[PATCH 09/17] lpfc: NVME Initiator: bind to nvme_fc api

James Smart jsmart2021 at gmail.com
Tue Jan 17 17:20:52 PST 2017


NVME Initiator: Tie in to NVME Fabrics nvme_fc LLDD initiator api

Adds the routines to:
- register and deregister the FC port as a nvme-fc initiator localport
- register and deregister remote FC ports as a nvme-fc remoteport.
- binding of nvme queues to adapter WQs
- send/perform NVME LS's
- send/perform NVME FCP initiator io operations

Signed-off-by: Dick Kennedy <dick.kennedy at broadcom.com>
Signed-off-by: James Smart <james.smart at broadcom.com>
---
 drivers/scsi/lpfc/Makefile         |    7 +-
 drivers/scsi/lpfc/lpfc.h           |    5 +
 drivers/scsi/lpfc/lpfc_crtn.h      |    9 +
 drivers/scsi/lpfc/lpfc_ct.c        |    2 +-
 drivers/scsi/lpfc/lpfc_els.c       |    4 +-
 drivers/scsi/lpfc/lpfc_hbadisc.c   |   15 +-
 drivers/scsi/lpfc/lpfc_init.c      |   20 +-
 drivers/scsi/lpfc/lpfc_nportdisc.c |    4 +-
 drivers/scsi/lpfc/lpfc_nvme.c      | 2249 ++++++++++++++++++++++++++++++++++++
 drivers/scsi/lpfc/lpfc_nvme.h      |    6 +
 drivers/scsi/lpfc/lpfc_sli.c       |   21 +-
 drivers/scsi/lpfc/lpfc_sli4.h      |    1 +
 drivers/scsi/lpfc/lpfc_vport.c     |   16 +-
 13 files changed, 2339 insertions(+), 20 deletions(-)
 create mode 100644 drivers/scsi/lpfc/lpfc_nvme.c

diff --git a/drivers/scsi/lpfc/Makefile b/drivers/scsi/lpfc/Makefile
index e2516ba..cd7e1fc 100644
--- a/drivers/scsi/lpfc/Makefile
+++ b/drivers/scsi/lpfc/Makefile
@@ -28,6 +28,7 @@ endif
 
 obj-$(CONFIG_SCSI_LPFC) := lpfc.o
 
-lpfc-objs := lpfc_mem.o lpfc_sli.o lpfc_ct.o lpfc_els.o lpfc_hbadisc.o	\
-	lpfc_init.o lpfc_mbox.o lpfc_nportdisc.o lpfc_scsi.o lpfc_attr.o \
-	lpfc_vport.o lpfc_debugfs.o lpfc_bsg.o
+lpfc-objs := lpfc_mem.o lpfc_sli.o lpfc_ct.o lpfc_els.o \
+	lpfc_hbadisc.o	lpfc_init.o lpfc_mbox.o lpfc_nportdisc.o   \
+	lpfc_scsi.o lpfc_attr.o lpfc_vport.o lpfc_debugfs.o lpfc_bsg.o \
+	lpfc_nvme.o
diff --git a/drivers/scsi/lpfc/lpfc.h b/drivers/scsi/lpfc/lpfc.h
index 72b4614..1dae886 100644
--- a/drivers/scsi/lpfc/lpfc.h
+++ b/drivers/scsi/lpfc/lpfc.h
@@ -123,6 +123,11 @@ struct perf_prof {
 	uint16_t wqidx[40];
 };
 
+/*
+ * Provide for FC4 TYPE x28 - NVME.  The
+ * bit mask for FCP and NVME is 0x8 identically
+ * because they are 32 bit positions distance.
+ */
 #define LPFC_FC4_TYPE_BITMASK	0x00000100
 
 /* Provide DMA memory definitions the driver uses per port instance. */
diff --git a/drivers/scsi/lpfc/lpfc_crtn.h b/drivers/scsi/lpfc/lpfc_crtn.h
index 4b7decb..8592d6d 100644
--- a/drivers/scsi/lpfc/lpfc_crtn.h
+++ b/drivers/scsi/lpfc/lpfc_crtn.h
@@ -531,4 +531,13 @@ struct lpfc_scsi_buf *lpfc_get_scsi_buf(struct lpfc_hba *phba,
 					struct lpfc_nodelist *ndlp);
 
 /* NVME interfaces. */
+void lpfc_nvme_unregister_port(struct lpfc_nodelist *ndlp);
+int lpfc_nvme_register_port(struct lpfc_vport *vport,
+			struct lpfc_nodelist *ndlp);
+int lpfc_nvme_create_localport(struct lpfc_vport *vport);
+void lpfc_nvme_destroy_localport(struct lpfc_vport *vport);
+void lpfc_nvme_update_localport(struct lpfc_vport *vport);
 void lpfc_nvme_mod_param_dep(struct lpfc_hba *phba);
+void lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba,
+				struct lpfc_iocbq *cmdiocb,
+				struct lpfc_wcqe_complete *abts_cmpl);
diff --git a/drivers/scsi/lpfc/lpfc_ct.c b/drivers/scsi/lpfc/lpfc_ct.c
index 23d6a3c..ed285fc 100644
--- a/drivers/scsi/lpfc/lpfc_ct.c
+++ b/drivers/scsi/lpfc/lpfc_ct.c
@@ -1412,7 +1412,7 @@ lpfc_ns_cmd(struct lpfc_vport *vport, int cmdcode,
 		if (((phba->cfg_enable_fc4_type == LPFC_ENABLE_BOTH) ||
 		     (phba->cfg_enable_fc4_type == LPFC_ENABLE_NVME)) &&
 		    (context == FC_TYPE_NVME)) {
-			/* todo: init: revise localport nvme attributes */
+			lpfc_nvme_update_localport(vport);
 			CtReq->un.rff.type_code = context;
 
 		} else if (((phba->cfg_enable_fc4_type == LPFC_ENABLE_BOTH) ||
diff --git a/drivers/scsi/lpfc/lpfc_els.c b/drivers/scsi/lpfc/lpfc_els.c
index 3d915ee..6d2b163 100644
--- a/drivers/scsi/lpfc/lpfc_els.c
+++ b/drivers/scsi/lpfc/lpfc_els.c
@@ -2625,7 +2625,9 @@ lpfc_cmpl_els_logo(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
 		!(vport->fc_flag & FC_PT2PT_PLOGI)) {
 		phba->pport->fc_myDID = 0;
 
-		/* todo: init: revise localport nvme attributes */
+		if ((phba->cfg_enable_fc4_type == LPFC_ENABLE_BOTH) ||
+		    (phba->cfg_enable_fc4_type == LPFC_ENABLE_NVME))
+			lpfc_nvme_update_localport(phba->pport);
 
 		mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
 		if (mbox) {
diff --git a/drivers/scsi/lpfc/lpfc_hbadisc.c b/drivers/scsi/lpfc/lpfc_hbadisc.c
index 55a16f2..291a4cb 100644
--- a/drivers/scsi/lpfc/lpfc_hbadisc.c
+++ b/drivers/scsi/lpfc/lpfc_hbadisc.c
@@ -912,7 +912,9 @@ lpfc_linkdown(struct lpfc_hba *phba)
 
 			vports[i]->fc_myDID = 0;
 
-			/* todo: init: revise localport nvme attributes */
+			if ((phba->cfg_enable_fc4_type == LPFC_ENABLE_BOTH) ||
+			    (phba->cfg_enable_fc4_type == LPFC_ENABLE_NVME))
+				lpfc_nvme_update_localport(vports[i]);
 		}
 	}
 	lpfc_destroy_vport_work_array(phba, vports);
@@ -3585,7 +3587,9 @@ lpfc_mbx_cmpl_reg_vpi(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
 		spin_unlock_irq(shost->host_lock);
 		vport->fc_myDID = 0;
 
-		/* todo: init: revise localport nvme attributes */
+		if ((phba->cfg_enable_fc4_type == LPFC_ENABLE_BOTH) ||
+		    (phba->cfg_enable_fc4_type == LPFC_ENABLE_NVME))
+			lpfc_nvme_update_localport(vport);
 		goto out;
 	}
 
@@ -3955,7 +3959,8 @@ lpfc_mbx_cmpl_ns_reg_login(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
 
 		if ((phba->cfg_enable_fc4_type == LPFC_ENABLE_BOTH) ||
 		    (phba->cfg_enable_fc4_type == LPFC_ENABLE_NVME))
-			lpfc_ns_cmd(vport, SLI_CTNS_RFF_ID, 0, FC_TYPE_NVME);
+			lpfc_ns_cmd(vport, SLI_CTNS_RFF_ID, 0,
+				    FC_TYPE_NVME);
 
 		/* Issue SCR just before NameServer GID_FT Query */
 		lpfc_issue_els_scr(vport, SCR_DID, 0);
@@ -4149,7 +4154,7 @@ lpfc_nlp_state_cleanup(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
 		    ((ndlp->nlp_fc4_type & NLP_FC4_NVME) ||
 		    (ndlp->nlp_DID == Fabric_DID))) {
 			vport->phba->nport_event_cnt++;
-			/* todo: init: unregister rport from nvme */
+			lpfc_nvme_unregister_port(ndlp);
 		}
 	}
 
@@ -4174,7 +4179,7 @@ lpfc_nlp_state_cleanup(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
 				 * the register.
 				 */
 				vport->phba->nport_event_cnt++;
-				/* todo: init: register rport with nvme */
+				lpfc_nvme_register_port(vport, ndlp);
 			}
 		}
 	}
diff --git a/drivers/scsi/lpfc/lpfc_init.c b/drivers/scsi/lpfc/lpfc_init.c
index ea12eca..7ff04b0 100644
--- a/drivers/scsi/lpfc/lpfc_init.c
+++ b/drivers/scsi/lpfc/lpfc_init.c
@@ -11157,7 +11157,23 @@ lpfc_pci_probe_one_s4(struct pci_dev *pdev, const struct pci_device_id *pid)
 	/* Perform post initialization setup */
 	lpfc_post_init_setup(phba);
 
-	/* todo: init: register port with nvme */
+	/* NVME support in FW earlier in the driver load corrects the
+	 * FC4 type making a check for nvme_support unnecessary.
+	 */
+	if ((phba->nvmet_support == 0) &&
+	    (phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME)) {
+		/* Create NVME binding with nvme_fc_transport. This
+		 * ensures the vport is initialized.
+		 */
+		error = lpfc_nvme_create_localport(vport);
+		if (error) {
+			lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
+					"6004 NVME registration failed, "
+					"error x%x\n",
+					error);
+			goto out_disable_intr;
+		}
+	}
 
 	/* check for firmware upgrade or downgrade */
 	if (phba->cfg_request_firmware_upgrade)
@@ -11234,7 +11250,7 @@ lpfc_pci_remove_one_s4(struct pci_dev *pdev)
 	 * is destroyed after to ensure all rports are io-disabled.
 	 */
 	lpfc_cleanup(vport);
-	/* todo: init: unregister port with nvme */
+	lpfc_nvme_destroy_localport(vport);
 
 	/*
 	 * Bring down the SLI Layer. This step disables all interrupts,
diff --git a/drivers/scsi/lpfc/lpfc_nportdisc.c b/drivers/scsi/lpfc/lpfc_nportdisc.c
index 09bd7e2..8aa12ba 100644
--- a/drivers/scsi/lpfc/lpfc_nportdisc.c
+++ b/drivers/scsi/lpfc/lpfc_nportdisc.c
@@ -1680,9 +1680,7 @@ lpfc_cmpl_reglogin_reglogin_issue(struct lpfc_vport *vport,
 			     (phba->cfg_enable_fc4_type == LPFC_ENABLE_NVME)) {
 				ndlp->nlp_fc4_type |= NLP_FC4_NVME;
 				/* We need to update the localport also */
-				/* todo: init: revise localport nvme
-				 * attributes
-				 */
+				lpfc_nvme_update_localport(vport);
 			}
 
