Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi

Pull SCSI updates from James Bottomley:
 "This is mostly update of the usual drivers: arcmsr, qla2xxx, lpfc,
  hisi_sas, target/iscsi and target/core.

  Additionally Christoph refactored gdth as part of the dma changes. The
  major mid-layer change this time is the removal of bidi commands and
  with them the whole of the osd/exofs driver and filesystem. This is a
  major simplification for block and mq in particular"

* tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi: (240 commits)
  scsi: cxgb4i: validate tcp sequence number only if chip version <= T5
  scsi: cxgb4i: get pf number from lldi->pf
  scsi: core: replace GFP_ATOMIC with GFP_KERNEL in scsi_scan.c
  scsi: mpt3sas: Add missing breaks in switch statements
  scsi: aacraid: Fix missing break in switch statement
  scsi: kill command serial number
  scsi: csiostor: drop serial_number usage
  scsi: mvumi: use request tag instead of serial_number
  scsi: dpt_i2o: remove serial number usage
  scsi: st: osst: Remove negative constant left-shifts
  scsi: ufs-bsg: Allow reading descriptors
  scsi: ufs: Allow reading descriptor via raw upiu
  scsi: ufs-bsg: Change the calling convention for write descriptor
  scsi: ufs: Remove unused device quirks
  Revert "scsi: ufs: disable vccq if it's not needed by UFS device"
  scsi: megaraid_sas: Remove a bunch of set but not used variables
  scsi: clean obsolete return values of eh_timed_out
  scsi: sd: Optimal I/O size should be a multiple of physical block size
  scsi: MAINTAINERS: SCSI initiator and target tweaks
  scsi: fcoe: make use of fip_mode enum complete
  ...

18 files changed, 1134 insertions, 516 deletions

diff --git a/drivers/scsi/qla2xxx/qla_attr.c b/drivers/scsi/qla2xxx/qla_attr.c
index ac504a1ff0ff..2eb1ae721a7d 100644
--- a/drivers/scsi/qla2xxx/qla_attr.c
+++ b/drivers/scsi/qla2xxx/qla_attr.c
@@ -543,6 +543,9 @@ qla2x00_sysfs_write_vpd(struct file *filp, struct kobject *kobj,
 	if (unlikely(pci_channel_offline(ha->pdev)))
 		return 0;
 
+	if (qla2x00_chip_is_down(vha))
+		return 0;
+
 	if (!capable(CAP_SYS_ADMIN) || off != 0 || count != ha->vpd_size ||
 	    !ha->isp_ops->write_nvram)
 		return 0;
@@ -1002,7 +1005,7 @@ qla2x00_free_sysfs_attr(scsi_qla_host_t *vha, bool stop_beacon)
 /* Scsi_Host attributes. */
 
 static ssize_t
-qla2x00_drvr_version_show(struct device *dev,
+qla2x00_driver_version_show(struct device *dev,
 			  struct device_attribute *attr, char *buf)
 {
 	return scnprintf(buf, PAGE_SIZE, "%s\n", qla2x00_version_str);
@@ -1632,6 +1635,92 @@ qla2x00_max_speed_sup_show(struct device *dev, struct device_attribute *attr,
 	    ha->max_speed_sup ? "32Gps" : "16Gps");
 }
 
