&scsi_sgl_list, psb,
+					 struct lpfc_scsi_buf, list);
+			pci_pool_free(phba->lpfc_scsi_dma_buf_pool, psb->data,
+				      psb->dma_handle);
+			kfree(psb);
+		}
+		spin_lock_irq(&phba->scsi_buf_list_lock);
+		phba->sli4_hba.scsi_xri_cnt -= scsi_xri_cnt;
+		spin_unlock_irq(&phba->scsi_buf_list_lock);
+	}
+
+	/* update xris associated to remaining allocated scsi buffers */
+	psb = NULL;
+	psb_next = NULL;
+	list_for_each_entry_safe(psb, psb_next, &scsi_sgl_list, list) {
+		lxri = lpfc_sli4_next_xritag(phba);
+		if (lxri == NO_XRI) {
+			lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
+					"2560 Failed to allocate xri for "
+					"scsi buffer\n");
+			rc = -ENOMEM;
+			goto out_free_mem;
+		}
+		psb->cur_iocbq.sli4_lxritag = lxri;
+		psb->cur_iocbq.sli4_xritag = phba->sli4_hba.xri_ids[lxri];
+	}
+	spin_lock(&phba->scsi_buf_list_lock);
+	list_splice_init(&scsi_sgl_list, &phba->lpfc_scsi_buf_list);
+	spin_unlock(&phba->scsi_buf_list_lock);
+
 	return 0;
+
+out_free_mem:
+	lpfc_free_els_sgl_list(phba);
+	lpfc_scsi_free(phba);
+	return rc;
 }
 
 /**
@@ -4636,18 +4774,15 @@ lpfc_sli4_driver_resource_setup(struct lpfc_hba *phba)
 	if (rc)
 		goto out_free_bsmbx;
 
-	/* Initialize and populate the iocb list per host */
-	rc = lpfc_init_sgl_list(phba);
-	if (rc) {
-		lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
-				"1400 Failed to initialize sgl list.\n");
-		goto out_destroy_cq_event_pool;
-	}
+	/* Initialize sgl lists per host */
+	lpfc_init_sgl_list(phba);
+
+	/* Allocate and initialize active sgl array */
 	rc = lpfc_init_active_sgl_array(phba);
 	if (rc) {
 		lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
 				"1430 Failed to initialize sgl list.\n");
-		goto out_free_sgl_list;
+		goto out_destroy_cq_event_pool;
 	}
 	rc = lpfc_sli4_init_rpi_hdrs(phba);
 	if (rc) {
@@ -4722,8 +4857,6 @@ lpfc_sli4_driver_resource_setup(struct lpfc_hba *phba)
 	lpfc_sli4_remove_rpi_hdrs(phba);
 out_free_active_sgl:
 	lpfc_free_active_sgl(phba);
-out_free_sgl_list:
-	lpfc_free_sgl_list(phba);
 out_destroy_cq_event_pool:
 	lpfc_sli4_cq_event_pool_destroy(phba);
 out_free_bsmbx:
@@ -4760,10 +4893,7 @@ lpfc_sli4_driver_resource_unset(struct lpfc_hba *phba)
 
 	/* Free the ELS sgl list */
 	lpfc_free_active_sgl(phba);
-	lpfc_free_sgl_list(phba);
-
-	/* Free the SCSI sgl management array */
-	kfree(phba->sli4_hba.lpfc_scsi_psb_array);
+	lpfc_free_els_sgl_list(phba);
 
 	/* Free the completion queue EQ event pool */
 	lpfc_sli4_cq_event_release_all(phba);
@@ -4990,29 +5120,42 @@ lpfc_init_iocb_list(struct lpfc_hba *phba, int iocb_count)
 }
 