 		} else if (ndlp->nlp_fc4_type == 0) {
diff --git a/drivers/scsi/lpfc/lpfc_nvme.c b/drivers/scsi/lpfc/lpfc_nvme.c
new file mode 100644
index 0000000..defc7f2
--- /dev/null
+++ b/drivers/scsi/lpfc/lpfc_nvme.c
@@ -0,0 +1,2249 @@
+/*******************************************************************
+ * This file is part of the Emulex Linux Device Driver for         *
+ * Fibre Channel Host Bus Adapters.                                *
+ * Copyright (C) 2004-2016 Emulex.  All rights reserved.           *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ * Portions Copyright (C) 2004-2005 Christoph Hellwig              *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ ********************************************************************/
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/interrupt.h>
+#include <linux/delay.h>
+#include <asm/unaligned.h>
+#include <linux/crc-t10dif.h>
+#include <net/checksum.h>
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_device.h>
+#include <scsi/scsi_eh.h>
+#include <scsi/scsi_host.h>
+#include <scsi/scsi_tcq.h>
+#include <scsi/scsi_transport_fc.h>
+#include <scsi/fc/fc_fs.h>
+
+#include <linux/nvme.h>
+#include <linux/nvme-fc-driver.h>
+#include <linux/nvme-fc.h>
+#include "lpfc_version.h"
+#include "lpfc_hw4.h"
+#include "lpfc_hw.h"
+#include "lpfc_sli.h"
+#include "lpfc_sli4.h"
+#include "lpfc_nl.h"
+#include "lpfc_disc.h"
+#include "lpfc.h"
+#include "lpfc_nvme.h"
+#include "lpfc_scsi.h"
+#include "lpfc_logmsg.h"
+#include "lpfc_crtn.h"
+#include "lpfc_vport.h"
+
+/* NVME initiator-based functions */
+
+static struct lpfc_nvme_buf *
+lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp);
+
+static void
+lpfc_release_nvme_buf(struct lpfc_hba *, struct lpfc_nvme_buf *);
+
+
+/**
+ * lpfc_nvme_create_queue -
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
+ * @handle: An opaque driver handle used in follow-up calls.
+ *
+ * Driver registers this routine to preallocate and initialize any
+ * internal data structures to bind the @qidx to its internal IO queues.
+ * A hardware queue maps (qidx) to a specific driver MSI-X vector/EQ/CQ/WQ.
+ *
+ * Return value :
+ *   0 - Success
+ *   -EINVAL - Unsupported input value.
+ *   -ENOMEM - Could not alloc necessary memory
+ **/
+static int
+lpfc_nvme_create_queue(struct nvme_fc_local_port *pnvme_lport,
+		       unsigned int qidx, u16 qsize,
+		       void **handle)
+{
+	struct lpfc_nvme_lport *lport;
+	struct lpfc_vport *vport;
+	struct lpfc_nvme_qhandle *qhandle;
+	char *str;
+
+	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
+	vport = lport->vport;
+	qhandle = kzalloc(sizeof(struct lpfc_nvme_qhandle), GFP_KERNEL);
+	if (qhandle == NULL)
+		return -ENOMEM;
+
+	qhandle->cpu_id = smp_processor_id();
+	qhandle->qidx = qidx;
+	/*
+	 * NVME qidx == 0 is the admin queue, so both admin queue
+	 * and first IO queue will use MSI-X vector and associated
+	 * EQ/CQ/WQ at index 0. After that they are sequentially assigned.
+	 */
+	if (qidx) {
+		str = "IO ";  /* IO queue */
+		qhandle->index = ((qidx - 1) %
+			vport->phba->cfg_nvme_io_channel);
+	} else {
+		str = "ADM";  /* Admin queue */
+		qhandle->index = qidx;
+	}
+
+	lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+			 "6073 Binding %s HdwQueue %d  (cpu %d) to "
+			 "io_channel %d qhandle %p\n", str,
+			 qidx, qhandle->cpu_id, qhandle->index, qhandle);
+	*handle = (void *)qhandle;
+	return 0;
+}
+
+/**
+ * lpfc_nvme_delete_queue -
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @qidx: An cpu index used to affinitize IO queues and MSIX vectors.
+ * @handle: An opaque driver handle from lpfc_nvme_create_queue
+ *
+ * Driver registers this routine to free
+ * any internal data structures to bind the @qidx to its internal
+ * IO queues.
+ *
+ * Return value :
+ *   0 - Success
+ *   TODO:  What are the failure codes.
+ **/
+static void
+lpfc_nvme_delete_queue(struct nvme_fc_local_port *pnvme_lport,
+		       unsigned int qidx,
+		       void *handle)
+{
+	struct lpfc_nvme_lport *lport;
+	struct lpfc_vport *vport;
+
+	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
+	vport = lport->vport;
+
+	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
+			"6001 ENTER.  lpfc_pnvme %p, qidx x%xi qhandle %p\n",
+			lport, qidx, handle);
+	kfree(handle);
+}
+
+static void
+lpfc_nvme_localport_delete(struct nvme_fc_local_port *localport)
+{
+	struct lpfc_nvme_lport *lport = localport->private;
+
+	/* release any threads waiting for the unreg to complete */
+	complete(&lport->lport_unreg_done);
+}
+
+static void
+lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport)
+{
+	struct lpfc_nvme_rport *rport = remoteport->private;
+
+	/* release any threads waiting for the unreg to complete */
+	complete(&rport->rport_unreg_done);
+}
+
+static void
+lpfc_nvme_cmpl_gen_req(struct lpfc_hba *phba, struct lpfc_iocbq *cmdwqe,
+		       struct lpfc_wcqe_complete *wcqe)
+{
+	struct lpfc_vport *vport = cmdwqe->vport;
+	uint32_t status;
+	struct nvmefc_ls_req *pnvme_lsreq;
+	struct lpfc_dmabuf *buf_ptr;
+	struct lpfc_nodelist *ndlp;
+
+	vport->phba->fc4NvmeLsCmpls++;
+
+	pnvme_lsreq = (struct nvmefc_ls_req *)cmdwqe->context2;
+	status = bf_get(lpfc_wcqe_c_status, wcqe) & LPFC_IOCB_STATUS_MASK;
+	ndlp = (struct lpfc_nodelist *)cmdwqe->context1;
+	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
+			 "6047 nvme cmpl Enter "
+			 "Data %p DID %x Xri: %x status %x cmd:%p lsreg:%p "
+			 "bmp:%p ndlp:%p\n",
+			 pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
+			 cmdwqe->sli4_xritag, status,
+			 cmdwqe, pnvme_lsreq, cmdwqe->context3, ndlp);
+
+	if (cmdwqe->context3) {
+		buf_ptr = (struct lpfc_dmabuf *)cmdwqe->context3;
+		lpfc_mbuf_free(phba, buf_ptr->virt, buf_ptr->phys);
+		kfree(buf_ptr);
+		cmdwqe->context3 = NULL;
+	}
+	if (pnvme_lsreq->done)
+		pnvme_lsreq->done(pnvme_lsreq, status);
+	else
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
+				 "6046 nvme cmpl without done call back? "
+				 "Data %p DID %x Xri: %x status %x\n",
+				pnvme_lsreq, ndlp ? ndlp->nlp_DID : 0,
+				cmdwqe->sli4_xritag, status);
+	if (ndlp) {
+		lpfc_nlp_put(ndlp);
+		cmdwqe->context1 = NULL;
+	}
+	lpfc_sli_release_iocbq(phba, cmdwqe);
+}
+
+static int
+lpfc_nvme_gen_req(struct lpfc_vport *vport, struct lpfc_dmabuf *bmp,
+		  struct lpfc_dmabuf *inp,
+		 struct nvmefc_ls_req *pnvme_lsreq,
+	     void (*cmpl)(struct lpfc_hba *, struct lpfc_iocbq *,
+			   struct lpfc_wcqe_complete *),
+	     struct lpfc_nodelist *ndlp, uint32_t num_entry,
+	     uint32_t tmo, uint8_t retry)
+{
+	struct lpfc_hba  *phba = vport->phba;
+	union lpfc_wqe *wqe;
+	struct lpfc_iocbq *genwqe;
+	struct ulp_bde64 *bpl;
+	struct ulp_bde64 bde;
+	int i, rc, xmit_len, first_len;
+
+	/* Allocate buffer for  command WQE */
+	genwqe = lpfc_sli_get_iocbq(phba);
+	if (genwqe == NULL)
+		return 1;
+
+	wqe = &genwqe->wqe;
+	memset(wqe, 0, sizeof(union lpfc_wqe));
+
+	genwqe->context3 = (uint8_t *)bmp;
+	genwqe->iocb_flag |= LPFC_IO_NVME_LS;
+
+	/* Save for completion so we can release these resources */
+	genwqe->context1 = lpfc_nlp_get(ndlp);
+	genwqe->context2 = (uint8_t *)pnvme_lsreq;
+	/* Fill in payload, bp points to frame payload */
+
+	if (!tmo)
+		/* FC spec states we need 3 * ratov for CT requests */
+		tmo = (3 * phba->fc_ratov);
+
+	/* For this command calculate the xmit length of the request bde. */
+	xmit_len = 0;
+	first_len = 0;
+	bpl = (struct ulp_bde64 *)bmp->virt;
+	for (i = 0; i < num_entry; i++) {
+		bde.tus.w = bpl[i].tus.w;
+		if (bde.tus.f.bdeFlags != BUFF_TYPE_BDE_64)
+			break;
+		xmit_len += bde.tus.f.bdeSize;
+		if (i == 0)
+			first_len = xmit_len;
+	}
+
+	genwqe->rsvd2 = num_entry;
+	genwqe->hba_wqidx = 0;
+
+	/* Words 0 - 2 */
+	wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
+	wqe->generic.bde.tus.f.bdeSize = first_len;
+	wqe->generic.bde.addrLow = bpl[0].addrLow;
+	wqe->generic.bde.addrHigh = bpl[0].addrHigh;
+
+	/* Word 3 */
+	wqe->gen_req.request_payload_len = first_len;
+
+	/* Word 4 */
+
+	/* Word 5 */
+	bf_set(wqe_dfctl, &wqe->gen_req.wge_ctl, 0);
+	bf_set(wqe_si, &wqe->gen_req.wge_ctl, 1);
+	bf_set(wqe_la, &wqe->gen_req.wge_ctl, 1);
+	bf_set(wqe_rctl, &wqe->gen_req.wge_ctl, FC_RCTL_DD_UNSOL_CTL);
+	bf_set(wqe_type, &wqe->gen_req.wge_ctl, FC_TYPE_NVME);
+
+	/* Word 6 */
+	bf_set(wqe_ctxt_tag, &wqe->gen_req.wqe_com,
+	       phba->sli4_hba.rpi_ids[ndlp->nlp_rpi]);
+	bf_set(wqe_xri_tag, &wqe->gen_req.wqe_com, genwqe->sli4_xritag);
+
+	/* Word 7 */
+	bf_set(wqe_tmo, &wqe->gen_req.wqe_com, (vport->phba->fc_ratov-1));
+	bf_set(wqe_class, &wqe->gen_req.wqe_com, CLASS3);
+	bf_set(wqe_cmnd, &wqe->gen_req.wqe_com, CMD_GEN_REQUEST64_WQE);
+	bf_set(wqe_ct, &wqe->gen_req.wqe_com, SLI4_CT_RPI);
+
+	/* Word 8 */
+	wqe->gen_req.wqe_com.abort_tag = genwqe->iotag;
+
+	/* Word 9 */
+	bf_set(wqe_reqtag, &wqe->gen_req.wqe_com, genwqe->iotag);
+
+	/* Word 10 */
+	bf_set(wqe_dbde, &wqe->gen_req.wqe_com, 1);
+	bf_set(wqe_iod, &wqe->gen_req.wqe_com, LPFC_WQE_IOD_READ);
+	bf_set(wqe_qosd, &wqe->gen_req.wqe_com, 1);
+	bf_set(wqe_lenloc, &wqe->gen_req.wqe_com, LPFC_WQE_LENLOC_NONE);
+	bf_set(wqe_ebde_cnt, &wqe->gen_req.wqe_com, 0);
+
+	/* Word 11 */
+	bf_set(wqe_cqid, &wqe->gen_req.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
+	bf_set(wqe_cmd_type, &wqe->gen_req.wqe_com, OTHER_COMMAND);
+
+
+	/* Issue GEN REQ WQE for NPORT <did> */
+	lpfc_printf_vlog(vport, KERN_INFO, LOG_ELS,
+			 "6050 Issue GEN REQ WQE to NPORT x%x "
+			 "Data: x%x x%x wq:%p lsreq:%p bmp:%p xmit:%d 1st:%d\n",
+			 ndlp->nlp_DID, genwqe->iotag,
+			 vport->port_state,
+			genwqe, pnvme_lsreq, bmp, xmit_len, first_len);
+	genwqe->wqe_cmpl = cmpl;
+	genwqe->iocb_cmpl = NULL;
+	genwqe->drvrTimeout = tmo + LPFC_DRVR_TIMEOUT;
+	genwqe->vport = vport;
+	genwqe->retry = retry;
+
+	rc = lpfc_sli4_issue_wqe(phba, LPFC_ELS_RING, genwqe);
+	if (rc == WQE_ERROR) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_ELS,
+				 "6045 Issue GEN REQ WQE to NPORT x%x "
+				 "Data: x%x x%x\n",
+				 ndlp->nlp_DID, genwqe->iotag,
+				 vport->port_state);
+		lpfc_sli_release_iocbq(phba, genwqe);
+		return 1;
+	}
+	return 0;
+}
+
+/**
+ * lpfc_nvme_ls_req - Issue an Link Service request
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @lpfc_nvme_lport: Pointer to the driver's local port data
+ * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
+ *
+ * Driver registers this routine to handle any link service request
+ * from the nvme_fc transport to a remote nvme-aware port.
+ *
+ * Return value :
+ *   0 - Success
+ *   TODO: What are the failure codes.
+ **/
+static int
+lpfc_nvme_ls_req(struct nvme_fc_local_port *pnvme_lport,
+		 struct nvme_fc_remote_port *pnvme_rport,
+		 struct nvmefc_ls_req *pnvme_lsreq)
+{
+	int ret = 0;
+	struct lpfc_nvme_lport *lport;
+	struct lpfc_vport *vport;
+	struct lpfc_nodelist *ndlp;
+	struct ulp_bde64 *bpl;
+	struct lpfc_dmabuf *bmp;
+
+	/* there are two dma buf in the request, actually there is one and
+	 * the second one is just the start address + cmd size.
+	 * Before calling lpfc_nvme_gen_req these buffers need to be wrapped
+	 * in a lpfc_dmabuf struct. When freeing we just free the wrapper
+	 * because the nvem layer owns the data bufs.
+	 * We do not have to break these packets open, we don't care what is in
+	 * them. And we do not have to look at the resonse data, we only care
+	 * that we got a response. All of the caring is going to happen in the
+	 * nvme-fc layer.
+	 */
+
+	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
+	vport = lport->vport;
+
+	ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
+	if (!ndlp) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
+				 "6043 Could not find node for DID %x\n",
+				 pnvme_rport->port_id);
+		return 1;
+	}
+	bmp = kmalloc(sizeof(struct lpfc_dmabuf), GFP_KERNEL);
+	if (!bmp) {
+
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
+				 "6044 Could not find node for DID %x\n",
+				 pnvme_rport->port_id);
+		return 2;
+	}
+	INIT_LIST_HEAD(&bmp->list);
