scsi: lpfc: Fix random heartbeat timeouts during heavy IO

NVME targets appear to randomly disconnect from the initiator when
running heavy IO.

The error is due to the host aggregate (across all controllers) io load
was beyond the maximum exchange count for nvme on the adapter. The
driver was properly returning a resource busy status, but the io load
was so great heartbeat commands would be bounced and not have a
successful retry within the fuzz amount for the nvme heartbeat (yes, a
very high io load!). Thus the target was terminating the controller due
to a keep alive failure.

Resolve by reserving a few exchanges (by counters) which can be used
when the adapter is out of normal exchanges and the command is a NVME
heartbeat command. As counters are used, while the reserved command is
outstanding, as soon as any other exchange completes, the counters are
adjusted and the reserved count is replenished. The heartbeat completes
execution in a normal fashion.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

drivers/scsi/lpfc/lpfc.h

@@ -945,6 +945,8 @@ struct lpfc_hba {
 	struct list_head lpfc_nvme_buf_list_get;
 	struct list_head lpfc_nvme_buf_list_put;
 	uint32_t total_nvme_bufs;
+	uint32_t get_nvme_bufs;
+	uint32_t put_nvme_bufs;
 	struct list_head lpfc_iocb_list;
 	uint32_t total_iocbq_bufs;
 	struct list_head active_rrq_list;

drivers/scsi/lpfc/lpfc_init.c

@@ -1034,6 +1034,7 @@ lpfc_hba_down_post_s4(struct lpfc_hba *phba)
 	LIST_HEAD(nvmet_aborts);
 	unsigned long iflag = 0;
 	struct lpfc_sglq *sglq_entry = NULL;
+	int cnt;
 
 
 	lpfc_sli_hbqbuf_free_all(phba);
@@ -1090,11 +1091,14 @@ lpfc_hba_down_post_s4(struct lpfc_hba *phba)
 	spin_unlock_irqrestore(&phba->scsi_buf_list_put_lock, iflag);
 
 	if (phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME) {
+		cnt = 0;
 		list_for_each_entry_safe(psb, psb_next, &nvme_aborts, list) {
 			psb->pCmd = NULL;
 			psb->status = IOSTAT_SUCCESS;
+			cnt++;
 		}
 		spin_lock_irqsave(&phba->nvme_buf_list_put_lock, iflag);
+		phba->put_nvme_bufs += cnt;
 		list_splice(&nvme_aborts, &phba->lpfc_nvme_buf_list_put);
 		spin_unlock_irqrestore(&phba->nvme_buf_list_put_lock, iflag);
 
@@ -3339,6 +3343,7 @@ lpfc_nvme_free(struct lpfc_hba *phba)
 	list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
 				 &phba->lpfc_nvme_buf_list_put, list) {
 		list_del(&lpfc_ncmd->list);
+		phba->put_nvme_bufs--;
 		dma_pool_free(phba->lpfc_sg_dma_buf_pool, lpfc_ncmd->data,
 			      lpfc_ncmd->dma_handle);
 		kfree(lpfc_ncmd);
@@ -3350,6 +3355,7 @@ lpfc_nvme_free(struct lpfc_hba *phba)
 	list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
 				 &phba->lpfc_nvme_buf_list_get, list) {
 		list_del(&lpfc_ncmd->list);
+		phba->get_nvme_bufs--;
 		dma_pool_free(phba->lpfc_sg_dma_buf_pool, lpfc_ncmd->data,
 			      lpfc_ncmd->dma_handle);
 		kfree(lpfc_ncmd);
@@ -3754,9 +3760,11 @@ lpfc_sli4_nvme_sgl_update(struct lpfc_hba *phba)
 	uint16_t i, lxri, els_xri_cnt;
 	uint16_t nvme_xri_cnt, nvme_xri_max;
 	LIST_HEAD(nvme_sgl_list);
-	int rc;
+	int rc, cnt;
 
 	phba->total_nvme_bufs = 0;
+	phba->get_nvme_bufs = 0;
+	phba->put_nvme_bufs = 0;
 