 /**
- * lpfc_free_sgl_list - Free sgl list.
+ * lpfc_free_sgl_list - Free a given sgl list.
  * @phba: pointer to lpfc hba data structure.
+ * @sglq_list: pointer to the head of sgl list.
  *
- * This routine is invoked to free the driver's sgl list and memory.
+ * This routine is invoked to free a give sgl list and memory.
  **/
-static void
-lpfc_free_sgl_list(struct lpfc_hba *phba)
+void
+lpfc_free_sgl_list(struct lpfc_hba *phba, struct list_head *sglq_list)
 {
 	struct lpfc_sglq *sglq_entry = NULL, *sglq_next = NULL;
+
+	list_for_each_entry_safe(sglq_entry, sglq_next, sglq_list, list) {
+		list_del(&sglq_entry->list);
+		lpfc_mbuf_free(phba, sglq_entry->virt, sglq_entry->phys);
+		kfree(sglq_entry);
+	}
+}
+
+/**
+ * lpfc_free_els_sgl_list - Free els sgl list.
+ * @phba: pointer to lpfc hba data structure.
+ *
+ * This routine is invoked to free the driver's els sgl list and memory.
+ **/
+static void
+lpfc_free_els_sgl_list(struct lpfc_hba *phba)
+{
 	LIST_HEAD(sglq_list);
 
+	/* Retrieve all els sgls from driver list */
 	spin_lock_irq(&phba->hbalock);
 	list_splice_init(&phba->sli4_hba.lpfc_sgl_list, &sglq_list);
 	spin_unlock_irq(&phba->hbalock);
 
-	list_for_each_entry_safe(sglq_entry, sglq_next,
-				 &sglq_list, list) {
-		list_del(&sglq_entry->list);
-		lpfc_mbuf_free(phba, sglq_entry->virt, sglq_entry->phys);
-		kfree(sglq_entry);
-		phba->sli4_hba.total_sglq_bufs--;
-	}
-	kfree(phba->sli4_hba.lpfc_els_sgl_array);
+	/* Now free the sgl list */
+	lpfc_free_sgl_list(phba, &sglq_list);
 }
 
 /**
@@ -5057,99 +5200,19 @@ lpfc_free_active_sgl(struct lpfc_hba *phba)
  * This routine is invoked to allocate and initizlize the driver's sgl
  * list and set up the sgl xritag tag array accordingly.
  *
- * Return codes
- *	0 - successful
- *	other values - error
  **/
-static int
+static void
 lpfc_init_sgl_list(struct lpfc_hba *phba)
 {
-	struct lpfc_sglq *sglq_entry = NULL;
-	int i;
-	int els_xri_cnt;
-
-	els_xri_cnt = lpfc_sli4_get_els_iocb_cnt(phba);
-	lpfc_printf_log(phba, KERN_INFO, LOG_SLI,
-				"2400 ELS XRI count %d.\n",
-				els_xri_cnt);
 	/* Initialize and populate the sglq list per host/VF. */
 	INIT_LIST_HEAD(&phba->sli4_hba.lpfc_sgl_list);
 	INIT_LIST_HEAD(&phba->sli4_hba.lpfc_abts_els_sgl_list);
 
-	/* Sanity check on XRI management */
-	if (phba->sli4_hba.max_cfg_param.max_xri <= els_xri_cnt) {
-		lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
-				"2562 No room left for SCSI XRI allocation: "
-				"max_xri=%d, els_xri=%d\n",
-				phba->sli4_hba.max_cfg_param.max_xri,
-				els_xri_cnt);
-		return -ENOMEM;
-	}
-
-	/* Allocate memory for the ELS XRI management array */
-	phba->sli4_hba.lpfc_els_sgl_array =
-			kzalloc((sizeof(struct lpfc_sglq *) * els_xri_cnt),
-			GFP_KERNEL);
+	/* els xri-sgl book keeping */
+	phba->sli4_hba.els_xri_cnt = 0;
 