+	bmp->virt = lpfc_mbuf_alloc(vport->phba, MEM_PRI, &(bmp->phys));
+	if (!bmp->virt) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
+				 "6042 Could not find node for DID %x\n",
+				 pnvme_rport->port_id);
+		kfree(bmp);
+		return 3;
+	}
+	bpl = (struct ulp_bde64 *)bmp->virt;
+	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rqstdma));
+	bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rqstdma));
+	bpl->tus.f.bdeFlags = 0;
+	bpl->tus.f.bdeSize = pnvme_lsreq->rqstlen;
+	bpl->tus.w = le32_to_cpu(bpl->tus.w);
+	bpl++;
+
+	bpl->addrHigh = le32_to_cpu(putPaddrHigh(pnvme_lsreq->rspdma));
+	bpl->addrLow = le32_to_cpu(putPaddrLow(pnvme_lsreq->rspdma));
+	bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
+	bpl->tus.f.bdeSize = pnvme_lsreq->rsplen;
+	bpl->tus.w = le32_to_cpu(bpl->tus.w);
+
+	/* Expand print to include key fields. */
+	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
+			 "6051 ENTER.  lport %p, rport %p lsreq%p rqstlen:%d "
+			 "rsplen:%d %llux %llux\n",
+			 pnvme_lport, pnvme_rport,
+			 pnvme_lsreq, pnvme_lsreq->rqstlen,
+			 pnvme_lsreq->rsplen, pnvme_lsreq->rqstdma,
+			 pnvme_lsreq->rspdma);
+
+	vport->phba->fc4NvmeLsRequests++;
+
+	/* Hardcode the wait to 30 seconds.  Connections are failing otherwise.
+	 * This code allows it all to work.
+	 */
+	ret = lpfc_nvme_gen_req(vport, bmp, pnvme_lsreq->rqstaddr,
+				pnvme_lsreq, lpfc_nvme_cmpl_gen_req,
+				ndlp, 2, 30, 0);
+	if (ret != WQE_SUCCESS) {
+		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
+				 "6052 EXIT. issue ls wqe failed lport %p, "
+				 "rport %p lsreq%p Status %x DID %x\n",
+				 pnvme_lport, pnvme_rport, pnvme_lsreq,
+				 ret, ndlp->nlp_DID);
+		lpfc_mbuf_free(vport->phba, bmp->virt, bmp->phys);
+		kfree(bmp);
+		return ret;
+	}
+
+	/* Stub in routine and return 0 for now. */
+	return ret;
+}
+
+/**
+ * lpfc_nvme_ls_abort - Issue an Link Service request
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @lpfc_nvme_lport: Pointer to the driver's local port data
+ * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
+ *
+ * Driver registers this routine to handle any link service request
+ * from the nvme_fc transport to a remote nvme-aware port.
+ *
+ * Return value :
+ *   0 - Success
+ *   TODO: What are the failure codes.
+ **/
+static void
+lpfc_nvme_ls_abort(struct nvme_fc_local_port *pnvme_lport,
+		   struct nvme_fc_remote_port *pnvme_rport,
+		   struct nvmefc_ls_req *pnvme_lsreq)
+{
+	struct lpfc_nvme_lport *lport;
+	struct lpfc_vport *vport;
+	struct lpfc_hba *phba;
+	struct lpfc_nodelist *ndlp;
+	LIST_HEAD(abort_list);
+	struct lpfc_sli_ring *pring;
+	struct lpfc_iocbq *wqe, *next_wqe;
+
+	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
+	vport = lport->vport;
+	phba = vport->phba;
+
+	ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
+	if (!ndlp) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
+				 "6049 Could not find node for DID %x\n",
+				 pnvme_rport->port_id);
+		return;
+	}
+
+	/* Expand print to include key fields. */
+	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_ABTS,
+			 "6040 ENTER.  lport %p, rport %p lsreq %p rqstlen:%d "
+			 "rsplen:%d %llux %llux\n",
+			 pnvme_lport, pnvme_rport,
+			 pnvme_lsreq, pnvme_lsreq->rqstlen,
+			 pnvme_lsreq->rsplen, pnvme_lsreq->rqstdma,
+			 pnvme_lsreq->rspdma);
+
+	/*
+	 * Lock the ELS ring txcmplq and build a local list of all ELS IOs
+	 * that need an ABTS.  The IOs need to stay on the txcmplq so that
+	 * the abort operation completes them successfully.
+	 */
+	pring = phba->sli4_hba.nvmels_wq->pring;
+	spin_lock_irq(&phba->hbalock);
+	spin_lock(&pring->ring_lock);
+	list_for_each_entry_safe(wqe, next_wqe, &pring->txcmplq, list) {
+		/* Add to abort_list on on NDLP match. */
+		if (lpfc_check_sli_ndlp(phba, pring, wqe, ndlp)) {
+			wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
+			list_add_tail(&wqe->dlist, &abort_list);
+		}
+	}
+	spin_unlock(&pring->ring_lock);
+	spin_unlock_irq(&phba->hbalock);
+
+	/* Abort the targeted IOs and remove them from the abort list. */
+	list_for_each_entry_safe(wqe, next_wqe, &abort_list, dlist) {
+		spin_lock_irq(&phba->hbalock);
+		list_del_init(&wqe->dlist);
+		lpfc_sli_issue_abort_iotag(phba, pring, wqe);
+		spin_unlock_irq(&phba->hbalock);
+	}
+}
+
+/* Fix up the existing sgls for NVME IO. */
+static void
+lpfc_nvme_adj_fcp_sgls(struct lpfc_vport *vport,
+		       struct lpfc_nvme_buf *lpfc_ncmd,
+		       struct nvmefc_fcp_req *nCmd)
+{
+	struct sli4_sge *sgl;
+	union lpfc_wqe128 *wqe;
+	uint32_t *wptr, *dptr;
+
+	/*
+	 * Adjust the FCP_CMD and FCP_RSP DMA data and sge_len to
+	 * match NVME.  NVME sends 96 bytes. Also, use the
+	 * nvme commands command and response dma addresses
+	 * rather than the virtual memory to ease the restore
+	 * operation.
+	 */
+	sgl = lpfc_ncmd->nvme_sgl;
+	sgl->sge_len = cpu_to_le32(nCmd->cmdlen);
+
+	sgl++;
+
+	/* Setup the physical region for the FCP RSP */
+	sgl->addr_hi = cpu_to_le32(putPaddrHigh(nCmd->rspdma));
+	sgl->addr_lo = cpu_to_le32(putPaddrLow(nCmd->rspdma));
+	sgl->word2 = le32_to_cpu(sgl->word2);
+	if (nCmd->sg_cnt)
+		bf_set(lpfc_sli4_sge_last, sgl, 0);
+	else
+		bf_set(lpfc_sli4_sge_last, sgl, 1);
+	sgl->word2 = cpu_to_le32(sgl->word2);
+	sgl->sge_len = cpu_to_le32(nCmd->rsplen);
+
+	/*
+	 * Get a local pointer to the built-in wqe and correct
+	 * the cmd size to match NVME's 96 bytes and fix
+	 * the dma address.
+	 */
+
+	/* 128 byte wqe support here */
+	wqe = (union lpfc_wqe128 *)&lpfc_ncmd->cur_iocbq.wqe;
+
+	/* Word 0-2 - NVME CMND IU (embedded payload) */
+	wqe->generic.bde.tus.f.bdeFlags = BUFF_TYPE_BDE_IMMED;
+	wqe->generic.bde.tus.f.bdeSize = 60;
+	wqe->generic.bde.addrHigh = 0;
+	wqe->generic.bde.addrLow =  64;  /* Word 16 */
+
+	/* Word 3 */
+	bf_set(payload_offset_len, &wqe->fcp_icmd,
+	       (nCmd->rsplen + nCmd->cmdlen));
+
+	/* Word 10 */
+	bf_set(wqe_nvme, &wqe->fcp_icmd.wqe_com, 1);
+	bf_set(wqe_wqes, &wqe->fcp_icmd.wqe_com, 1);
+
+	/*
+	 * Embed the payload in the last half of the WQE
+	 * WQE words 16-30 get the NVME CMD IU payload
+	 *
+	 * WQE Word 16 is already setup with flags
+	 * WQE words 17-19 get payload Words 2-4
+	 * WQE words 20-21 get payload Words 6-7
+	 * WQE words 22-29 get payload Words 16-23
+	 */
+	wptr = &wqe->words[17];  /* WQE ptr */
+	dptr = (uint32_t *)nCmd->cmdaddr;  /* payload ptr */
+	dptr += 2;		/* Skip Words 0-1 in payload */
+
+	*wptr++ = *dptr++;	/* Word 2 */
+	*wptr++ = *dptr++;	/* Word 3 */
+	*wptr++ = *dptr++;	/* Word 4 */
+	dptr++;			/* Skip Word 5 in payload */
+	*wptr++ = *dptr++;	/* Word 6 */
+	*wptr++ = *dptr++;	/* Word 7 */
+	dptr += 8;		/* Skip Words 8-15 in payload */
+	*wptr++ = *dptr++;	/* Word 16 */
+	*wptr++ = *dptr++;	/* Word 17 */
+	*wptr++ = *dptr++;	/* Word 18 */
+	*wptr++ = *dptr++;	/* Word 19 */
+	*wptr++ = *dptr++;	/* Word 20 */
+	*wptr++ = *dptr++;	/* Word 21 */
+	*wptr++ = *dptr++;	/* Word 22 */
+	*wptr   = *dptr;	/* Word 23 */
+}
+
+/**
+ * lpfc_nvme_io_cmd_wqe_cmpl - Complete an NVME-over-FCP IO
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @lpfc_nvme_lport: Pointer to the driver's local port data
+ * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
+ *
+ * Driver registers this routine as it io request handler.  This
+ * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
+ * data structure to the rport indicated in @lpfc_nvme_rport.
+ *
+ * Return value :
+ *   0 - Success
+ *   TODO: What are the failure codes.
+ **/
+static void
+lpfc_nvme_io_cmd_wqe_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pwqeIn,
+			  struct lpfc_wcqe_complete *wcqe)
+{
+	struct lpfc_nvme_buf *lpfc_ncmd =
+		(struct lpfc_nvme_buf *)pwqeIn->context1;
+	struct lpfc_vport *vport = pwqeIn->vport;
+	struct nvmefc_fcp_req *nCmd;
+	struct nvme_fc_ersp_iu *ep;
+	struct nvme_fc_cmd_iu *cp;
+	struct lpfc_nvme_rport *rport;
+	struct lpfc_nodelist *ndlp;
+	unsigned long flags;
+	uint32_t code;
+	uint16_t cid, sqhd, data;
+	uint32_t *ptr;
+
+	/* Sanity check on return of outstanding command */
+	if (!lpfc_ncmd || !lpfc_ncmd->nvmeCmd || !lpfc_ncmd->nrport) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
+				 "6071 Completion pointers bad on wqe %p.\n",
+				 wcqe);
+		return;
+	}
+	phba->fc4NvmeIoCmpls++;
+
+	nCmd = lpfc_ncmd->nvmeCmd;
+	rport = lpfc_ncmd->nrport;
+
+	/*
+	 * Catch race where our node has transitioned, but the
+	 * transport is still transitioning.
+	 */
+	ndlp = rport->ndlp;
+	if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
+				 "6061 rport %p, ndlp %p, DID x%06x ndlp "
+				 "not ready.\n",
+				 rport, ndlp, rport->remoteport->port_id);
+
+		ndlp = lpfc_findnode_did(vport, rport->remoteport->port_id);
+		if (!ndlp) {
+			lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
+					 "6062 Ignoring NVME cmpl.  No ndlp\n");
+			goto out_err;
+		}
+	}
+
+	code = bf_get(lpfc_wcqe_c_code, wcqe);
+	if (code == CQE_CODE_NVME_ERSP) {
+		/* For this type of CQE, we need to rebuild the rsp */
+		ep = (struct nvme_fc_ersp_iu *)nCmd->rspaddr;
+
+		/*
+		 * Get Command Id from cmd to plug into response. This
+		 * code is not needed in the next NVME Transport drop.
+		 */
+		cp = (struct nvme_fc_cmd_iu *)nCmd->cmdaddr;
+		cid = cp->sqe.common.command_id;
+
+		/*
+		 * RSN is in CQE word 2
+		 * SQHD is in CQE Word 3 bits 15:0
+		 * Cmd Specific info is in CQE Word 1
+		 * and in CQE Word 0 bits 15:0
+		 */
+		sqhd = bf_get(lpfc_wcqe_c_sqhead, wcqe);
+
+		/* Now lets build the NVME ERSP IU */
+		ep->iu_len = cpu_to_be16(8);
+		ep->rsn = wcqe->parameter;
+		ep->xfrd_len = cpu_to_be32(nCmd->payload_length);
+		ep->rsvd12 = 0;
+		ptr = (uint32_t *)&ep->cqe.result.u64;
+		*ptr++ = wcqe->total_data_placed;
+		data = bf_get(lpfc_wcqe_c_ersp0, wcqe);
+		*ptr = (uint32_t)data;
+		ep->cqe.sq_head = sqhd;
+		ep->cqe.sq_id =  nCmd->sqid;
+		ep->cqe.command_id = cid;
+		ep->cqe.status = 0;
+
+		lpfc_ncmd->status = IOSTAT_SUCCESS;
+		lpfc_ncmd->result = 0;
+		nCmd->rcv_rsplen = LPFC_NVME_ERSP_LEN;
+		nCmd->transferred_length = nCmd->payload_length;
+	} else {
+		lpfc_ncmd->status = (bf_get(lpfc_wcqe_c_status, wcqe) &
+			    LPFC_IOCB_STATUS_MASK);
+		lpfc_ncmd->result = wcqe->parameter;
+
+		/* For NVME, the only failure path that results in an
+		 * IO error is when the adapter rejects it.  All other
+		 * conditions are a success case and resolved by the
+		 * transport.
+		 * IOSTAT_FCP_RSP_ERROR means:
+		 * 1. Length of data received doesn't match total
+		 *    transfer length in WQE
+		 * 2. If the RSP payload does NOT match these cases:
+		 *    a. RSP length 12/24 bytes and all zeros
+		 *    b. NVME ERSP
+		 */
+		switch (lpfc_ncmd->status) {
+		case IOSTAT_SUCCESS:
+			nCmd->transferred_length = wcqe->total_data_placed;
+			nCmd->rcv_rsplen = 0;
+			nCmd->status = 0;
+			break;
+		case IOSTAT_FCP_RSP_ERROR:
+			nCmd->transferred_length = wcqe->total_data_placed;
+			nCmd->rcv_rsplen = wcqe->parameter;
+			nCmd->status = 0;
+			/* Sanity check */
+			if (nCmd->rcv_rsplen == LPFC_NVME_ERSP_LEN)
+				break;
+			lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
+					 "6081 NVME Completion Protocol Error: "
+					 "status x%x result x%x placed x%x\n",
+					 lpfc_ncmd->status, lpfc_ncmd->result,
+					 wcqe->total_data_placed);
+			break;
+		default:
+out_err:
+			lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
+					 "6072 NVME Completion Error: "
+					 "status x%x result x%x placed x%x\n",
+					 lpfc_ncmd->status, lpfc_ncmd->result,
+					 wcqe->total_data_placed);
+			nCmd->transferred_length = 0;
+			nCmd->rcv_rsplen = 0;
+			nCmd->status = NVME_SC_FC_TRANSPORT_ERROR;
+		}
+	}
+
+	/* pick up SLI4 exhange busy condition */
+	if (bf_get(lpfc_wcqe_c_xb, wcqe))
+		lpfc_ncmd->flags |= LPFC_SBUF_XBUSY;
+	else
+		lpfc_ncmd->flags &= ~LPFC_SBUF_XBUSY;
+
+	if (ndlp && NLP_CHK_NODE_ACT(ndlp))
+		atomic_dec(&ndlp->cmd_pending);
+
+	/* Update stats and complete the IO.  There is
+	 * no need for dma unprep because the nvme_transport