+static ssize_t
+qla2x00_port_speed_store(struct device *dev, struct device_attribute *attr,
+    const char *buf, size_t count)
+{
+	struct scsi_qla_host *vha = shost_priv(dev_to_shost(dev));
+	ulong type, speed;
+	int oldspeed, rval;
+	int mode = QLA_SET_DATA_RATE_LR;
+	struct qla_hw_data *ha = vha->hw;
+
+	if (!IS_QLA27XX(vha->hw)) {
+		ql_log(ql_log_warn, vha, 0x70d8,
+		    "Speed setting not supported \n");
+		return -EINVAL;
+	}
+
+	rval = kstrtol(buf, 10, &type);
+	speed = type;
+	if (type == 40 || type == 80 || type == 160 ||
+	    type == 320) {
+		ql_dbg(ql_dbg_user, vha, 0x70d9,
+		    "Setting will be affected after a loss of sync\n");
+		type = type/10;
+		mode = QLA_SET_DATA_RATE_NOLR;
+	}
+
+	oldspeed = ha->set_data_rate;
+
+	switch (type) {
+	case 0:
+		ha->set_data_rate = PORT_SPEED_AUTO;
+		break;
+	case 4:
+		ha->set_data_rate = PORT_SPEED_4GB;
+		break;
+	case 8:
+		ha->set_data_rate = PORT_SPEED_8GB;
+		break;
+	case 16:
+		ha->set_data_rate = PORT_SPEED_16GB;
+		break;
+	case 32:
+		ha->set_data_rate = PORT_SPEED_32GB;
+		break;
+	default:
+		ql_log(ql_log_warn, vha, 0x1199,
+		    "Unrecognized speed setting:%lx. Setting Autoneg\n",
+		    speed);
+		ha->set_data_rate = PORT_SPEED_AUTO;
+	}
+
+	if (qla2x00_chip_is_down(vha) || (oldspeed == ha->set_data_rate))
+		return -EINVAL;
+
+	ql_log(ql_log_info, vha, 0x70da,
+	    "Setting speed to %lx Gbps \n", type);
+
+	rval = qla2x00_set_data_rate(vha, mode);
+	if (rval != QLA_SUCCESS)
+		return -EIO;
+
+	return strlen(buf);
+}
+
+static ssize_t
+qla2x00_port_speed_show(struct device *dev, struct device_attribute *attr,
+    char *buf)
+{
+	struct scsi_qla_host *vha = shost_priv(dev_to_shost(dev));
+	struct qla_hw_data *ha = vha->hw;
+	ssize_t rval;
+	char *spd[7] = {"0", "0", "0", "4", "8", "16", "32"};
+
+	rval = qla2x00_get_data_rate(vha);
+	if (rval != QLA_SUCCESS) {
+		ql_log(ql_log_warn, vha, 0x70db,
+		    "Unable to get port speed rval:%zd\n", rval);
+		return -EINVAL;
+	}
+
+	ql_log(ql_log_info, vha, 0x70d6,
+	    "port speed:%d\n", ha->link_data_rate);
+
+	return scnprintf(buf, PAGE_SIZE, "%s\n", spd[ha->link_data_rate]);
+}
+
 /* ----- */
 
 static ssize_t
@@ -2059,7 +2148,21 @@ ql2xiniexchg_store(struct device *dev, struct device_attribute *attr,
 	return strlen(buf);
 }
 
-static DEVICE_ATTR(driver_version, S_IRUGO, qla2x00_drvr_version_show, NULL);
+static ssize_t
+qla2x00_dif_bundle_statistics_show(struct device *dev,
+    struct device_attribute *attr, char *buf)
+{
+	scsi_qla_host_t *vha = shost_priv(class_to_shost(dev));
+	struct qla_hw_data *ha = vha->hw;
+
+	return scnprintf(buf, PAGE_SIZE,
+	    "cross=%llu read=%llu write=%llu kalloc=%llu dma_alloc=%llu unusable=%u\n",
+	    ha->dif_bundle_crossed_pages, ha->dif_bundle_reads,
+	    ha->dif_bundle_writes, ha->dif_bundle_kallocs,
+	    ha->dif_bundle_dma_allocs, ha->pool.unusable.count);
+}
+
+static DEVICE_ATTR(driver_version, S_IRUGO, qla2x00_driver_version_show, NULL);
 static DEVICE_ATTR(fw_version, S_IRUGO, qla2x00_fw_version_show, NULL);
 static DEVICE_ATTR(serial_num, S_IRUGO, qla2x00_serial_num_show, NULL);
 static DEVICE_ATTR(isp_name, S_IRUGO, qla2x00_isp_name_show, NULL);
@@ -2112,6 +2215,10 @@ static DEVICE_ATTR(zio_threshold, 0644,
 static DEVICE_ATTR_RW(qlini_mode);
 static DEVICE_ATTR_RW(ql2xexchoffld);
 static DEVICE_ATTR_RW(ql2xiniexchg);
+static DEVICE_ATTR(dif_bundle_statistics, 0444,
+    qla2x00_dif_bundle_statistics_show, NULL);
+static DEVICE_ATTR(port_speed, 0644, qla2x00_port_speed_show,
+    qla2x00_port_speed_store);
 