-	if (!phba->sli4_hba.lpfc_els_sgl_array) {
-		lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
-				"2401 Failed to allocate memory for ELS "
-				"XRI management array of size %d.\n",
-				els_xri_cnt);
-		return -ENOMEM;
-	}
-
-	/* Keep the SCSI XRI into the XRI management array */
-	phba->sli4_hba.scsi_xri_max =
-			phba->sli4_hba.max_cfg_param.max_xri - els_xri_cnt;
+	/* scsi xri-buffer book keeping */
 	phba->sli4_hba.scsi_xri_cnt = 0;
-	phba->sli4_hba.lpfc_scsi_psb_array =
-			kzalloc((sizeof(struct lpfc_scsi_buf *) *
-			phba->sli4_hba.scsi_xri_max), GFP_KERNEL);
-
-	if (!phba->sli4_hba.lpfc_scsi_psb_array) {
-		lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
-				"2563 Failed to allocate memory for SCSI "
-				"XRI management array of size %d.\n",
-				phba->sli4_hba.scsi_xri_max);
-		kfree(phba->sli4_hba.lpfc_els_sgl_array);
-		return -ENOMEM;
-	}
-
-	for (i = 0; i < els_xri_cnt; i++) {
-		sglq_entry = kzalloc(sizeof(struct lpfc_sglq), GFP_KERNEL);
-		if (sglq_entry == NULL) {
-			printk(KERN_ERR "%s: only allocated %d sgls of "
-				"expected %d count. Unloading driver.\n",
-				__func__, i, els_xri_cnt);
-			goto out_free_mem;
-		}
-
-		sglq_entry->buff_type = GEN_BUFF_TYPE;
-		sglq_entry->virt = lpfc_mbuf_alloc(phba, 0, &sglq_entry->phys);
-		if (sglq_entry->virt == NULL) {
-			kfree(sglq_entry);
-			printk(KERN_ERR "%s: failed to allocate mbuf.\n"
-				"Unloading driver.\n", __func__);
-			goto out_free_mem;
-		}
-		sglq_entry->sgl = sglq_entry->virt;
-		memset(sglq_entry->sgl, 0, LPFC_BPL_SIZE);
-
-		/* The list order is used by later block SGL registraton */
-		spin_lock_irq(&phba->hbalock);
-		sglq_entry->state = SGL_FREED;
-		list_add_tail(&sglq_entry->list, &phba->sli4_hba.lpfc_sgl_list);
-		phba->sli4_hba.lpfc_els_sgl_array[i] = sglq_entry;
-		phba->sli4_hba.total_sglq_bufs++;
-		spin_unlock_irq(&phba->hbalock);
-	}
-	return 0;
-
-out_free_mem:
-	kfree(phba->sli4_hba.lpfc_scsi_psb_array);
-	lpfc_free_sgl_list(phba);
-	return -ENOMEM;
 }
 
 /**
@@ -7320,9 +7383,11 @@ lpfc_pci_function_reset(struct lpfc_hba *phba)
 					phba->sli4_hba.u.if_type2.ERR2regaddr);
 				lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
 					"2890 Port error detected during port "
-					"reset(%d): port status reg 0x%x, "
+					"reset(%d): wait_tmo:%d ms, "
+					"port status reg 0x%x, "
 					"error 1=0x%x, error 2=0x%x\n",
-					num_resets, reg_data.word0,
+					num_resets, rdy_chk*10,
+					reg_data.word0,
 					phba->work_status[0],
 					phba->work_status[1]);
 				rc = -ENODEV;
@@ -9118,8 +9183,12 @@ lpfc_sli4_get_els_iocb_cnt(struct lpfc_hba *phba)
 			return 50;
 		else if (max_xri <= 1024)
 			return 100;
-		else
+		else if (max_xri <= 1536)
 			return 150;
+		else if (max_xri <= 2048)
+			return 200;
+		else
+			return 250;
 	} else
 		return 0;
 }