+	 * owns the dma address.
+	 */
+	nCmd->done(nCmd);
+
+	spin_lock_irqsave(&phba->hbalock, flags);
+	lpfc_ncmd->nrport = NULL;
+	spin_unlock_irqrestore(&phba->hbalock, flags);
+
+	lpfc_release_nvme_buf(phba, lpfc_ncmd);
+}
+
+
+/**
+ * lpfc_nvme_prep_io_cmd - Issue an NVME-over-FCP IO
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @lpfc_nvme_lport: Pointer to the driver's local port data
+ * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
+ * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
+ * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
+ *
+ * Driver registers this routine as it io request handler.  This
+ * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
+ * data structure to the rport indicated in @lpfc_nvme_rport.
+ *
+ * Return value :
+ *   0 - Success
+ *   TODO: What are the failure codes.
+ **/
+static int
+lpfc_nvme_prep_io_cmd(struct lpfc_vport *vport,
+		      struct lpfc_nvme_buf *lpfc_ncmd,
+		      struct lpfc_nodelist *pnode)
+{
+	struct lpfc_hba *phba = vport->phba;
+	struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
+	struct lpfc_iocbq *pwqeq = &(lpfc_ncmd->cur_iocbq);
+	union lpfc_wqe128 *wqe = (union lpfc_wqe128 *)&pwqeq->wqe;
+	uint32_t req_len;
+
+	if (!pnode || !NLP_CHK_NODE_ACT(pnode))
+		return -EINVAL;
+
+	/*
+	 * There are three possibilities here - use scatter-gather segment, use
+	 * the single mapping, or neither.
+	 */
+	wqe->fcp_iwrite.initial_xfer_len = 0;
+	if (nCmd->sg_cnt) {
+		if (nCmd->io_dir == NVMEFC_FCP_WRITE) {
+			/* Word 5 */
+			if ((phba->cfg_nvme_enable_fb) &&
+			    (pnode->nlp_flag & NLP_FIRSTBURST)) {
+				req_len = lpfc_ncmd->nvmeCmd->payload_length;
+				if (req_len < pnode->nvme_fb_size)
+					wqe->fcp_iwrite.initial_xfer_len =
+						req_len;
+				else
+					wqe->fcp_iwrite.initial_xfer_len =
+						pnode->nvme_fb_size;
+			}
+
+			/* Word 7 */
+			bf_set(wqe_cmnd, &wqe->generic.wqe_com,
+			       CMD_FCP_IWRITE64_WQE);
+			bf_set(wqe_pu, &wqe->generic.wqe_com,
+			       PARM_READ_CHECK);
+
+			/* Word 10 */
+			bf_set(wqe_qosd, &wqe->fcp_iwrite.wqe_com, 0);
+			bf_set(wqe_iod, &wqe->fcp_iwrite.wqe_com,
+			       LPFC_WQE_IOD_WRITE);
+			bf_set(wqe_lenloc, &wqe->fcp_iwrite.wqe_com,
+			       LPFC_WQE_LENLOC_WORD4);
+			if (phba->cfg_nvme_oas)
+				bf_set(wqe_oas, &wqe->fcp_iwrite.wqe_com, 1);
+
+			/* Word 11 */
+			bf_set(wqe_cmd_type, &wqe->generic.wqe_com,
+			       NVME_WRITE_CMD);
+
+			/* Word 16 */
+			wqe->words[16] = LPFC_NVME_EMBED_WRITE;
+
+			phba->fc4NvmeOutputRequests++;
+		} else {
+			/* Word 7 */
+			bf_set(wqe_cmnd, &wqe->generic.wqe_com,
+			       CMD_FCP_IREAD64_WQE);
+			bf_set(wqe_pu, &wqe->generic.wqe_com,
+			       PARM_READ_CHECK);
+
+			/* Word 10 */
+			bf_set(wqe_qosd, &wqe->fcp_iread.wqe_com, 0);
+			bf_set(wqe_iod, &wqe->fcp_iread.wqe_com,
+			       LPFC_WQE_IOD_READ);
+			bf_set(wqe_lenloc, &wqe->fcp_iread.wqe_com,
+			       LPFC_WQE_LENLOC_WORD4);
+			if (phba->cfg_nvme_oas)
+				bf_set(wqe_oas, &wqe->fcp_iread.wqe_com, 1);
+
+			/* Word 11 */
+			bf_set(wqe_cmd_type, &wqe->generic.wqe_com,
+			       NVME_READ_CMD);
+
+			/* Word 16 */
+			wqe->words[16] = LPFC_NVME_EMBED_READ;
+
+			phba->fc4NvmeInputRequests++;
+		}
+	} else {
+		/* Word 4 */
+		wqe->fcp_icmd.rsrvd4 = 0;
+
+		/* Word 7 */
+		bf_set(wqe_cmnd, &wqe->generic.wqe_com, CMD_FCP_ICMND64_WQE);
+		bf_set(wqe_pu, &wqe->generic.wqe_com, 0);
+
+		/* Word 10 */
+		bf_set(wqe_qosd, &wqe->fcp_icmd.wqe_com, 1);
+		bf_set(wqe_iod, &wqe->fcp_icmd.wqe_com, LPFC_WQE_IOD_WRITE);
+		bf_set(wqe_lenloc, &wqe->fcp_icmd.wqe_com,
+		       LPFC_WQE_LENLOC_NONE);
+		if (phba->cfg_nvme_oas)
+			bf_set(wqe_oas, &wqe->fcp_icmd.wqe_com, 1);
+
+		/* Word 11 */
+		bf_set(wqe_cmd_type, &wqe->generic.wqe_com, NVME_READ_CMD);
+
+		/* Word 16 */
+		wqe->words[16] = LPFC_NVME_EMBED_CMD;
+
+		phba->fc4NvmeControlRequests++;
+	}
+	/*
+	 * Finish initializing those WQE fields that are independent
+	 * of the nvme_cmnd request_buffer
+	 */
+
+	/* Word 6 */
+	bf_set(wqe_ctxt_tag, &wqe->generic.wqe_com,
+	       phba->sli4_hba.rpi_ids[pnode->nlp_rpi]);
+	bf_set(wqe_xri_tag, &wqe->generic.wqe_com, pwqeq->sli4_xritag);
+
+	/* Word 7 */
+	/* Preserve Class data in the ndlp. */
+	bf_set(wqe_class, &wqe->generic.wqe_com,
+	       (pnode->nlp_fcp_info & 0x0f));
+
+	/* Word 8 */
+	wqe->generic.wqe_com.abort_tag = pwqeq->iotag;
+
+	/* Word 9 */
+	bf_set(wqe_reqtag, &wqe->generic.wqe_com, pwqeq->iotag);
+
+	/* Word 11 */
+	bf_set(wqe_cqid, &wqe->generic.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
+
+	pwqeq->vport = vport;
+	return 0;
+}
+
+
+/**
+ * lpfc_nvme_prep_io_dma - Issue an NVME-over-FCP IO
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @lpfc_nvme_lport: Pointer to the driver's local port data
+ * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
+ * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
+ * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
+ *
+ * Driver registers this routine as it io request handler.  This
+ * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
+ * data structure to the rport indicated in @lpfc_nvme_rport.
+ *
+ * Return value :
+ *   0 - Success
+ *   TODO: What are the failure codes.
+ **/
+static int
+lpfc_nvme_prep_io_dma(struct lpfc_vport *vport,
+		      struct lpfc_nvme_buf *lpfc_ncmd)
+{
+	struct lpfc_hba *phba = vport->phba;
+	struct nvmefc_fcp_req *nCmd = lpfc_ncmd->nvmeCmd;
+	union lpfc_wqe128 *wqe = (union lpfc_wqe128 *)&lpfc_ncmd->cur_iocbq.wqe;
+	struct sli4_sge *sgl = lpfc_ncmd->nvme_sgl;
+	struct scatterlist *data_sg;
+	struct sli4_sge *first_data_sgl;
+	dma_addr_t physaddr;
+	uint32_t num_bde = 0;
+	uint32_t dma_len;
+	uint32_t dma_offset = 0;
+	int nseg, i;
+
+	/* Fix up the command and response DMA stuff. */
+	lpfc_nvme_adj_fcp_sgls(vport, lpfc_ncmd, nCmd);
+
+	/*
+	 * There are three possibilities here - use scatter-gather segment, use
+	 * the single mapping, or neither.
+	 */
+	if (nCmd->sg_cnt) {
+		/*
+		 * Jump over the cmd and rsp SGEs.  The fix routine
+		 * has already adjusted for this.
+		 */
+		sgl += 2;
+
+		first_data_sgl = sgl;
+		lpfc_ncmd->seg_cnt = nCmd->sg_cnt;
+		if (lpfc_ncmd->seg_cnt > phba->cfg_sg_seg_cnt) {
+			lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
+					"6058 Too many sg segments from "
+					"NVME Transport.  Max %d, "
+					"nvmeIO sg_cnt %d\n",
+					phba->cfg_sg_seg_cnt,
+					lpfc_ncmd->seg_cnt);
+			lpfc_ncmd->seg_cnt = 0;
+			return 1;
+		}
+
+		/*
+		 * The driver established a maximum scatter-gather segment count
+		 * during probe that limits the number of sg elements in any
+		 * single nvme command.  Just run through the seg_cnt and format
+		 * the sge's.
+		 */
+		nseg = nCmd->sg_cnt;
+		data_sg = nCmd->first_sgl;
+		for (i = 0; i < nseg; i++) {
+			if (data_sg == NULL) {
+				lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
+						"6059 dptr err %d, nseg %d\n",
+						i, nseg);
+				lpfc_ncmd->seg_cnt = 0;
+				return 1;
+			}
+			physaddr = data_sg->dma_address;
+			dma_len = data_sg->length;
+			sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr));
+			sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr));
+			sgl->word2 = le32_to_cpu(sgl->word2);
+			if ((num_bde + 1) == nseg)
+				bf_set(lpfc_sli4_sge_last, sgl, 1);
+			else
+				bf_set(lpfc_sli4_sge_last, sgl, 0);
+			bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
+			bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_DATA);
+			sgl->word2 = cpu_to_le32(sgl->word2);
+			sgl->sge_len = cpu_to_le32(dma_len);
+
+			dma_offset += dma_len;
+			data_sg = sg_next(data_sg);
+			sgl++;
+		}
+	} else {
+		/* For this clause to be valid, the payload_length
+		 * and sg_cnt must zero.
+		 */
+		if (nCmd->payload_length != 0) {
+			lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
+					"6063 NVME DMA Prep Err: sg_cnt %d "
+					"payload_length x%x\n",
+					nCmd->sg_cnt, nCmd->payload_length);
+			return 1;
+		}
+	}
+
+	/*
+	 * Due to difference in data length between DIF/non-DIF paths,
+	 * we need to set word 4 of WQE here
+	 */
+	wqe->fcp_iread.total_xfer_len = nCmd->payload_length;
+	return 0;
+}
+
+/**
+ * lpfc_nvme_fcp_io_submit - Issue an NVME-over-FCP IO
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @lpfc_nvme_lport: Pointer to the driver's local port data
+ * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
+ * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
+ * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
+ *
+ * Driver registers this routine as it io request handler.  This
+ * routine issues an fcp WQE with data from the @lpfc_nvme_fcpreq
+ * data structure to the rport
+ indicated in @lpfc_nvme_rport.
+ *
+ * Return value :
+ *   0 - Success
+ *   TODO: What are the failure codes.
+ **/
+static int
+lpfc_nvme_fcp_io_submit(struct nvme_fc_local_port *pnvme_lport,
+			struct nvme_fc_remote_port *pnvme_rport,
+			void *hw_queue_handle,
+			struct nvmefc_fcp_req *pnvme_fcreq)
+{
+	int ret = 0;
+	struct lpfc_nvme_lport *lport;
+	struct lpfc_vport *vport;
+	struct lpfc_hba *phba;
+	struct lpfc_nodelist *ndlp;
+	struct lpfc_nvme_buf *lpfc_ncmd;
+	struct lpfc_nvme_rport *rport;
+	struct lpfc_nvme_qhandle *lpfc_queue_info;
+
+	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
+	vport = lport->vport;
+	phba = vport->phba;
+
+	rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
+	lpfc_queue_info = (struct lpfc_nvme_qhandle *)hw_queue_handle;
+
+	/*
+	 * Catch race where our node has transitioned, but the
+	 * transport is still transitioning.
+	 */
+	ndlp = rport->ndlp;
+	if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
+				 "6053 rport %p, ndlp %p, DID x%06x "
+				 "ndlp not ready.\n",
+				 rport, ndlp, pnvme_rport->port_id);
+
+		ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
+		if (!ndlp) {
+			lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
+					 "6066 Missing node for DID %x\n",
+					 pnvme_rport->port_id);
+			ret = -ENODEV;
+			goto out_fail;
+		}
+	}
+
+	/* The remote node has to be a mapped target or it's an error. */
+	if ((ndlp->nlp_type & NLP_NVME_TARGET) &&
+	    (ndlp->nlp_state != NLP_STE_MAPPED_NODE)) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_IOERR,
+				 "6036 rport %p, DID x%06x not ready for "
+				 "IO. State x%x, Type x%x\n",
+				 rport, pnvme_rport->port_id,
+				 ndlp->nlp_state, ndlp->nlp_type);
+		ret = -ENODEV;
+		goto out_fail;
+
+	}
+
+	/* 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) {
+		ret = -EAGAIN;
+		goto out_fail;
+	}
+
+	lpfc_ncmd = lpfc_get_nvme_buf(phba, ndlp);
+	if (lpfc_ncmd == NULL) {
+		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_IOERR,
+				 "6065 driver's buffer pool is empty, "
+				 "IO failed\n");
+		ret = -ENOMEM;
+		goto out_fail;
+	}
+
+	/*
+	 * Store the data needed by the driver to issue, abort, and complete
+	 * an IO.
+	 * Do not let the IO hang out forever.  There is no midlayer issuing
+	 * an abort so inform the FW of the maximum IO pending time.
+	 */
+	pnvme_fcreq->private = (void *)lpfc_ncmd;
+	lpfc_ncmd->nvmeCmd = pnvme_fcreq;
+	lpfc_ncmd->nrport = rport;
+	lpfc_ncmd->start_time = jiffies;
+
+	lpfc_nvme_prep_io_cmd(vport, lpfc_ncmd, ndlp);
+	ret = lpfc_nvme_prep_io_dma(vport, lpfc_ncmd);
+	if (ret) {
+		ret = -ENOMEM;
+		goto out_free_nvme_buf;
+	}
+
+	atomic_inc(&ndlp->cmd_pending);
+
+	/*
+	 * Issue the IO on the WQ indicated by index in the hw_queue_handle.
+	 * This identfier was create in our hardware queue create callback
+	 * routine. The driver now is dependent on the IO queue steering from
+	 * the transport.  We are trusting the upper NVME layers know which
+	 * index to use and that they have affinitized a CPU to this hardware
+	 * queue. A hardware queue maps to a driver MSI-X vector/EQ/CQ/WQ.
+	 */
+	lpfc_ncmd->cur_iocbq.hba_wqidx = lpfc_queue_info->index;
+
+	ret = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, &lpfc_ncmd->cur_iocbq);
+	if (ret) {
+		atomic_dec(&ndlp->cmd_pending);