 
 struct device_attribute *qla2x00_host_attrs[] = {
@@ -2150,6 +2257,8 @@ struct device_attribute *qla2x00_host_attrs[] = {
 	&dev_attr_min_link_speed,
 	&dev_attr_max_speed_sup,
 	&dev_attr_zio_threshold,
+	&dev_attr_dif_bundle_statistics,
+	&dev_attr_port_speed,
 	NULL, /* reserve for qlini_mode */
 	NULL, /* reserve for ql2xiniexchg */
 	NULL, /* reserve for ql2xexchoffld */
diff --git a/drivers/scsi/qla2xxx/qla_def.h b/drivers/scsi/qla2xxx/qla_def.h
index d1fc4958222a..3d46975a5e5c 100644
--- a/drivers/scsi/qla2xxx/qla_def.h
+++ b/drivers/scsi/qla2xxx/qla_def.h
@@ -314,6 +314,7 @@ struct srb_cmd {
 #define SRB_CRC_PROT_DMA_VALID		BIT_4	/* DIF: prot DMA valid */
 #define SRB_CRC_CTX_DSD_VALID		BIT_5	/* DIF: dsd_list valid */
 #define SRB_WAKEUP_ON_COMP		BIT_6
+#define SRB_DIF_BUNDL_DMA_VALID		BIT_7   /* DIF: DMA list valid */
 
 /* To identify if a srb is of T10-CRC type. @sp => srb_t pointer */
 #define IS_PROT_IO(sp)	(sp->flags & SRB_CRC_CTX_DSD_VALID)
@@ -1892,6 +1893,13 @@ struct crc_context {
 	/* List of DMA context transfers */
 	struct list_head dsd_list;
 
+	/* List of DIF Bundling context DMA address */
+	struct list_head ldif_dsd_list;
+	u8 no_ldif_dsd;
+
+	struct list_head ldif_dma_hndl_list;
+	u32 dif_bundl_len;
+	u8 no_dif_bundl;
 	/* This structure should not exceed 512 bytes */
 };
 
@@ -2359,7 +2367,9 @@ typedef struct fc_port {
 #define NVME_PRLI_SP_INITIATOR  BIT_5
 #define NVME_PRLI_SP_TARGET     BIT_4
 #define NVME_PRLI_SP_DISCOVERY  BIT_3
+#define NVME_PRLI_SP_FIRST_BURST	BIT_0
 	uint8_t nvme_flag;
+	uint32_t nvme_first_burst_size;
 #define NVME_FLAG_REGISTERED 4
 #define NVME_FLAG_DELETING 2
 #define NVME_FLAG_RESETTING 1
@@ -3688,12 +3698,14 @@ struct qla_hw_data {
 #define PORT_SPEED_UNKNOWN 0xFFFF
 #define PORT_SPEED_1GB  0x00
 #define PORT_SPEED_2GB  0x01
+#define PORT_SPEED_AUTO 0x02
 #define PORT_SPEED_4GB  0x03
 #define PORT_SPEED_8GB  0x04
 #define PORT_SPEED_16GB 0x05
 #define PORT_SPEED_32GB 0x06
 #define PORT_SPEED_10GB	0x13
 	uint16_t	link_data_rate;         /* F/W operating speed */
+	uint16_t	set_data_rate;		/* Set by user */
 
 	uint8_t		current_topology;
 	uint8_t		prev_topology;
@@ -3958,6 +3970,10 @@ struct qla_hw_data {
 	uint16_t	fw_subminor_version;
 	uint16_t	fw_attributes;
 	uint16_t	fw_attributes_h;
+#define FW_ATTR_H_NVME_FBURST 	BIT_1
+#define FW_ATTR_H_NVME		BIT_10
+#define FW_ATTR_H_NVME_UPDATED  BIT_14
+
 	uint16_t	fw_attributes_ext[2];
 	uint32_t	fw_memory_size;
 	uint32_t	fw_transfer_size;
@@ -4184,12 +4200,32 @@ struct qla_hw_data {
 	uint16_t min_link_speed;
 	uint16_t max_speed_sup;
 