drivers/scsi/lpfc/lpfc_scsi.c

@@ -718,72 +718,162 @@ lpfc_sli4_fcp_xri_aborted(struct lpfc_hba *phba,
 }
 
 /**
- * lpfc_sli4_repost_scsi_sgl_list - Repsot the Scsi buffers sgl pages as block
+ * lpfc_sli4_post_scsi_sgl_list - Psot blocks of scsi buffer sgls from a list
  * @phba: pointer to lpfc hba data structure.
+ * @post_sblist: pointer to the scsi buffer list.
  *
- * This routine walks the list of scsi buffers that have been allocated and
- * repost them to the HBA by using SGL block post. This is needed after a
- * pci_function_reset/warm_start or start. The lpfc_hba_down_post_s4 routine
- * is responsible for moving all scsi buffers on the lpfc_abts_scsi_sgl_list
- * to the lpfc_scsi_buf_list. If the repost fails, reject all scsi buffers.
+ * This routine walks a list of scsi buffers that was passed in. It attempts
+ * to construct blocks of scsi buffer sgls which contains contiguous xris and
+ * uses the non-embedded SGL block post mailbox commands to post to the port.
+ * For single SCSI buffer sgl with non-contiguous xri, if any, it shall use
+ * embedded SGL post mailbox command for posting. The @post_sblist passed in
+ * must be local list, thus no lock is needed when manipulate the list.
  *
- * Returns: 0 = success, non-zero failure.
+ * Returns: 0 = failure, non-zero number of successfully posted buffers.
  **/
 int
-lpfc_sli4_repost_scsi_sgl_list(struct lpfc_hba *phba)
+lpfc_sli4_post_scsi_sgl_list(struct lpfc_hba *phba,
+			     struct list_head *post_sblist, int sb_count)
 {
-	struct lpfc_scsi_buf *psb;
-	int index, status, bcnt = 0, rcnt = 0, rc = 0;
-	LIST_HEAD(sblist);
-
-	for (index = 0; index < phba->sli4_hba.scsi_xri_cnt; index++) {
-		psb = phba->sli4_hba.lpfc_scsi_psb_array[index];
-		if (psb) {
-			/* Remove from SCSI buffer list */
-			list_del(&psb->list);
-			/* Add it to a local SCSI buffer list */
-			list_add_tail(&psb->list, &sblist);
-			if (++rcnt == LPFC_NEMBED_MBOX_SGL_CNT) {
-				bcnt = rcnt;
-				rcnt = 0;
+	struct lpfc_scsi_buf *psb, *psb_next;
+	int status;
+	int post_cnt = 0, block_cnt = 0, num_posting = 0, num_posted = 0;
+	dma_addr_t pdma_phys_bpl1;
+	int last_xritag = NO_XRI;
+	LIST_HEAD(prep_sblist);
+	LIST_HEAD(blck_sblist);
+	LIST_HEAD(scsi_sblist);
+
+	/* sanity check */
+	if (sb_count <= 0)
+		return -EINVAL;
+
+	list_for_each_entry_safe(psb, psb_next, post_sblist, list) {
+		list_del_init(&psb->list);
+		block_cnt++;
+		if ((last_xritag != NO_XRI) &&
+		    (psb->cur_iocbq.sli4_xritag != last_xritag + 1)) {
+			/* a hole in xri block, form a sgl posting block */
+			list_splice_init(&prep_sblist, &blck_sblist);
+			post_cnt = block_cnt - 1;
+			/* prepare list for next posting block */
+			list_add_tail(&psb->list, &prep_sblist);
+			block_cnt = 1;
+		} else {
+			/* prepare list for next posting block */
+			list_add_tail(&psb->list, &prep_sblist);
+			/* enough sgls for non-embed sgl mbox command */
+			if (block_cnt == LPFC_NEMBED_MBOX_SGL_CNT) {
+				list_splice_init(&prep_sblist, &blck_sblist);
+				post_cnt = block_cnt;
+				block_cnt = 0;
 			}
-		} else
-			/* A hole present in the XRI array, need to skip */
-			bcnt = rcnt;
+		}
+		num_posting++;
+		last_xritag = psb->cur_iocbq.sli4_xritag;
 