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_IOERR,
+				 "6113 FCP could not issue WQE err %x "
+				 "sid: x%x did: x%x oxid: x%x\n",
+				 ret, vport->fc_myDID, ndlp->nlp_DID,
+				 lpfc_ncmd->cur_iocbq.sli4_xritag);
+		ret = -EINVAL;
+		goto out_free_nvme_buf;
+	}
+
+	return 0;
+
+ out_free_nvme_buf:
+	lpfc_release_nvme_buf(phba, lpfc_ncmd);
+ out_fail:
+	return ret;
+}
+
+/**
+ * lpfc_nvme_abort_fcreq_cmpl - Complete an NVME FCP abort request.
+ * @phba: Pointer to HBA context object
+ * @cmdiocb: Pointer to command iocb object.
+ * @rspiocb: Pointer to response iocb object.
+ *
+ * This is the callback function for any NVME FCP IO that was aborted.
+ *
+ * Return value:
+ *   None
+ **/
+void
+lpfc_nvme_abort_fcreq_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *cmdiocb,
+			   struct lpfc_wcqe_complete *abts_cmpl)
+{
+	lpfc_printf_log(phba, KERN_ERR, LOG_NVME,
+			"6145 ABORT_XRI_CN completing on rpi x%x "
+			"original iotag x%x, abort cmd iotag x%x "
+			"req_tag x%x, status x%x, hwstatus x%x\n",
+			cmdiocb->iocb.un.acxri.abortContextTag,
+			cmdiocb->iocb.un.acxri.abortIoTag,
+			cmdiocb->iotag,
+			bf_get(lpfc_wcqe_c_request_tag, abts_cmpl),
+			bf_get(lpfc_wcqe_c_status, abts_cmpl),
+			bf_get(lpfc_wcqe_c_hw_status, abts_cmpl));
+	lpfc_sli_release_iocbq(phba, cmdiocb);
+}
+
+/**
+ * lpfc_nvme_fcp_abort - Issue an NVME-over-FCP ABTS
+ * @lpfc_pnvme: Pointer to the driver's nvme instance data
+ * @lpfc_nvme_lport: Pointer to the driver's local port data
+ * @lpfc_nvme_rport: Pointer to the rport getting the @lpfc_nvme_ereq
+ * @lpfc_nvme_fcreq: IO request from nvme fc to driver.
+ * @hw_queue_handle: Driver-returned handle in lpfc_nvme_create_queue
+ *
+ * Driver registers this routine as its nvme request io abort handler.  This
+ * routine issues an fcp Abort WQE with data from the @lpfc_nvme_fcpreq
+ * data structure to the rport indicated in @lpfc_nvme_rport.  This routine
+ * is executed asynchronously - one the target is validated as "MAPPED" and
+ * ready for IO, the driver issues the abort request and returns.
+ *
+ * Return value:
+ *   None
+ **/
+static void
+lpfc_nvme_fcp_abort(struct nvme_fc_local_port *pnvme_lport,
+		    struct nvme_fc_remote_port *pnvme_rport,
+		    void *hw_queue_handle,
+		    struct nvmefc_fcp_req *pnvme_fcreq)
+{
+	struct lpfc_nvme_lport *lport;
+	struct lpfc_vport *vport;
+	struct lpfc_hba *phba;
+	struct lpfc_nodelist *ndlp;
+	struct lpfc_nvme_rport *rport;
+	struct lpfc_nvme_buf *lpfc_nbuf;
+	struct lpfc_iocbq *abts_buf;
+	struct lpfc_iocbq *nvmereq_wqe;
+	union lpfc_wqe *abts_wqe;
+	unsigned long flags;
+	int ret_val;
+
+	lport = (struct lpfc_nvme_lport *)pnvme_lport->private;
+	rport = (struct lpfc_nvme_rport *)pnvme_rport->private;
+	vport = lport->vport;
+	phba = vport->phba;
+
+	/* Announce entry to new IO submit field. */
+	lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
+			 "6002 Abort Request to rport DID x%06x "
+			 "for nvme_fc_req %p\n",
+			 pnvme_rport->port_id,
+			 pnvme_fcreq);
+
+	/*
+	 * Catch race where our node has transitioned, but the
+	 * transport is still transitioning.
+	 */
+	ndlp = rport->ndlp;
+	if (!ndlp || !NLP_CHK_NODE_ACT(ndlp)) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_ABTS,
+				 "6054 rport %p, ndlp %p, DID x%06x ndlp "
+				 " not ready.\n",
+				 rport, ndlp, pnvme_rport->port_id);
+
+		ndlp = lpfc_findnode_did(vport, pnvme_rport->port_id);
+		if (!ndlp) {
+			lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_ABTS,
+					 "6055 Could not find node for "
+					 "DID %x\n",
+					 pnvme_rport->port_id);
+			return;
+		}
+	}
+
+	/* The remote node has to be ready to send an abort. */
+	if ((ndlp->nlp_state != NLP_STE_MAPPED_NODE) &&
+	    !(ndlp->nlp_type & NLP_NVME_TARGET)) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NODE | LOG_NVME_ABTS,
+				 "6048 rport %p, DID x%06x not ready for "
+				 "IO. State x%x, Type x%x\n",
+				 rport, pnvme_rport->port_id,
+				 ndlp->nlp_state, ndlp->nlp_type);
+		return;
+	}
+
+	/* If the hba is getting reset, this flag is set.  It is
+	 * cleared when the reset is complete and rings reestablished.
+	 */
+	spin_lock_irqsave(&phba->hbalock, flags);
+	/* driver queued commands are in process of being flushed */
+	if (phba->hba_flag & HBA_NVME_IOQ_FLUSH) {
+		spin_unlock_irqrestore(&phba->hbalock, flags);
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+				 "6139 Driver in reset cleanup - flushing "
+				 "NVME Req now.  hba_flag x%x\n",
+				 phba->hba_flag);
+		return;
+	}
+
+	lpfc_nbuf = (struct lpfc_nvme_buf *)pnvme_fcreq->private;
+	if (!lpfc_nbuf) {
+		spin_unlock_irqrestore(&phba->hbalock, flags);
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+				 "6140 NVME IO req has no matching lpfc nvme "
+				 "io buffer.  Skipping abort req.\n");
+		return;
+	} else if (!lpfc_nbuf->nvmeCmd) {
+		spin_unlock_irqrestore(&phba->hbalock, flags);
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+				 "6141 lpfc NVME IO req has no nvme_fcreq "
+				 "io buffer.  Skipping abort req.\n");
+		return;
+	}
+
+	/*
+	 * The lpfc_nbuf and the mapped nvme_fcreq in the driver's
+	 * state must match the nvme_fcreq passed by the nvme
+	 * transport.  If they don't match, it is likely the driver
+	 * has already completed the NVME IO and the nvme transport
+	 * has not seen it yet.
+	 */
+	if (lpfc_nbuf->nvmeCmd != pnvme_fcreq) {
+		spin_unlock_irqrestore(&phba->hbalock, flags);
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+				 "6143 NVME req mismatch: "
+				 "lpfc_nbuf %p nvmeCmd %p, "
+				 "pnvme_fcreq %p.  Skipping Abort\n",
+				 lpfc_nbuf, lpfc_nbuf->nvmeCmd,
+				 pnvme_fcreq);
+		return;
+	}
+
+	/* Don't abort IOs no longer on the pending queue. */
+	nvmereq_wqe = &lpfc_nbuf->cur_iocbq;
+	if (!(nvmereq_wqe->iocb_flag & LPFC_IO_ON_TXCMPLQ)) {
+		spin_unlock_irqrestore(&phba->hbalock, flags);
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+				 "6142 NVME IO req %p not queued - skipping "
+				 "abort req\n",
+				 pnvme_fcreq);
+		return;
+	}
+
+	/* Outstanding abort is in progress */
+	if (nvmereq_wqe->iocb_flag & LPFC_DRIVER_ABORTED) {
+		spin_unlock_irqrestore(&phba->hbalock, flags);
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+				 "6144 Outstanding NVME I/O Abort Request "
+				 "still pending on nvme_fcreq %p, "
+				 "lpfc_ncmd %p\n",
+				 pnvme_fcreq, lpfc_nbuf);
+		return;
+	}
+
+	abts_buf = __lpfc_sli_get_iocbq(phba);
+	if (!abts_buf) {
+		spin_unlock_irqrestore(&phba->hbalock, flags);
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+				 "6136 No available abort wqes. Skipping "
+				 "Abts req for nvme_fcreq %p.\n",
+				 pnvme_fcreq);
+		return;
+	}
+
+	/* Ready - mark outstanding as aborted by driver. */
+	nvmereq_wqe->iocb_flag |= LPFC_DRIVER_ABORTED;
+
+	/* Complete prepping the abort wqe and issue to the FW. */
+	abts_wqe = &abts_buf->wqe;
+
+	/* WQEs are reused.  Clear stale data and set key fields to
+	 * zero like ia, iaab, iaar, xri_tag, and ctxt_tag.
+	 */
+	memset(abts_wqe, 0, sizeof(union lpfc_wqe));
+	bf_set(abort_cmd_criteria, &abts_wqe->abort_cmd, T_XRI_TAG);
+
+	/* word 7 */
+	bf_set(wqe_ct, &abts_wqe->abort_cmd.wqe_com, 0);
+	bf_set(wqe_cmnd, &abts_wqe->abort_cmd.wqe_com, CMD_ABORT_XRI_CX);
+	bf_set(wqe_class, &abts_wqe->abort_cmd.wqe_com,
+	       nvmereq_wqe->iocb.ulpClass);
+
+	/* word 8 - tell the FW to abort the IO associated with this
+	 * outstanding exchange ID.
+	 */
+	abts_wqe->abort_cmd.wqe_com.abort_tag = nvmereq_wqe->sli4_xritag;
+
+	/* word 9 - this is the iotag for the abts_wqe completion. */
+	bf_set(wqe_reqtag, &abts_wqe->abort_cmd.wqe_com,
+	       abts_buf->iotag);
+
+	/* word 10 */
+	bf_set(wqe_wqid, &abts_wqe->abort_cmd.wqe_com, nvmereq_wqe->hba_wqidx);
+	bf_set(wqe_qosd, &abts_wqe->abort_cmd.wqe_com, 1);
+	bf_set(wqe_lenloc, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_LENLOC_NONE);
+
+	/* word 11 */
+	bf_set(wqe_cmd_type, &abts_wqe->abort_cmd.wqe_com, OTHER_COMMAND);
+	bf_set(wqe_wqec, &abts_wqe->abort_cmd.wqe_com, 1);
+	bf_set(wqe_cqid, &abts_wqe->abort_cmd.wqe_com, LPFC_WQE_CQ_ID_DEFAULT);
+
+	/* ABTS WQE must go to the same WQ as the WQE to be aborted */
+	abts_buf->iocb_flag |= LPFC_IO_NVME;
+	abts_buf->hba_wqidx = nvmereq_wqe->hba_wqidx;
+	abts_buf->vport = vport;
+	abts_buf->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl;
+	ret_val = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abts_buf);
+	spin_unlock_irqrestore(&phba->hbalock, flags);
+	if (ret_val == IOCB_ERROR) {
+		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+				 "6137 Failed abts issue_wqe with status x%x "
+				 "for nvme_fcreq %p.\n",
+				 ret_val, pnvme_fcreq);
+		lpfc_sli_release_iocbq(phba, abts_buf);
+		return;
+	}
+
+	lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
+			 "6138 Transport Abort NVME Request Issued for\n"
+			 "ox_id x%x on reqtag x%x\n",
+			 nvmereq_wqe->sli4_xritag,
+			 abts_buf->iotag);
+}
+
+/* Declare and initialization an instance of the FC NVME template. */
+static struct nvme_fc_port_template lpfc_nvme_template = {
+	/* initiator-based functions */
+	.localport_delete  = lpfc_nvme_localport_delete,
+	.remoteport_delete = lpfc_nvme_remoteport_delete,
+	.create_queue = lpfc_nvme_create_queue,
+	.delete_queue = lpfc_nvme_delete_queue,
+	.ls_req       = lpfc_nvme_ls_req,
+	.fcp_io       = lpfc_nvme_fcp_io_submit,
+	.ls_abort     = lpfc_nvme_ls_abort,
+	.fcp_abort    = lpfc_nvme_fcp_abort,
+
+	.max_hw_queues = 1,
+	.max_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
+	.max_dif_sgl_segments = LPFC_NVME_DEFAULT_SEGS,
+	.dma_boundary = 0xFFFFFFFF,
+
+	/* Sizes of additional private data for data structures.
+	 * No use for the last two sizes at this time.
+	 */
+	.local_priv_sz = sizeof(struct lpfc_nvme_lport),
+	.remote_priv_sz = sizeof(struct lpfc_nvme_rport),
+	.lsrqst_priv_sz = 0,
+	.fcprqst_priv_sz = 0,
+};
+
+/**
+ * lpfc_sli4_post_nvme_sgl_block - post a block of nvme sgl list to firmware
+ * @phba: pointer to lpfc hba data structure.
+ * @nblist: pointer to nvme buffer list.
+ * @count: number of scsi buffers on the list.
+ *
+ * This routine is invoked to post a block of @count scsi sgl pages from a
+ * SCSI buffer list @nblist to the HBA using non-embedded mailbox command.
+ * No Lock is held.
+ *
+ **/
+static int
+lpfc_sli4_post_nvme_sgl_block(struct lpfc_hba *phba,
+			      struct list_head *nblist,
+			      int count)
+{
+	struct lpfc_nvme_buf *lpfc_ncmd;
+	struct lpfc_mbx_post_uembed_sgl_page1 *sgl;
+	struct sgl_page_pairs *sgl_pg_pairs;
+	void *viraddr;
+	LPFC_MBOXQ_t *mbox;
+	uint32_t reqlen, alloclen, pg_pairs;
+	uint32_t mbox_tmo;
+	uint16_t xritag_start = 0;
+	int rc = 0;
+	uint32_t shdr_status, shdr_add_status;
+	dma_addr_t pdma_phys_bpl1;
+	union lpfc_sli4_cfg_shdr *shdr;
+
+	/* Calculate the requested length of the dma memory */
+	reqlen = count * sizeof(struct sgl_page_pairs) +
+		 sizeof(union lpfc_sli4_cfg_shdr) + sizeof(uint32_t);
+	if (reqlen > SLI4_PAGE_SIZE) {
+		lpfc_printf_log(phba, KERN_WARNING, LOG_INIT,
+				"6118 Block sgl registration required DMA "
+				"size (%d) great than a page\n", reqlen);
+		return -ENOMEM;
+	}
+	mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
+	if (!mbox) {
+		lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
+				"6119 Failed to allocate mbox cmd memory\n");
+		return -ENOMEM;
+	}
+
+	/* Allocate DMA memory and set up the non-embedded mailbox command */
+	alloclen = lpfc_sli4_config(phba, mbox, LPFC_MBOX_SUBSYSTEM_FCOE,
+				LPFC_MBOX_OPCODE_FCOE_POST_SGL_PAGES, reqlen,
+				LPFC_SLI4_MBX_NEMBED);
+
+	if (alloclen < reqlen) {
+		lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
+				"6120 Allocated DMA memory size (%d) is "
+				"less than the requested DMA memory "
+				"size (%d)\n", alloclen, reqlen);
+		lpfc_sli4_mbox_cmd_free(phba, mbox);
+		return -ENOMEM;
+	}
+
+	/* Get the first SGE entry from the non-embedded DMA memory */
+	viraddr = mbox->sge_array->addr[0];
+
+	/* Set up the SGL pages in the non-embedded DMA pages */