+	/* DMA pool for the DIF bundling buffers */
+	struct dma_pool *dif_bundl_pool;
+	#define DIF_BUNDLING_DMA_POOL_SIZE  1024
+	struct {
+		struct {
+			struct list_head head;
+			uint count;
+		} good;
+		struct {
+			struct list_head head;
+			uint count;
+		} unusable;
+	} pool;
+
+	unsigned long long dif_bundle_crossed_pages;
+	unsigned long long dif_bundle_reads;
+	unsigned long long dif_bundle_writes;
+	unsigned long long dif_bundle_kallocs;
+	unsigned long long dif_bundle_dma_allocs;
+
 	atomic_t        nvme_active_aen_cnt;
 	uint16_t        nvme_last_rptd_aen;             /* Last recorded aen count */
 
 	atomic_t zio_threshold;
 	uint16_t last_zio_threshold;
-#define DEFAULT_ZIO_THRESHOLD 64
+#define DEFAULT_ZIO_THRESHOLD 5
 };
 
 #define FW_ABILITY_MAX_SPEED_MASK	0xFUL
@@ -4198,6 +4234,10 @@ struct qla_hw_data {
 #define FW_ABILITY_MAX_SPEED(ha)	\
 	(ha->fw_ability_mask & FW_ABILITY_MAX_SPEED_MASK)
 
+#define QLA_GET_DATA_RATE	0
+#define QLA_SET_DATA_RATE_NOLR	1
+#define QLA_SET_DATA_RATE_LR	2 /* Set speed and initiate LR */
+
 /*
  * Qlogic scsi host structure
  */
@@ -4229,6 +4269,7 @@ typedef struct scsi_qla_host {
 		uint32_t	qpairs_req_created:1;
 		uint32_t	qpairs_rsp_created:1;
 		uint32_t	nvme_enabled:1;
+		uint32_t        nvme_first_burst:1;
 	} flags;
 
 	atomic_t	loop_state;
diff --git a/drivers/scsi/qla2xxx/qla_dfs.c b/drivers/scsi/qla2xxx/qla_dfs.c
index 0b190082aa8d..ead17288e2a7 100644
--- a/drivers/scsi/qla2xxx/qla_dfs.c
+++ b/drivers/scsi/qla2xxx/qla_dfs.c
@@ -446,11 +446,6 @@ qla2x00_dfs_setup(scsi_qla_host_t *vha)
 
 	atomic_set(&qla2x00_dfs_root_count, 0);
 	qla2x00_dfs_root = debugfs_create_dir(QLA2XXX_DRIVER_NAME, NULL);
-	if (!qla2x00_dfs_root) {
-		ql_log(ql_log_warn, vha, 0x00f7,
-		    "Unable to create debugfs root directory.\n");
-		goto out;
-	}
 
 create_dir:
 	if (ha->dfs_dir)
@@ -458,64 +453,28 @@ create_dir:
 
 	mutex_init(&ha->fce_mutex);
 	ha->dfs_dir = debugfs_create_dir(vha->host_str, qla2x00_dfs_root);
-	if (!ha->dfs_dir) {
-		ql_log(ql_log_warn, vha, 0x00f8,
-		    "Unable to create debugfs ha directory.\n");
-		goto out;
-	}
 
 	atomic_inc(&qla2x00_dfs_root_count);
 
 create_nodes:
 	ha->dfs_fw_resource_cnt = debugfs_create_file("fw_resource_count",
 	    S_IRUSR, ha->dfs_dir, vha, &dfs_fw_resource_cnt_ops);
-	if (!ha->dfs_fw_resource_cnt) {
-		ql_log(ql_log_warn, vha, 0x00fd,
-		    "Unable to create debugFS fw_resource_count node.\n");
-		goto out;
-	}
 
 	ha->dfs_tgt_counters = debugfs_create_file("tgt_counters", S_IRUSR,
 	    ha->dfs_dir, vha, &dfs_tgt_counters_ops);
-	if (!ha->dfs_tgt_counters) {
-		ql_log(ql_log_warn, vha, 0xd301,
-		    "Unable to create debugFS tgt_counters node.\n");
-		goto out;
-	}
 