-		if (index == phba->sli4_hba.scsi_xri_cnt - 1)
-			/* End of XRI array for SCSI buffer, complete */
-			bcnt = rcnt;
+		/* end of repost sgl list condition for SCSI buffers */
+		if (num_posting == sb_count) {
+			if (post_cnt == 0) {
+				/* last sgl posting block */
+				list_splice_init(&prep_sblist, &blck_sblist);
+				post_cnt = block_cnt;
+			} else if (block_cnt == 1) {
+				/* last single sgl with non-contiguous xri */
+				if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE)
+					pdma_phys_bpl1 = psb->dma_phys_bpl +
+								SGL_PAGE_SIZE;
+				else
+					pdma_phys_bpl1 = 0;
+				status = lpfc_sli4_post_sgl(phba,
+						psb->dma_phys_bpl,
+						pdma_phys_bpl1,
+						psb->cur_iocbq.sli4_xritag);
+				if (status) {
+					/* failure, put on abort scsi list */
+					psb->exch_busy = 1;
+				} else {
+					/* success, put on SCSI buffer list */
+					psb->exch_busy = 0;
+					psb->status = IOSTAT_SUCCESS;
+					num_posted++;
+				}
+				/* success, put on SCSI buffer sgl list */
+				list_add_tail(&psb->list, &scsi_sblist);
+			}
+		}
 
-		/* Continue until collect up to a nembed page worth of sgls */
-		if (bcnt == 0)
+		/* continue until a nembed page worth of sgls */
+		if (post_cnt == 0)
 			continue;
-		/* Now, post the SCSI buffer list sgls as a block */
-		if (!phba->sli4_hba.extents_in_use)
-			status = lpfc_sli4_post_scsi_sgl_block(phba,
-							&sblist,
-							bcnt);
-		else
-			status = lpfc_sli4_post_scsi_sgl_blk_ext(phba,
-							&sblist,
-							bcnt);
-		/* Reset SCSI buffer count for next round of posting */
-		bcnt = 0;
-		while (!list_empty(&sblist)) {
-			list_remove_head(&sblist, psb, struct lpfc_scsi_buf,
-					 list);
+
+		/* post block of SCSI buffer list sgls */
+		status = lpfc_sli4_post_scsi_sgl_block(phba, &blck_sblist,
+						       post_cnt);
+
+		/* don't reset xirtag due to hole in xri block */
+		if (block_cnt == 0)
+			last_xritag = NO_XRI;
+
+		/* reset SCSI buffer post count for next round of posting */
+		post_cnt = 0;
+
+		/* put posted SCSI buffer-sgl posted on SCSI buffer sgl list */
+		while (!list_empty(&blck_sblist)) {
+			list_remove_head(&blck_sblist, psb,
+					 struct lpfc_scsi_buf, list);
 			if (status) {
-				/* Put this back on the abort scsi list */
+				/* failure, put on abort scsi list */
 				psb->exch_busy = 1;
-				rc++;
 			} else {
+				/* success, put on SCSI buffer list */
 				psb->exch_busy = 0;
 				psb->status = IOSTAT_SUCCESS;
+				num_posted++;
 			}
-			/* Put it back into the SCSI buffer list */
-			lpfc_release_scsi_buf_s4(phba, psb);
+			list_add_tail(&psb->list, &scsi_sblist);
 		}
 	}
+	/* Push SCSI buffers with sgl posted to the availble list */
+	while (!list_empty(&scsi_sblist)) {
+		list_remove_head(&scsi_sblist, psb,
+				 struct lpfc_scsi_buf, list);
+		lpfc_release_scsi_buf_s4(phba, psb);
+	}
+	return num_posted;
+}
+
+/**
+ * lpfc_sli4_repost_scsi_sgl_list - Repsot all the allocated scsi buffer sgls
+ * @phba: pointer to lpfc hba data structure.
+ *
+ * This routine walks the list of scsi buffers that have been allocated and
+ * repost them to the port by using SGL block post. This is needed after a
+ * pci_function_reset/warm_start or start. The lpfc_hba_down_post_s4 routine
+ * is responsible for moving all scsi buffers on the lpfc_abts_scsi_sgl_list
+ * to the lpfc_scsi_buf_list. If the repost fails, reject all scsi buffers.
+ *
+ * Returns: 0 = success, non-zero failure.
+ **/
+int
+lpfc_sli4_repost_scsi_sgl_list(struct lpfc_hba *phba)
+{
+	LIST_HEAD(post_sblist);
+	int num_posted, rc = 0;
+
+	/* get all SCSI buffers need to repost to a local list */
+	spin_lock(&phba->scsi_buf_list_lock);
+	list_splice_init(&phba->lpfc_scsi_buf_list, &post_sblist);
+	spin_unlock(&phba->scsi_buf_list_lock);
+
+	/* post the list of scsi buffer sgls to port if available */
+	if (!list_empty(&post_sblist)) {
+		num_posted = lpfc_sli4_post_scsi_sgl_list(phba, &post_sblist,
+						phba->sli4_hba.scsi_xri_cnt);
+		/* failed to post any scsi buffer, return error */
+		if (num_posted == 0)
+			rc = -EIO;
+	}
 	return rc;
 }
 