+	sgl = (struct lpfc_mbx_post_uembed_sgl_page1 *)viraddr;
+	sgl_pg_pairs = &sgl->sgl_pg_pairs;
+
+	pg_pairs = 0;
+	list_for_each_entry(lpfc_ncmd, nblist, list) {
+		/* Set up the sge entry */
+		sgl_pg_pairs->sgl_pg0_addr_lo =
+			cpu_to_le32(putPaddrLow(lpfc_ncmd->dma_phys_sgl));
+		sgl_pg_pairs->sgl_pg0_addr_hi =
+			cpu_to_le32(putPaddrHigh(lpfc_ncmd->dma_phys_sgl));
+		if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE)
+			pdma_phys_bpl1 = lpfc_ncmd->dma_phys_sgl +
+						SGL_PAGE_SIZE;
+		else
+			pdma_phys_bpl1 = 0;
+		sgl_pg_pairs->sgl_pg1_addr_lo =
+			cpu_to_le32(putPaddrLow(pdma_phys_bpl1));
+		sgl_pg_pairs->sgl_pg1_addr_hi =
+			cpu_to_le32(putPaddrHigh(pdma_phys_bpl1));
+		/* Keep the first xritag on the list */
+		if (pg_pairs == 0)
+			xritag_start = lpfc_ncmd->cur_iocbq.sli4_xritag;
+		sgl_pg_pairs++;
+		pg_pairs++;
+	}
+	bf_set(lpfc_post_sgl_pages_xri, sgl, xritag_start);
+	bf_set(lpfc_post_sgl_pages_xricnt, sgl, pg_pairs);
+	/* Perform endian conversion if necessary */
+	sgl->word0 = cpu_to_le32(sgl->word0);
+
+	if (!phba->sli4_hba.intr_enable)
+		rc = lpfc_sli_issue_mbox(phba, mbox, MBX_POLL);
+	else {
+		mbox_tmo = lpfc_mbox_tmo_val(phba, mbox);
+		rc = lpfc_sli_issue_mbox_wait(phba, mbox, mbox_tmo);
+	}
+	shdr = (union lpfc_sli4_cfg_shdr *)&sgl->cfg_shdr;
+	shdr_status = bf_get(lpfc_mbox_hdr_status, &shdr->response);
+	shdr_add_status = bf_get(lpfc_mbox_hdr_add_status, &shdr->response);
+	if (rc != MBX_TIMEOUT)
+		lpfc_sli4_mbox_cmd_free(phba, mbox);
+	if (shdr_status || shdr_add_status || rc) {
+		lpfc_printf_log(phba, KERN_ERR, LOG_SLI,
+				"6125 POST_SGL_BLOCK mailbox command failed "
+				"status x%x add_status x%x mbx status x%x\n",
+				shdr_status, shdr_add_status, rc);
+		rc = -ENXIO;
+	}
+	return rc;
+}
+
+/**
+ * lpfc_post_nvme_sgl_list - Post blocks of nvme buffer sgls from a list
+ * @phba: pointer to lpfc hba data structure.
+ * @post_nblist: pointer to the nvme buffer list.
+ *
+ * This routine walks a list of nvme buffers that was passed in. It attempts
+ * to construct blocks of nvme buffer sgls which contains contiguous xris and
+ * uses the non-embedded SGL block post mailbox commands to post to the port.
+ * For single NVME buffer sgl with non-contiguous xri, if any, it shall use
+ * embedded SGL post mailbox command for posting. The @post_nblist passed in
+ * must be local list, thus no lock is needed when manipulate the list.
+ *
+ * Returns: 0 = failure, non-zero number of successfully posted buffers.
+ **/
+static int
+lpfc_post_nvme_sgl_list(struct lpfc_hba *phba,
+			     struct list_head *post_nblist, int sb_count)
+{
+	struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
+	int status, sgl_size;
+	int post_cnt = 0, block_cnt = 0, num_posting = 0, num_posted = 0;
+	dma_addr_t pdma_phys_sgl1;
+	int last_xritag = NO_XRI;
+	int cur_xritag;
+	LIST_HEAD(prep_nblist);
+	LIST_HEAD(blck_nblist);
+	LIST_HEAD(nvme_nblist);
+
+	/* sanity check */
+	if (sb_count <= 0)
+		return -EINVAL;
+
+	sgl_size = phba->cfg_sg_dma_buf_size;
+
+	list_for_each_entry_safe(lpfc_ncmd, lpfc_ncmd_next, post_nblist, list) {
+		list_del_init(&lpfc_ncmd->list);
+		block_cnt++;
+		if ((last_xritag != NO_XRI) &&
+		    (lpfc_ncmd->cur_iocbq.sli4_xritag != last_xritag + 1)) {
+			/* a hole in xri block, form a sgl posting block */
+			list_splice_init(&prep_nblist, &blck_nblist);
+			post_cnt = block_cnt - 1;
+			/* prepare list for next posting block */
+			list_add_tail(&lpfc_ncmd->list, &prep_nblist);
+			block_cnt = 1;
+		} else {
+			/* prepare list for next posting block */
+			list_add_tail(&lpfc_ncmd->list, &prep_nblist);
+			/* enough sgls for non-embed sgl mbox command */
+			if (block_cnt == LPFC_NEMBED_MBOX_SGL_CNT) {
+				list_splice_init(&prep_nblist, &blck_nblist);
+				post_cnt = block_cnt;
+				block_cnt = 0;
+			}
+		}
+		num_posting++;
+		last_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag;
+
+		/* end of repost sgl list condition for NVME buffers */
+		if (num_posting == sb_count) {
+			if (post_cnt == 0) {
+				/* last sgl posting block */
+				list_splice_init(&prep_nblist, &blck_nblist);
+				post_cnt = block_cnt;
+			} else if (block_cnt == 1) {
+				/* last single sgl with non-contiguous xri */
+				if (sgl_size > SGL_PAGE_SIZE)
+					pdma_phys_sgl1 =
+						lpfc_ncmd->dma_phys_sgl +
+						SGL_PAGE_SIZE;
+				else
+					pdma_phys_sgl1 = 0;
+				cur_xritag = lpfc_ncmd->cur_iocbq.sli4_xritag;
+				status = lpfc_sli4_post_sgl(phba,
+						lpfc_ncmd->dma_phys_sgl,
+						pdma_phys_sgl1, cur_xritag);
+				if (status) {
+					/* failure, put on abort nvme list */
+					lpfc_ncmd->exch_busy = 1;
+				} else {
+					/* success, put on NVME buffer list */
+					lpfc_ncmd->exch_busy = 0;
+					lpfc_ncmd->status = IOSTAT_SUCCESS;
+					num_posted++;
+				}
+				/* success, put on NVME buffer sgl list */
+				list_add_tail(&lpfc_ncmd->list, &nvme_nblist);
+			}
+		}
+
+		/* continue until a nembed page worth of sgls */
+		if (post_cnt == 0)
+			continue;
+
+		/* post block of NVME buffer list sgls */
+		status = lpfc_sli4_post_nvme_sgl_block(phba, &blck_nblist,
+						       post_cnt);
+
+		/* don't reset xirtag due to hole in xri block */
+		if (block_cnt == 0)
+			last_xritag = NO_XRI;
+
+		/* reset NVME buffer post count for next round of posting */
+		post_cnt = 0;
+
+		/* put posted NVME buffer-sgl posted on NVME buffer sgl list */
+		while (!list_empty(&blck_nblist)) {
+			list_remove_head(&blck_nblist, lpfc_ncmd,
+					 struct lpfc_nvme_buf, list);
+			if (status) {
+				/* failure, put on abort nvme list */
+				lpfc_ncmd->exch_busy = 1;
+			} else {
+				/* success, put on NVME buffer list */
+				lpfc_ncmd->exch_busy = 0;
+				lpfc_ncmd->status = IOSTAT_SUCCESS;
+				num_posted++;
+			}
+			list_add_tail(&lpfc_ncmd->list, &nvme_nblist);
+		}
+	}
+	/* Push NVME buffers with sgl posted to the available list */
+	while (!list_empty(&nvme_nblist)) {
+		list_remove_head(&nvme_nblist, lpfc_ncmd,
+				 struct lpfc_nvme_buf, list);
+		lpfc_release_nvme_buf(phba, lpfc_ncmd);
+	}
+	return num_posted;
+}
+
+/**
+ * lpfc_repost_nvme_sgl_list - Repost all the allocated nvme buffer sgls
+ * @phba: pointer to lpfc hba data structure.
+ *
+ * This routine walks the list of nvme 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 nvme buffers on the lpfc_abts_nvme_sgl_list
+ * to the lpfc_nvme_buf_list. If the repost fails, reject all nvme buffers.
+ *
+ * Returns: 0 = success, non-zero failure.
+ **/
+int
+lpfc_repost_nvme_sgl_list(struct lpfc_hba *phba)
+{
+	LIST_HEAD(post_nblist);
+	int num_posted, rc = 0;
+
+	/* get all NVME buffers need to repost to a local list */
+	spin_lock_irq(&phba->nvme_buf_list_get_lock);
+	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);
+	spin_unlock(&phba->nvme_buf_list_put_lock);
+	spin_unlock_irq(&phba->nvme_buf_list_get_lock);
+
+	/* post the list of nvme buffer sgls to port if available */
+	if (!list_empty(&post_nblist)) {
+		num_posted = lpfc_post_nvme_sgl_list(phba, &post_nblist,
+						phba->sli4_hba.nvme_xri_cnt);
+		/* failed to post any nvme buffer, return error */
+		if (num_posted == 0)
+			rc = -EIO;
+	}
+	return rc;
+}
+
+/**
+ * lpfc_new_nvme_buf - Scsi buffer allocator for HBA with SLI4 IF spec
+ * @vport: The virtual port for which this call being executed.
+ * @num_to_allocate: The requested number of buffers to allocate.
+ *
+ * This routine allocates nvme buffers for device with SLI-4 interface spec,
+ * the nvme buffer contains all the necessary information needed to initiate
+ * a NVME I/O. After allocating up to @num_to_allocate NVME buffers and put
+ * them on a list, it post them to the port by using SGL block post.
+ *
+ * Return codes:
+ *   int - number of nvme buffers that were allocated and posted.
+ *   0 = failure, less than num_to_alloc is a partial failure.
+ **/
+static int
+lpfc_new_nvme_buf(struct lpfc_vport *vport, int num_to_alloc)
+{
+	struct lpfc_hba *phba = vport->phba;
+	struct lpfc_nvme_buf *lpfc_ncmd;
+	struct lpfc_iocbq *pwqeq;
+	union lpfc_wqe128 *wqe;
+	struct sli4_sge *sgl;
+	dma_addr_t pdma_phys_sgl;
+	uint16_t iotag, lxri = 0;
+	int bcnt, num_posted, sgl_size;
+	LIST_HEAD(prep_nblist);
+	LIST_HEAD(post_nblist);
+	LIST_HEAD(nvme_nblist);
+
+	sgl_size = phba->cfg_sg_dma_buf_size;
+
+	for (bcnt = 0; bcnt < num_to_alloc; bcnt++) {
+		lpfc_ncmd = kzalloc(sizeof(struct lpfc_nvme_buf), GFP_KERNEL);
+		if (!lpfc_ncmd)
+			break;
+		/*
+		 * Get memory from the pci pool to map the virt space to
+		 * pci bus space for an I/O. The DMA buffer includes the
+		 * number of SGE's necessary to support the sg_tablesize.
+		 */
+		lpfc_ncmd->data = pci_pool_alloc(phba->lpfc_sg_dma_buf_pool,
+						 GFP_KERNEL,
+						 &lpfc_ncmd->dma_handle);
+		if (!lpfc_ncmd->data) {
+			kfree(lpfc_ncmd);
+			break;
+		}
+		memset(lpfc_ncmd->data, 0, phba->cfg_sg_dma_buf_size);
+
+		lxri = lpfc_sli4_next_xritag(phba);
+		if (lxri == NO_XRI) {
+			pci_pool_free(phba->lpfc_sg_dma_buf_pool,
+				      lpfc_ncmd->data, lpfc_ncmd->dma_handle);
+			kfree(lpfc_ncmd);
+			break;
+		}
+		pwqeq = &(lpfc_ncmd->cur_iocbq);
+		wqe = (union lpfc_wqe128 *)&pwqeq->wqe;
+
+		/* Allocate iotag for lpfc_ncmd->cur_iocbq. */
+		iotag = lpfc_sli_next_iotag(phba, pwqeq);
+		if (iotag == 0) {
+			pci_pool_free(phba->lpfc_sg_dma_buf_pool,
+				      lpfc_ncmd->data, lpfc_ncmd->dma_handle);
+			kfree(lpfc_ncmd);
+			lpfc_printf_log(phba, KERN_ERR, LOG_NVME_IOERR,
+					"6121 Failed to allocated IOTAG for"
+					" XRI:0x%x\n", lxri);
+			lpfc_sli4_free_xri(phba, lxri);
+			break;
+		}
+		pwqeq->sli4_lxritag = lxri;
+		pwqeq->sli4_xritag = phba->sli4_hba.xri_ids[lxri];
+		pwqeq->iocb_flag |= LPFC_IO_NVME;
+		pwqeq->context1 = lpfc_ncmd;
+		pwqeq->wqe_cmpl = lpfc_nvme_io_cmd_wqe_cmpl;
+
+		/* Initialize local short-hand pointers. */
+		lpfc_ncmd->nvme_sgl = lpfc_ncmd->data;
+		sgl = lpfc_ncmd->nvme_sgl;
+		pdma_phys_sgl = lpfc_ncmd->dma_handle;
+		lpfc_ncmd->dma_phys_sgl = pdma_phys_sgl;
+
+		/* Rsp SGE will be filled in when we rcv an IO
+		 * from the NVME Layer to be sent.
+		 * The cmd is going to be embedded so we need a SKIP SGE.
+		 */
+		bf_set(lpfc_sli4_sge_type, sgl, LPFC_SGE_TYPE_SKIP);
+		bf_set(lpfc_sli4_sge_last, sgl, 0);
+		sgl->word2 = cpu_to_le32(sgl->word2);
+		/* Fill in word 3 / sgl_len during cmd submission */
+
+		lpfc_ncmd->cur_iocbq.context1 = lpfc_ncmd;
+
+		/* Word 7 */
+		bf_set(wqe_erp, &wqe->generic.wqe_com, 0);
+		/* NVME upper layers will time things out, if needed */
+		bf_set(wqe_tmo, &wqe->generic.wqe_com, 0);
+
+		/* Word 10 */
+		bf_set(wqe_ebde_cnt, &wqe->generic.wqe_com, 0);
+		bf_set(wqe_dbde, &wqe->generic.wqe_com, 1);
+
+		/* add the nvme buffer to a post list */
+		list_add_tail(&lpfc_ncmd->list, &post_nblist);
+		spin_lock_irq(&phba->nvme_buf_list_get_lock);
+		phba->sli4_hba.nvme_xri_cnt++;
+		spin_unlock_irq(&phba->nvme_buf_list_get_lock);
+	}
+	lpfc_printf_log(phba, KERN_INFO, LOG_NVME,
+			"6114 Allocate %d out of %d requested new NVME "
+			"buffers\n", bcnt, num_to_alloc);
+
+	/* post the list of nvme buffer sgls to port if available */
+	if (!list_empty(&post_nblist))
+		num_posted = lpfc_post_nvme_sgl_list(phba,
+						     &post_nblist, bcnt);
+	else
+		num_posted = 0;
+
+	return num_posted;
+}
+
+/**
+ * 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.
+ *
+ * This routine removes a nvme buffer from head of @phba lpfc_nvme_buf_list list
+ * and returns to caller.
+ *
+ * Return codes:
+ *   NULL - Error
+ *   Pointer to lpfc_nvme_buf - Success
+ **/
+static struct lpfc_nvme_buf *
+lpfc_get_nvme_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp)
+{
+	struct lpfc_nvme_buf *lpfc_ncmd, *lpfc_ncmd_next;
+	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) {
+		if (lpfc_test_rrq_active(phba, ndlp,
+					 lpfc_ncmd->cur_iocbq.sli4_lxritag))
+			continue;
+		list_del(&lpfc_ncmd->list);
+		found = 1;
+		break;
+	}
+	if (!found) {
+		spin_lock(&phba->nvme_buf_list_put_lock);