 	if (!(phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME))
 		return 0;
@@ -3780,6 +3788,9 @@ lpfc_sli4_nvme_sgl_update(struct lpfc_hba *phba)
 	spin_lock(&phba->nvme_buf_list_put_lock);
 	list_splice_init(&phba->lpfc_nvme_buf_list_get, &nvme_sgl_list);
 	list_splice(&phba->lpfc_nvme_buf_list_put, &nvme_sgl_list);
+	cnt = phba->get_nvme_bufs + phba->put_nvme_bufs;
+	phba->get_nvme_bufs = 0;
+	phba->put_nvme_bufs = 0;
 	spin_unlock(&phba->nvme_buf_list_put_lock);
 	spin_unlock_irq(&phba->nvme_buf_list_get_lock);
 
@@ -3824,6 +3835,7 @@ lpfc_sli4_nvme_sgl_update(struct lpfc_hba *phba)
 	spin_lock_irq(&phba->nvme_buf_list_get_lock);
 	spin_lock(&phba->nvme_buf_list_put_lock);
 	list_splice_init(&nvme_sgl_list, &phba->lpfc_nvme_buf_list_get);
+	phba->get_nvme_bufs = cnt;
 	INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put);
 	spin_unlock(&phba->nvme_buf_list_put_lock);
 	spin_unlock_irq(&phba->nvme_buf_list_get_lock);
@@ -5609,8 +5621,10 @@ lpfc_setup_driver_resource_phase1(struct lpfc_hba *phba)
 		/* Initialize the NVME buffer list used by driver for NVME IO */
 		spin_lock_init(&phba->nvme_buf_list_get_lock);
 		INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_get);
+		phba->get_nvme_bufs = 0;
 		spin_lock_init(&phba->nvme_buf_list_put_lock);
 		INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put);
+		phba->put_nvme_bufs = 0;
 	}
 
 	/* Initialize the fabric iocb list */

drivers/scsi/lpfc/lpfc_nvme.c

@@ -57,7 +57,8 @@
 /* NVME initiator-based functions */
 
 static struct lpfc_nvme_buf *
-lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp);
+lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
+		  int expedite);
 
 static void
 lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_nvme_buf *);
@@ -1265,6 +1266,7 @@ lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
 			struct nvmefc_fcp_req *pnvme_fcreq)
 {
 	int ret = 0;
+	int expedite = 0;
 	struct lpfc_nvme_lport *lport;
 	struct lpfc_vport *vport;
 	struct lpfc_hba *phba;
@@ -1273,6 +1275,7 @@ lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
 	struct lpfc_nvme_rport *rport;
 	struct lpfc_nvme_qhandle *lpfc_queue_info;
 	struct lpfc_nvme_fcpreq_priv *freqpriv;
+	struct nvme_common_command *sqe;
 #ifdef CONFIG_SCSI_LPFC_DEBUG_FS
 	uint64_t start = 0;
 #endif
@@ -1354,15 +1357,27 @@ lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
 
 	}
 
+	/* Currently only NVME Keep alive commands should be expedited
+	 * if the driver runs out of a resource. These should only be
+	 * issued on the admin queue, qidx 0
+	 */
+	if (!lpfc_queue_info->qidx && !pnvme_fcreq->sg_cnt) {
+		sqe = &((struct nvme_fc_cmd_iu *)
+			pnvme_fcreq->cmdaddr)->sqe.common;
+		if (sqe->opcode == nvme_admin_keep_alive)
+			expedite = 1;
+	}
+
 	/* The node is shared with FCP IO, make sure the IO pending count does
 	 * not exceed the programmed depth.
 	 */
-	if (atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) {
+	if ((atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth) &&
+	    !expedite) {
 		ret = -EBUSY;
 		goto out_fail;
 	}
 