@@ -792,12 +882,13 @@ lpfc_sli4_repost_scsi_sgl_list(struct lpfc_hba *phba)
  * @vport: The virtual port for which this call being executed.
  * @num_to_allocate: The requested number of buffers to allocate.
  *
- * This routine allocates a scsi buffer for device with SLI-4 interface spec,
+ * This routine allocates scsi buffers for device with SLI-4 interface spec,
  * the scsi buffer contains all the necessary information needed to initiate
- * a SCSI I/O.
+ * a SCSI I/O. After allocating up to @num_to_allocate SCSI buffers and put
+ * them on a list, it post them to the port by using SGL block post.
  *
  * Return codes:
- *   int - number of scsi buffers that were allocated.
+ *   int - number of scsi buffers that were allocated and posted.
  *   0 = failure, less than num_to_alloc is a partial failure.
  **/
 static int
@@ -810,22 +901,21 @@ lpfc_new_scsi_buf_s4(struct lpfc_vport *vport, int num_to_alloc)
 	dma_addr_t pdma_phys_fcp_cmd;
 	dma_addr_t pdma_phys_fcp_rsp;
 	dma_addr_t pdma_phys_bpl, pdma_phys_bpl1;
-	uint16_t iotag, last_xritag = NO_XRI, lxri = 0;
-	int status = 0, index;
-	int bcnt;
-	int non_sequential_xri = 0;
-	LIST_HEAD(sblist);
+	uint16_t iotag, lxri = 0;
+	int bcnt, num_posted;
+	LIST_HEAD(prep_sblist);
+	LIST_HEAD(post_sblist);
+	LIST_HEAD(scsi_sblist);
 
 	for (bcnt = 0; bcnt < num_to_alloc; bcnt++) {
 		psb = kzalloc(sizeof(struct lpfc_scsi_buf), GFP_KERNEL);
 		if (!psb)
 			break;
-
 		/*
-		 * Get memory from the pci pool to map the virt space to pci bus
-		 * space for an I/O.  The DMA buffer includes space for the
-		 * struct fcp_cmnd, struct fcp_rsp and the number of bde's
-		 * necessary to support the sg_tablesize.
+		 * Get memory from the pci pool to map the virt space to
+		 * pci bus space for an I/O. The DMA buffer includes space
+		 * for the struct fcp_cmnd, struct fcp_rsp and the number
+		 * of bde's necessary to support the sg_tablesize.
 		 */
 		psb->data = pci_pool_alloc(phba->lpfc_scsi_dma_buf_pool,
 						GFP_KERNEL, &psb->dma_handle);
@@ -833,8 +923,6 @@ lpfc_new_scsi_buf_s4(struct lpfc_vport *vport, int num_to_alloc)
 			kfree(psb);
 			break;
 		}
-
-		/* Initialize virtual ptrs to dma_buf region. */
 		memset(psb->data, 0, phba->cfg_sg_dma_buf_size);
 