+		list_splice(&phba->lpfc_nvme_buf_list_put,
+			    &phba->lpfc_nvme_buf_list_get);
+		INIT_LIST_HEAD(&phba->lpfc_nvme_buf_list_put);
+		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) {
+			if (lpfc_test_rrq_active(
+				phba, ndlp, lpfc_ncmd->cur_iocbq.sli4_lxritag))
+				continue;
+			list_del(&lpfc_ncmd->list);
+			found = 1;
+			break;
+		}
+	}
+	spin_unlock_irqrestore(&phba->nvme_buf_list_get_lock, iflag);
+	if (!found)
+		return NULL;
+	return  lpfc_ncmd;
+}
+
+/**
+ * lpfc_release_nvme_buf: Return a nvme buffer back to hba nvme buf list.
+ * @phba: The Hba for which this call is being executed.
+ * @lpfc_ncmd: The nvme buffer which is being released.
+ *
+ * This routine releases @lpfc_ncmd nvme buffer by adding it to tail of @phba
+ * lpfc_nvme_buf_list list. For SLI4 XRI's are tied to the nvme buffer
+ * and cannot be reused for at least RA_TOV amount of time if it was
+ * aborted.
+ **/
+static void
+lpfc_release_nvme_buf(struct lpfc_hba *phba, struct lpfc_nvme_buf *lpfc_ncmd)
+{
+	unsigned long iflag = 0;
+
+	lpfc_ncmd->nonsg_phys = 0;
+	if (lpfc_ncmd->exch_busy) {
+		spin_lock_irqsave(&phba->sli4_hba.abts_nvme_buf_list_lock,
+					iflag);
+		lpfc_ncmd->nvmeCmd = NULL;
+		list_add_tail(&lpfc_ncmd->list,
+			&phba->sli4_hba.lpfc_abts_nvme_buf_list);
+		spin_unlock_irqrestore(&phba->sli4_hba.abts_nvme_buf_list_lock,
+					iflag);
+	} else {
+		lpfc_ncmd->nvmeCmd = NULL;
+		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);
+		spin_unlock_irqrestore(&phba->nvme_buf_list_put_lock, iflag);
+	}
+}
+
+/**
+ * lpfc_nvme_create_localport - Create/Bind an nvme localport instance.
+ * @pvport - the lpfc_vport instance requesting a localport.
+ *
+ * This routine is invoked to create an nvme localport instance to bind
+ * to the nvme_fc_transport.  It is called once during driver load
+ * like lpfc_create_shost after all other services are initialized.
+ * It requires a vport, vpi, and wwns at call time.  Other localport
+ * parameters are modified as the driver's FCID and the Fabric WWN
+ * are established.
+ *
+ * Return codes
+ *      0 - successful
+ *      -ENOMEM - no heap memory available
+ *      other values - from nvme registration upcall
+ **/
+int
+lpfc_nvme_create_localport(struct lpfc_vport *vport)
+{
+	struct lpfc_hba  *phba = vport->phba;
+	struct nvme_fc_port_info nfcp_info;
+	struct nvme_fc_local_port *localport;
+	struct lpfc_nvme_lport *lport;
+	int len, ret = 0;
+
+	/* Initialize this localport instance.  The vport wwn usage ensures
+	 * that NPIV is accounted for.
+	 */
+	memset(&nfcp_info, 0, sizeof(struct nvme_fc_port_info));
+	nfcp_info.port_role = FC_PORT_ROLE_NVME_INITIATOR;
+	nfcp_info.node_name = wwn_to_u64(vport->fc_nodename.u.wwn);
+	nfcp_info.port_name = wwn_to_u64(vport->fc_portname.u.wwn);
+
+	/* For now need + 1 to get around NVME transport logic */
+	lpfc_nvme_template.max_sgl_segments = phba->cfg_sg_seg_cnt + 1;
+	lpfc_nvme_template.max_hw_queues = phba->cfg_nvme_max_hw_queue;
+
+	/* localport is allocated from the stack, but the registration
+	 * call allocates heap memory as well as the private area.
+	 */
+	ret = nvme_fc_register_localport(&nfcp_info, &lpfc_nvme_template,
+					 &vport->phba->pcidev->dev, &localport);
+	if (!ret) {
+		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME | LOG_NVME_DISC,
+				 "6005 Successfully registered local "
+				 "NVME port num %d, localP %p, private %p, "
+				 "sg_seg %d\n",
+				 localport->port_num, localport,
+				 localport->private,
+				 lpfc_nvme_template.max_sgl_segments);
+
+		/* Private is our lport size declared in the template. */
+		lport = (struct lpfc_nvme_lport *)localport->private;
+		vport->localport = localport;
+		lport->vport = vport;
+		INIT_LIST_HEAD(&lport->rport_list);
+		vport->nvmei_support = 1;
+	}
+
+	len  = lpfc_new_nvme_buf(vport, phba->cfg_nvme_posted_buf);
+	vport->phba->total_nvme_bufs += len;
+	return ret;
+}
+
+/**
+ * lpfc_nvme_destroy_localport - Destroy lpfc_nvme bound to nvme transport.
+ * @pnvme: pointer to lpfc nvme data structure.
+ *
+ * This routine is invoked to destroy all lports bound to the phba.
+ * The lport memory was allocated by the nvme fc transport and is
+ * released there.  This routine ensures all rports bound to the
+ * lport have been disconnected.
+ *
+ **/
+void
+lpfc_nvme_destroy_localport(struct lpfc_vport *vport)
+{
+	struct nvme_fc_local_port *localport;
+	struct lpfc_nvme_lport *lport;
+	struct lpfc_nvme_rport *rport;
+	int ret;
+
+	if (vport->nvmei_support == 0)
+		return;
+
+	localport = vport->localport;
+	vport->localport = NULL;
+	lport = (struct lpfc_nvme_lport *)localport->private;
+
+	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
+			 "6011 Destroying NVME localport %p\n",
+			 localport);
+
+	list_for_each_entry(rport, &lport->rport_list, list) {
+		init_completion(&rport->rport_unreg_done);
+		ret = nvme_fc_unregister_remoteport(rport->remoteport);
+		if (ret)
+			lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME_DISC,
+					 "6008 rport fail destroy %x\n", ret);
+		wait_for_completion(&rport->rport_unreg_done);
+	}
+	/* lport's rport list is clear.  Unregister
+	 * lport and release resources.
+	 */
+	init_completion(&lport->lport_unreg_done);
+	ret = nvme_fc_unregister_localport(localport);
+	wait_for_completion(&lport->lport_unreg_done);
+
+	/* Regardless of the unregister upcall response, clear
+	 * nvmei_support.  All rports are unregistered and the
+	 * driver will clean up.
+	 */
+	vport->nvmei_support = 0;
+	if (ret == 0) {
+		lpfc_printf_vlog(vport,
+				 KERN_INFO, LOG_NVME_DISC,
+				 "6009 Unregistered lport Success\n");
+	} else {
+		lpfc_printf_vlog(vport,
+				 KERN_INFO, LOG_NVME_DISC,
+				 "6010 Unregistered lport "
+				 "Failed, status x%x\n",
+				 ret);
+	}
+}
+
+void
+lpfc_nvme_update_localport(struct lpfc_vport *vport)
+{
+	struct nvme_fc_local_port *localport;
+	struct lpfc_nvme_lport *lport;
+
+	localport = vport->localport;
+	lport = (struct lpfc_nvme_lport *)localport->private;
+
+	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME,
+			 "6012 Update NVME lport %p did x%x\n",
+			 localport, vport->fc_myDID);
+
+	localport->port_id = vport->fc_myDID;
+	if (localport->port_id == 0)
+		localport->port_role = FC_PORT_ROLE_NVME_DISCOVERY;
+	else
+		localport->port_role = FC_PORT_ROLE_NVME_INITIATOR;
+
+	lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
+			 "6030 bound lport %p to DID x%06x\n",
+			 lport, localport->port_id);
+
+}
+
+int
+lpfc_nvme_register_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp)
+{
+	int ret = 0;
+	struct nvme_fc_local_port *localport;
+	struct lpfc_nvme_lport *lport;
+	struct lpfc_nvme_rport *rport;
+	struct nvme_fc_remote_port *remote_port;
+	struct nvme_fc_port_info rpinfo;
+
+	lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
+			 "6006 Register NVME PORT. DID x%06x nlptype x%x\n",
+			 ndlp->nlp_DID, ndlp->nlp_type);
+
+	localport = vport->localport;
+	lport = (struct lpfc_nvme_lport *)localport->private;
+
+	if (ndlp->nlp_type & (NLP_NVME_TARGET | NLP_NVME_INITIATOR)) {
+
+		/* The driver isn't expecting the rport wwn to change
+		 * but it might get a different DID on a different
+		 * fabric.
+		 */
+		list_for_each_entry(rport, &lport->rport_list, list) {
+			if (rport->remoteport->port_name !=
+			    wwn_to_u64(ndlp->nlp_portname.u.wwn))
+				continue;
+			lpfc_printf_vlog(ndlp->vport, KERN_INFO, LOG_NVME_DISC,
+					 "6035 lport %p, found matching rport "
+					 "at wwpn 0x%llx, Data: x%x x%x x%x "
+					 "x%06x\n",
+					 lport,
+					 rport->remoteport->port_name,
+					 rport->remoteport->port_id,
+					 rport->remoteport->port_role,
+					 ndlp->nlp_type,
+					 ndlp->nlp_DID);
+			remote_port = rport->remoteport;
+			if ((remote_port->port_id == 0) &&
+			    (remote_port->port_role ==
+			     FC_PORT_ROLE_NVME_DISCOVERY)) {
+				remote_port->port_id = ndlp->nlp_DID;
+				remote_port->port_role &=
+					~FC_PORT_ROLE_NVME_DISCOVERY;
+				if (ndlp->nlp_type & NLP_NVME_TARGET)
+					remote_port->port_role |=
+						FC_PORT_ROLE_NVME_TARGET;
+				if (ndlp->nlp_type & NLP_NVME_INITIATOR)
+					remote_port->port_role |=
+						FC_PORT_ROLE_NVME_INITIATOR;
+
+				lpfc_printf_vlog(ndlp->vport, KERN_INFO,
+						 LOG_NVME_DISC,
+						 "6014 Rebinding lport to "
+						 "rport wwpn 0x%llx, "
+						 "Data: x%x x%x x%x x%06x\n",
+						 remote_port->port_name,
+						 remote_port->port_id,
+						 remote_port->port_role,
+						 ndlp->nlp_type,
+						 ndlp->nlp_DID);
+			}
+			return 0;
+		}
+
+		/* NVME rports are not preserved across devloss.
+		 * Just register this instance.
+		 */
+		rpinfo.port_id = ndlp->nlp_DID;
+		rpinfo.port_role = 0;
+		if (ndlp->nlp_type & NLP_NVME_TARGET)
+			rpinfo.port_role |= FC_PORT_ROLE_NVME_TARGET;
+		if (ndlp->nlp_type & NLP_NVME_INITIATOR)
+			rpinfo.port_role |= FC_PORT_ROLE_NVME_INITIATOR;
+		rpinfo.port_name = wwn_to_u64(ndlp->nlp_portname.u.wwn);
+		rpinfo.node_name = wwn_to_u64(ndlp->nlp_nodename.u.wwn);
+
+		ret = nvme_fc_register_remoteport(localport, &rpinfo,
+						  &remote_port);
+		if (!ret) {
+			rport = remote_port->private;
+			rport->remoteport = remote_port;
+			rport->lport = lport;
+			rport->ndlp = lpfc_nlp_get(ndlp);
+			if (!rport->ndlp)
+				return -1;
+			ndlp->nrport = rport;
+			INIT_LIST_HEAD(&rport->list);
+			list_add_tail(&rport->list, &lport->rport_list);
+			lpfc_printf_vlog(vport, KERN_INFO,
+					 LOG_NVME_DISC | LOG_NODE,
+					 "6022 Binding new rport to lport %p "
+					 "Rport WWNN 0x%llx, Rport WWPN 0x%llx "
+					 "DID x%06x Role x%x\n",
+					 lport,
+					 rpinfo.node_name, rpinfo.port_name,
+					 rpinfo.port_id, rpinfo.port_role);
+		} else {
+			lpfc_printf_vlog(vport, KERN_ERR,
+					 LOG_NVME_DISC | LOG_NODE,
+					 "6031 RemotePort Registration failed "
+					 "err: %d, DID x%06x\n",
+					 ret, ndlp->nlp_DID);
+		}
+	} else {
+		ret = -EINVAL;
+		lpfc_printf_vlog(vport, KERN_INFO, LOG_NVME_DISC,
+				 "6027 Unknown nlp_type x%x on DID x%06x "
+				 "ndlp %p.  Not Registering nvme rport\n",
+				 ndlp->nlp_type, ndlp->nlp_DID, ndlp);
+	}
+	return ret;
+}
+
+/* lpfc_nvme_unregister_port - unbind the DID and port_role from this rport.
+ *
+ * There is no notion of Devloss or rport recovery from the current
+ * nvme_transport perspective.  Loss of an rport just means IO cannot
+ * be sent and recovery is completely up to the initator.
+ * For now, the driver just unbinds the DID and port_role so that
+ * no further IO can be issued.  Changes are planned for later.
+ *
+ * Notes - the ndlp reference count is not decremented here since
+ * since there is no nvme_transport api for devloss.  Node ref count
+ * is only adjusted in driver unload.
+ */
+void
+lpfc_nvme_unregister_port(struct lpfc_nodelist *ndlp)
+{
+	struct lpfc_vport *vport = ndlp->vport;
+	struct nvme_fc_local_port *localport = NULL;
+	struct lpfc_nvme_lport *lport = NULL;
+	struct lpfc_nvme_rport *rport = NULL;
+
+	localport = vport->localport;
+	lport = (struct lpfc_nvme_lport *)localport->private;
+
+	if (ndlp->nlp_type & (NLP_NVME_TARGET | NLP_NVME_INITIATOR)) {
+		/* The rports are not expected to change WWNs but
+		 * but the port_id and fabric name could be different
+		 * during recovery.  For NPIV-based rports that do
+		 * recover with new WWNs, a new rport should get
+		 * registered and added to the nvme_transport.
+		 */
+
+		list_for_each_entry(rport, &lport->rport_list, list) {
+			if (rport->remoteport->port_name !=
+			    wwn_to_u64(ndlp->nlp_portname.u.wwn))
+				continue;
+
+			rport->remoteport->port_id = 0;
+			rport->remoteport->port_role =
+				FC_PORT_ROLE_NVME_DISCOVERY;
+			lpfc_printf_vlog(vport,
+					 KERN_INFO, LOG_NVME_DISC,
+					 "6033 Unreg rport %p DID x%06x\n",
+					 rport, ndlp->nlp_DID);
+		}
+	}
+}
diff --git a/drivers/scsi/lpfc/lpfc_nvme.h b/drivers/scsi/lpfc/lpfc_nvme.h
index b5d9504..6e44950 100644
--- a/drivers/scsi/lpfc/lpfc_nvme.h
+++ b/drivers/scsi/lpfc/lpfc_nvme.h
@@ -28,6 +28,12 @@
 