 	ha->tgt.dfs_tgt_port_database = debugfs_create_file("tgt_port_database",
 	    S_IRUSR,  ha->dfs_dir, vha, &dfs_tgt_port_database_ops);
-	if (!ha->tgt.dfs_tgt_port_database) {
-		ql_log(ql_log_warn, vha, 0xd03f,
-		    "Unable to create debugFS tgt_port_database node.\n");
-		goto out;
-	}
 
 	ha->dfs_fce = debugfs_create_file("fce", S_IRUSR, ha->dfs_dir, vha,
 	    &dfs_fce_ops);
-	if (!ha->dfs_fce) {
-		ql_log(ql_log_warn, vha, 0x00f9,
-		    "Unable to create debugfs fce node.\n");
-		goto out;
-	}
 
 	ha->tgt.dfs_tgt_sess = debugfs_create_file("tgt_sess",
 		S_IRUSR, ha->dfs_dir, vha, &dfs_tgt_sess_ops);
-	if (!ha->tgt.dfs_tgt_sess) {
-		ql_log(ql_log_warn, vha, 0xd040,
-		    "Unable to create debugFS tgt_sess node.\n");
-		goto out;
-	}
 
-	if (IS_QLA27XX(ha) || IS_QLA83XX(ha)) {
+	if (IS_QLA27XX(ha) || IS_QLA83XX(ha))
 		ha->tgt.dfs_naqp = debugfs_create_file("naqp",
 		    0400, ha->dfs_dir, vha, &dfs_naqp_ops);
-		if (!ha->tgt.dfs_naqp) {
-			ql_log(ql_log_warn, vha, 0xd011,
-			    "Unable to create debugFS naqp node.\n");
-			goto out;
-		}
-	}
 out:
 	return 0;
 }
diff --git a/drivers/scsi/qla2xxx/qla_gbl.h b/drivers/scsi/qla2xxx/qla_gbl.h
index 3673fcdb033a..4eefe69ca807 100644
--- a/drivers/scsi/qla2xxx/qla_gbl.h
+++ b/drivers/scsi/qla2xxx/qla_gbl.h
@@ -160,6 +160,7 @@ extern int ql2xautodetectsfp;
 extern int ql2xenablemsix;
 extern int qla2xuseresexchforels;
 extern int ql2xexlogins;
+extern int ql2xdifbundlinginternalbuffers;
 
 extern int qla2x00_loop_reset(scsi_qla_host_t *);
 extern void qla2x00_abort_all_cmds(scsi_qla_host_t *, int);
@@ -269,8 +270,8 @@ extern void qla24xx_build_scsi_iocbs(srb_t *, struct cmd_type_7 *,
 	uint16_t, struct req_que *);
 extern int qla2x00_start_scsi(srb_t *sp);
 extern int qla24xx_start_scsi(srb_t *sp);
-int qla2x00_marker(struct scsi_qla_host *, struct req_que *, struct rsp_que *,
-						uint16_t, uint64_t, uint8_t);
+int qla2x00_marker(struct scsi_qla_host *, struct qla_qpair *,
+    uint16_t, uint64_t, uint8_t);
 extern int qla2x00_start_sp(srb_t *);
 extern int qla24xx_dif_start_scsi(srb_t *);
 extern int qla2x00_start_bidir(srb_t *, struct scsi_qla_host *, uint32_t);
@@ -285,7 +286,7 @@ extern int qla24xx_walk_and_build_sglist_no_difb(struct qla_hw_data *, srb_t *,
 extern int qla24xx_walk_and_build_sglist(struct qla_hw_data *, srb_t *,
 	uint32_t *, uint16_t, struct qla_tc_param *);
 extern int qla24xx_walk_and_build_prot_sglist(struct qla_hw_data *, srb_t *,
-	uint32_t *, uint16_t, struct qla_tc_param *);
+	uint32_t *, uint16_t, struct qla_tgt_cmd *);
 extern int qla24xx_get_one_block_sg(uint32_t, struct qla2_sgx *, uint32_t *);
 extern int qla24xx_configure_prot_mode(srb_t *, uint16_t *);
 extern int qla24xx_build_scsi_crc_2_iocbs(srb_t *,
@@ -898,5 +899,6 @@ void qlt_update_host_map(struct scsi_qla_host *, port_id_t);
 void qlt_remove_target_resources(struct qla_hw_data *);
 void qlt_clr_qp_table(struct scsi_qla_host *vha);
 void qlt_set_mode(struct scsi_qla_host *);
+int qla2x00_set_data_rate(scsi_qla_host_t *vha, uint16_t mode);
 