-	lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp);
+	lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp, expedite);
 	if (lpfc_ncmd == NULL) {
 		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
 				 "6065 driver's buffer pool is empty, "
@@ -1991,6 +2006,8 @@ lpfc_repost_nvme_sgl_list(struct lpfc_hba *phba)
 	spin_lock(&phba->nvme_buf_list_put_lock);
 	list_splice_init(&phba->lpfc_nvme_buf_list_get, &post_nblist);
 	list_splice(&phba->lpfc_nvme_buf_list_put, &post_nblist);
+	phba->get_nvme_bufs = 0;
+	phba->put_nvme_bufs = 0;
 	spin_unlock(&phba->nvme_buf_list_put_lock);
 	spin_unlock_irq(&phba->nvme_buf_list_get_lock);
 
@@ -2127,6 +2144,20 @@ lpfc_new_nvme_buf(struct lpfc_vport *vport, int num_to_alloc)
 	return num_posted;
 }
 
+static inline struct lpfc_nvme_buf *
+lpfc_nvme_buf(struct lpfc_hba *phba)
+{
+	struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
+
+	list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
+				 &phba->lpfc_nvme_buf_list_get, list) {
+		list_del_init(&lpfc_ncmd->list);
+		phba->get_nvme_bufs--;
+		return lpfc_ncmd;
+	}
+	return NULL;
+}
+
 /**
  * lpfc_get_nvme_buf - Get a nvme buffer from lpfc_nvme_buf_list of the HBA
  * @phba: The HBA for which this call is being executed.
@@ -2139,35 +2170,27 @@ lpfc_new_nvme_buf(struct lpfc_vport *vport, int num_to_alloc)
  *   Pointer to lpfc_nvme_buf - Success
  **/
 static struct lpfc_nvme_buf *
-lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp)
+lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp,
+		  int expedite)
 {
-	struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
+	struct lpfc_nvme_buf *lpfc_ncmd = NULL;
 	unsigned long iflag = 0;
-	int found = 0;
 
 	spin_lock_irqsave(&phba->nvme_buf_list_get_lock, iflag);
-	list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
-				 &phba->lpfc_nvme_buf_list_get, list) {
-		list_del_init(&lpfc_ncmd->list);
-		found = 1;
-		break;
-	}
-	if (!found) {
+	if (phba->get_nvme_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite)
+		lpfc_ncmd = lpfc_nvme_buf(phba);
+	if (!lpfc_ncmd) {
 		spin_lock(&phba->nvme_buf_list_put_lock);
 		list_splice(&phba->lpfc_nvme_buf_list_put,
 			    &phba->lpfc_nvme_buf_list_get);
+		phba->get_nvme_bufs += phba->put_nvme_bufs;
 		INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put);
+		phba->put_nvme_bufs = 0;
 		spin_unlock(&phba->nvme_buf_list_put_lock);
-		list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next,
-					 &phba->lpfc_nvme_buf_list_get, list) {
-			list_del_init(&lpfc_ncmd->list);
-			found = 1;
-			break;
-		}
+		if (phba->get_nvme_bufs > LPFC_NVME_EXPEDITE_XRICNT || expedite)
+			lpfc_ncmd = lpfc_nvme_buf(phba);
 	}
 	spin_unlock_irqrestore(&phba->nvme_buf_list_get_lock, iflag);
-	if (!found)
-		return NULL;
 	return  lpfc_ncmd;
 }
 
@@ -2205,6 +2228,7 @@ lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_nvme_buf *lpfc_ncmd)
 		lpfc_ncmd->cur_iocbq.iocb_flag = LPFC_IO_NVME;
 		spin_lock_irqsave(&phba->nvme_buf_list_put_lock, iflag);
 		list_add_tail(&lpfc_ncmd->list, &phba->lpfc_nvme_buf_list_put);
+		phba->put_nvme_bufs++;
 		spin_unlock_irqrestore(&phba->nvme_buf_list_put_lock, iflag);
 	}
 }

drivers/scsi/lpfc/lpfc_nvme.h

@@ -28,6 +28,7 @@
 #define LPFC_NVME_ERSP_LEN		0x20
 
 #define LPFC_NVME_WAIT_TMO              10
+#define LPFC_NVME_EXPEDITE_XRICNT	8
 
 struct lpfc_nvme_qhandle {
 	uint32_t index;		/* WQ index to use */