 #define LPFC_NVME_ERSP_LEN		0x20
 
+struct lpfc_nvme_qhandle {
+	uint32_t index;		/* WQ index to use */
+	uint32_t qidx;		/* queue index passed to create */
+	uint32_t cpu_id;	/* current cpu id at time of create */
+};
+
 /* Declare nvme-based local and remote port definitions. */
 struct lpfc_nvme_lport {
 	struct lpfc_vport *vport;
diff --git a/drivers/scsi/lpfc/lpfc_sli.c b/drivers/scsi/lpfc/lpfc_sli.c
index 076810d..446b6f5 100644
--- a/drivers/scsi/lpfc/lpfc_sli.c
+++ b/drivers/scsi/lpfc/lpfc_sli.c
@@ -6807,6 +6807,22 @@ lpfc_sli4_hba_setup(struct lpfc_hba *phba)
 		}
 	}
 
+	if ((phba->cfg_enable_fc4_type & LPFC_ENABLE_NVME) &&
+	    (phba->nvmet_support == 0)) {
+
+		/* register the allocated nvme sgl pool to the port */
+		rc = lpfc_repost_nvme_sgl_list(phba);
+		if (unlikely(rc)) {
+			lpfc_printf_log(phba, KERN_ERR, LOG_MBOX | LOG_SLI,
+					"6116 Error %d during nvme sgl post "
+					"operation\n", rc);
+			/* Some NVME buffers were moved to abort nvme list */
+			/* A pci function reset will repost them */
+			rc = -ENODEV;
+			goto out_destroy_queue;
+		}
+	}
+
 	/* Post the rpi header region to the device. */
 	rc = lpfc_sli4_post_all_rpi_hdrs(phba);
 	if (unlikely(rc)) {
@@ -10514,10 +10530,7 @@ lpfc_sli4_abort_nvme_io(struct lpfc_hba *phba, struct lpfc_sli_ring *pring,
 	/* ABTS WQE must go to the same WQ as the WQE to be aborted */
 	abtsiocbp->iocb_flag |= LPFC_IO_NVME;
 	abtsiocbp->vport = vport;
-	/* todo: assign wqe_cmpl to lpfc_nvme_abort_fcreq_cmpl
-	 * subsequent patch will add routine. For now, just skip assignment
-	 * as won't ever be called.
-	 */
+	abtsiocbp->wqe_cmpl = lpfc_nvme_abort_fcreq_cmpl;
 	retval = lpfc_sli4_issue_wqe(phba, LPFC_FCP_RING, abtsiocbp);
 	if (retval == IOCB_ERROR) {
 		lpfc_printf_vlog(vport, KERN_ERR, LOG_NVME,
diff --git a/drivers/scsi/lpfc/lpfc_sli4.h b/drivers/scsi/lpfc/lpfc_sli4.h
index 0a56fc6..0bfb61d 100644
--- a/drivers/scsi/lpfc/lpfc_sli4.h
+++ b/drivers/scsi/lpfc/lpfc_sli4.h
@@ -754,6 +754,7 @@ int lpfc_sli4_queue_setup(struct lpfc_hba *);
 void lpfc_sli4_queue_unset(struct lpfc_hba *);
 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 *);
+int lpfc_repost_nvme_sgl_list(struct lpfc_hba *phba);
 uint16_t lpfc_sli4_next_xritag(struct lpfc_hba *);
 void lpfc_sli4_free_xri(struct lpfc_hba *, int);
 int lpfc_sli4_post_async_mbox(struct lpfc_hba *);
diff --git a/drivers/scsi/lpfc/lpfc_vport.c b/drivers/scsi/lpfc/lpfc_vport.c
index 07c7c5f..b0d94f2 100644
--- a/drivers/scsi/lpfc/lpfc_vport.c
+++ b/drivers/scsi/lpfc/lpfc_vport.c
@@ -403,7 +403,21 @@ lpfc_vport_create(struct fc_vport *fc_vport, bool disable)
 		vport->fdmi_port_mask = phba->pport->fdmi_port_mask;
 	}
 
-	/* todo: init: register port with nvme */
+	if ((phba->nvmet_support == 0) &&
+	    ((phba->cfg_enable_fc4_type == LPFC_ENABLE_BOTH) ||
+	     (phba->cfg_enable_fc4_type == LPFC_ENABLE_NVME))) {
+		/* Create NVME binding with nvme_fc_transport. This
+		 * ensures the vport is initialized.
+		 */
+		rc = lpfc_nvme_create_localport(vport);
+		if (rc) {
+			lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
+					"6003 %s status x%x\n",
+					"NVME registration failed, ",
+					rc);
+			goto error_out;
+		}
+	}
 
 	/*
 	 * In SLI4, the vpi must be activated before it can be used
-- 
2.5.0




More information about the Linux-nvme mailing list