 		/* Allocate iotag for psb->cur_iocbq. */
@@ -855,16 +943,7 @@ lpfc_new_scsi_buf_s4(struct lpfc_vport *vport, int num_to_alloc)
 		}
 		psb->cur_iocbq.sli4_lxritag = lxri;
 		psb->cur_iocbq.sli4_xritag = phba->sli4_hba.xri_ids[lxri];
-		if (last_xritag != NO_XRI
-			&& psb->cur_iocbq.sli4_xritag != (last_xritag+1)) {
-			non_sequential_xri = 1;
-		} else
-			list_add_tail(&psb->list, &sblist);
-		last_xritag = psb->cur_iocbq.sli4_xritag;
-
-		index = phba->sli4_hba.scsi_xri_cnt++;
 		psb->cur_iocbq.iocb_flag |= LPFC_IO_FCP;
-
 		psb->fcp_bpl = psb->data;
 		psb->fcp_cmnd = (psb->data + phba->cfg_sg_dma_buf_size)
 			- (sizeof(struct fcp_cmnd) + sizeof(struct fcp_rsp));
@@ -880,9 +959,9 @@ lpfc_new_scsi_buf_s4(struct lpfc_vport *vport, int num_to_alloc)
 		pdma_phys_fcp_rsp = pdma_phys_fcp_cmd + sizeof(struct fcp_cmnd);
 
 		/*
-		 * The first two bdes are the FCP_CMD and FCP_RSP.  The balance
-		 * are sg list bdes.  Initialize the first two and leave the
-		 * rest for queuecommand.
+		 * The first two bdes are the FCP_CMD and FCP_RSP.
+		 * The balance are sg list bdes. Initialize the
+		 * first two and leave the rest for queuecommand.
 		 */
 		sgl->addr_hi = cpu_to_le32(putPaddrHigh(pdma_phys_fcp_cmd));
 		sgl->addr_lo = cpu_to_le32(putPaddrLow(pdma_phys_fcp_cmd));
@@ -917,62 +996,31 @@ lpfc_new_scsi_buf_s4(struct lpfc_vport *vport, int num_to_alloc)
 		iocb->ulpBdeCount = 1;
 		iocb->ulpLe = 1;
 		iocb->ulpClass = CLASS3;
-		psb->cur_iocbq.context1  = psb;
+		psb->cur_iocbq.context1 = psb;
 		if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE)
 			pdma_phys_bpl1 = pdma_phys_bpl + SGL_PAGE_SIZE;
 		else
 			pdma_phys_bpl1 = 0;
 		psb->dma_phys_bpl = pdma_phys_bpl;
-		phba->sli4_hba.lpfc_scsi_psb_array[index] = psb;
-		if (non_sequential_xri) {
-			status = lpfc_sli4_post_sgl(phba, pdma_phys_bpl,
-						pdma_phys_bpl1,
-						psb->cur_iocbq.sli4_xritag);
-			if (status) {
-				/* Put this back on the abort scsi list */
-				psb->exch_busy = 1;
-			} else {
-				psb->exch_busy = 0;
-				psb->status = IOSTAT_SUCCESS;
-			}
-			/* Put it back into the SCSI buffer list */
-			lpfc_release_scsi_buf_s4(phba, psb);
-			break;
-		}
-	}
-	if (bcnt) {
-		if (!phba->sli4_hba.extents_in_use)
-			status = lpfc_sli4_post_scsi_sgl_block(phba,
-								&sblist,
-								bcnt);
-		else
-			status = lpfc_sli4_post_scsi_sgl_blk_ext(phba,
-								&sblist,
-								bcnt);
-
-		if (status) {
-			lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
-					"3021 SCSI SGL post error %d\n",
-					status);
-			bcnt = 0;
-		}
-		/* Reset SCSI buffer count for next round of posting */
-		while (!list_empty(&sblist)) {
-			list_remove_head(&sblist, psb, struct lpfc_scsi_buf,
-				 list);
-			if (status) {
-				/* Put this back on the abort scsi list */
-				psb->exch_busy = 1;
-			} else {
-				psb->exch_busy = 0;
-				psb->status = IOSTAT_SUCCESS;
-			}
-			/* Put it back into the SCSI buffer list */
-			lpfc_release_scsi_buf_s4(phba, psb);
-		}
+
+		/* add the scsi buffer to a post list */
+		list_add_tail(&psb->list, &post_sblist);
+		spin_lock_irq(&phba->scsi_buf_list_lock);
+		phba->sli4_hba.scsi_xri_cnt++;
+		spin_unlock_irq(&phba->scsi_buf_list_lock);
 	}
+	lpfc_printf_log(phba, KERN_INFO, LOG_BG,
+			"3021 Allocate %d out of %d requested new SCSI "
+			"buffers\n", bcnt, num_to_alloc);
+
+	/* post the list of scsi buffer sgls to port if available */
+	if (!list_empty(&post_sblist))
+		num_posted = lpfc_sli4_post_scsi_sgl_list(phba,
+							  &post_sblist, bcnt);
+	else
+		num_posted = 0;
 