 #endif /* _QLA_GBL_H */
diff --git a/drivers/scsi/qla2xxx/qla_gs.c b/drivers/scsi/qla2xxx/qla_gs.c
index cbc3bc49d4d1..c6fdad12428e 100644
--- a/drivers/scsi/qla2xxx/qla_gs.c
+++ b/drivers/scsi/qla2xxx/qla_gs.c
@@ -657,15 +657,16 @@ static int qla_async_rftid(scsi_qla_host_t *vha, port_id_t *d_id)
 	sp->u.iocb_cmd.timeout = qla2x00_async_iocb_timeout;
 	sp->done = qla2x00_async_sns_sp_done;
 
+	ql_dbg(ql_dbg_disc, vha, 0xffff,
+	    "Async-%s - hdl=%x portid %06x.\n",
+	    sp->name, sp->handle, d_id->b24);
+
 	rval = qla2x00_start_sp(sp);
 	if (rval != QLA_SUCCESS) {
 		ql_dbg(ql_dbg_disc, vha, 0x2043,
 		    "RFT_ID issue IOCB failed (%d).\n", rval);
 		goto done_free_sp;
 	}
-	ql_dbg(ql_dbg_disc, vha, 0xffff,
-	    "Async-%s - hdl=%x portid %06x.\n",
-	    sp->name, sp->handle, d_id->b24);
 	return rval;
 done_free_sp:
 	sp->free(sp);
@@ -752,6 +753,10 @@ static int qla_async_rffid(scsi_qla_host_t *vha, port_id_t *d_id,
 	sp->u.iocb_cmd.timeout = qla2x00_async_iocb_timeout;
 	sp->done = qla2x00_async_sns_sp_done;
 
+	ql_dbg(ql_dbg_disc, vha, 0xffff,
+	    "Async-%s - hdl=%x portid %06x feature %x type %x.\n",
+	    sp->name, sp->handle, d_id->b24, fc4feature, fc4type);
+
 	rval = qla2x00_start_sp(sp);
 	if (rval != QLA_SUCCESS) {
 		ql_dbg(ql_dbg_disc, vha, 0x2047,
@@ -759,9 +764,6 @@ static int qla_async_rffid(scsi_qla_host_t *vha, port_id_t *d_id,
 		goto done_free_sp;
 	}
 
-	ql_dbg(ql_dbg_disc, vha, 0xffff,
-	    "Async-%s - hdl=%x portid %06x feature %x type %x.\n",
-	    sp->name, sp->handle, d_id->b24, fc4feature, fc4type);
 	return rval;
 
 done_free_sp:
@@ -844,15 +846,16 @@ static int qla_async_rnnid(scsi_qla_host_t *vha, port_id_t *d_id,
 	sp->u.iocb_cmd.timeout = qla2x00_async_iocb_timeout;
 	sp->done = qla2x00_async_sns_sp_done;
 
+	ql_dbg(ql_dbg_disc, vha, 0xffff,
+	    "Async-%s - hdl=%x portid %06x\n",
+	    sp->name, sp->handle, d_id->b24);
+
 	rval = qla2x00_start_sp(sp);
 	if (rval != QLA_SUCCESS) {
 		ql_dbg(ql_dbg_disc, vha, 0x204d,
 		    "RNN_ID issue IOCB failed (%d).\n", rval);
 		goto done_free_sp;
 	}
-	ql_dbg(ql_dbg_disc, vha, 0xffff,
-	    "Async-%s - hdl=%x portid %06x\n",
-	    sp->name, sp->handle, d_id->b24);
 
 	return rval;
 