-	return bcnt + non_sequential_xri;
+	return num_posted;
 }
 
 /**

drivers/scsi/lpfc/lpfc_sli4.h

@@ -493,14 +493,12 @@ struct lpfc_sli4_hba {
 	uint16_t next_rpi;
 	uint16_t scsi_xri_max;
 	uint16_t scsi_xri_cnt;
+	uint16_t els_xri_cnt;
 	uint16_t scsi_xri_start;
 	struct list_head lpfc_free_sgl_list;
 	struct list_head lpfc_sgl_list;
-	struct lpfc_sglq **lpfc_els_sgl_array;
 	struct list_head lpfc_abts_els_sgl_list;
-	struct lpfc_scsi_buf **lpfc_scsi_psb_array;
 	struct list_head lpfc_abts_scsi_buf_list;
-	uint32_t total_sglq_bufs;
 	struct lpfc_sglq **lpfc_sglq_active_list;
 	struct list_head lpfc_rpi_hdr_list;
 	unsigned long *rpi_bmask;
@@ -509,7 +507,6 @@ struct lpfc_sli4_hba {
 	struct list_head lpfc_rpi_blk_list;
 	unsigned long *xri_bmask;
 	uint16_t *xri_ids;
-	uint16_t xri_count;
 	struct list_head lpfc_xri_blk_list;
 	unsigned long *vfi_bmask;
 	uint16_t *vfi_ids;
@@ -614,11 +611,7 @@ int lpfc_sli4_post_sgl(struct lpfc_hba *, dma_addr_t, dma_addr_t, uint16_t);
 int lpfc_sli4_repost_scsi_sgl_list(struct lpfc_hba *);
 uint16_t lpfc_sli4_next_xritag(struct lpfc_hba *);
 int lpfc_sli4_post_async_mbox(struct lpfc_hba *);
-int lpfc_sli4_post_els_sgl_list(struct lpfc_hba *phba);
-int lpfc_sli4_post_els_sgl_list_ext(struct lpfc_hba *phba);
 int lpfc_sli4_post_scsi_sgl_block(struct lpfc_hba *, struct list_head *, int);
-int lpfc_sli4_post_scsi_sgl_blk_ext(struct lpfc_hba *, struct list_head *,
-				    int);
 struct lpfc_cq_event *__lpfc_sli4_cq_event_alloc(struct lpfc_hba *);
 struct lpfc_cq_event *lpfc_sli4_cq_event_alloc(struct lpfc_hba *);
 void __lpfc_sli4_cq_event_release(struct lpfc_hba *, struct lpfc_cq_event *);