@@ -957,15 +960,16 @@ static int qla_async_rsnn_nn(scsi_qla_host_t *vha)
 	sp->u.iocb_cmd.timeout = qla2x00_async_iocb_timeout;
 	sp->done = qla2x00_async_sns_sp_done;
 
+	ql_dbg(ql_dbg_disc, vha, 0xffff,
+	    "Async-%s - hdl=%x.\n",
+	    sp->name, sp->handle);
+
 	rval = qla2x00_start_sp(sp);
 	if (rval != QLA_SUCCESS) {
 		ql_dbg(ql_dbg_disc, vha, 0x2043,
 		    "RFT_ID issue IOCB failed (%d).\n", rval);
 		goto done_free_sp;
 	}
-	ql_dbg(ql_dbg_disc, vha, 0xffff,
-	    "Async-%s - hdl=%x.\n",
-	    sp->name, sp->handle);
 
 	return rval;
 
@@ -3578,14 +3582,14 @@ int qla24xx_async_gffid(scsi_qla_host_t *vha, fc_port_t *fcport)
 
 	sp->done = qla24xx_async_gffid_sp_done;
 
-	rval = qla2x00_start_sp(sp);
-	if (rval != QLA_SUCCESS)
-		goto done_free_sp;
-
 	ql_dbg(ql_dbg_disc, vha, 0x2132,
 	    "Async-%s hdl=%x  %8phC.\n", sp->name,
 	    sp->handle, fcport->port_name);
 
+	rval = qla2x00_start_sp(sp);
+	if (rval != QLA_SUCCESS)
+		goto done_free_sp;
+
 	return rval;
 done_free_sp:
 	sp->free(sp);
@@ -4067,6 +4071,10 @@ static int qla24xx_async_gnnft(scsi_qla_host_t *vha, struct srb *sp,
 
 	sp->done = qla2x00_async_gpnft_gnnft_sp_done;
 
+	ql_dbg(ql_dbg_disc, vha, 0xffff,
+	    "Async-%s hdl=%x FC4Type %x.\n", sp->name,
+	    sp->handle, ct_req->req.gpn_ft.port_type);
+
 	rval = qla2x00_start_sp(sp);
 	if (rval != QLA_SUCCESS) {
 		spin_lock_irqsave(&vha->work_lock, flags);
@@ -4075,9 +4083,6 @@ static int qla24xx_async_gnnft(scsi_qla_host_t *vha, struct srb *sp,
 		goto done_free_sp;
 	}
 
-	ql_dbg(ql_dbg_disc, vha, 0xffff,
-	    "Async-%s hdl=%x FC4Type %x.\n", sp->name,
-	    sp->handle, ct_req->req.gpn_ft.port_type);
 	return rval;
 
 done_free_sp:
@@ -4158,7 +4163,8 @@ int qla24xx_async_gpnft(scsi_qla_host_t *vha, u8 fc4_type, srb_t *sp)
 			spin_lock_irqsave(&vha->work_lock, flags);
 			vha->scan.scan_flags &= ~SF_SCANNING;
 			spin_unlock_irqrestore(&vha->work_lock, flags);
-			goto done_free_sp;
+			qla2x00_rel_sp(sp);
+			return rval;
 		}
 		sp->u.iocb_cmd.u.ctarg.req_size = GPN_FT_REQ_SIZE;
 
@@ -4177,7 +4183,13 @@ int qla24xx_async_gpnft(scsi_qla_host_t *vha, u8 fc4_type, srb_t *sp)
 			spin_lock_irqsave(&vha->work_lock, flags);
 			vha->scan.scan_flags &= ~SF_SCANNING;
 			spin_unlock_irqrestore(&vha->work_lock, flags);
-			goto done_free_sp;
+			dma_free_coherent(&vha->hw->pdev->dev,
+			    sp->u.iocb_cmd.u.ctarg.req_allocated_size,
+			    sp->u.iocb_cmd.u.ctarg.req,
+			    sp->u.iocb_cmd.u.ctarg.req_dma);
+			sp->u.iocb_cmd.u.ctarg.req = NULL;
+			qla2x00_rel_sp(sp);
+			return rval;
 		}
 		sp->u.iocb_cmd.u.ctarg.rsp_size = rspsz;
 
@@ -4214,6 +4226,10 @@ int qla24xx_async_gpnft(scsi_qla_host_t *vha, u8 fc4_type, srb_t *sp)
 
 	sp->done = qla2x00_async_gpnft_gnnft_sp_done;
 
+	ql_dbg(ql_dbg_disc, vha, 0xffff,
+	    "Async-%s hdl=%x FC4Type %x.\n", sp->name,
+	    sp->handle, ct_req->req.gpn_ft.port_type);
+
 	rval = qla2x00_start_sp(sp);
 	if (rval != QLA_SUCCESS) {
 		spin_lock_irqsave(&vha->work_lock, flags);
@@ -4222,9 +4238,6 @@ int qla24xx_async_gpnft(scsi_qla_host_t *vha, u8 fc4_type, srb_t *sp)
 		goto done_free_sp;
 	}
 
-	ql_dbg(ql_dbg_disc, vha, 0xffff,
-	    "Async-%s hdl=%x FC4Type %x.\n", sp->name,
-	    sp->handle, ct_req->req.gpn_ft.port_type);
 	return rval;
 
 done_free_sp:
@@ -4345,13 +4358,14 @@ int qla24xx_async_gnnid(scsi_qla_host_t *vha, fc_port_t *fcport)
 
 	sp->done = qla2x00_async_gnnid_sp_done;
 
-	rval = qla2x00_start_sp(sp);
-	if (rval != QLA_SUCCESS)
-		goto done_free_sp;
 	ql_dbg(ql_dbg_disc, vha, 0xffff,
 	    "Async-%s - %8phC hdl=%x loopid=%x portid %06x.\n",
 	    sp->name, fcport->port_name,
 	    sp->handle, fcport->loop_id, fcport->d_id.b24);
+
+	rval = qla2x00_start_sp(sp);
+	if (rval != QLA_SUCCESS)
+		goto done_free_sp;
 	return rval;
 
 done_free_sp:
@@ -4475,14 +4489,15 @@ int qla24xx_async_gfpnid(scsi_qla_host_t *vha, fc_port_t *fcport)
 
 	sp->done = qla2x00_async_gfpnid_sp_done;
 
-	rval = qla2x00_start_sp(sp);
-	if (rval != QLA_SUCCESS)
-		goto done_free_sp;
-
 	ql_dbg(ql_dbg_disc, vha, 0xffff,
 	    "Async-%s - %8phC hdl=%x loopid=%x portid %06x.\n",
 	    sp->name, fcport->port_name,
 	    sp->handle, fcport->loop_id, fcport->d_id.b24);
+
+	rval = qla2x00_start_sp(sp);
+	if (rval != QLA_SUCCESS)
+		goto done_free_sp;
+
 	return rval;
 
 done_free_sp:
diff --git a/drivers/scsi/qla2xxx/qla_init.c b/drivers/scsi/qla2xxx/qla_init.c
index 8d1acc802a67..420045155ba0 100644
--- a/drivers/scsi/qla2xxx/qla_init.c
+++ b/drivers/scsi/qla2xxx/qla_init.c
@@ -366,14 +366,16 @@ qla2x00_async_prlo(struct scsi_qla_host *vha, fc_port_t *fcport)
 	qla2x00_init_timer(sp, qla2x00_get_async_timeout(vha) + 2);
 
 	sp->done = qla2x00_async_prlo_sp_done;
-	rval = qla2x00_start_sp(sp);
-	if (rval != QLA_SUCCESS)
-		goto done_free_sp;
 
 	ql_dbg(ql_dbg_disc, vha, 0x2070,
 	    "Async-prlo - hdl=%x loop-id=%x portid=%02x%02x%02x.\n",
 	    sp->handle, fcport->loop_id, fcport->d_id.b.domain,
 	    fcport->d_id.b.area, fcport->d_id.b.al_pa);
+
+	rval = qla2x00_start_sp(sp);
+	if (rval != QLA_SUCCESS)
+		goto done_free_sp;
+
 	return rval;
 
 done_free_sp:
@@ -471,9 +473,11 @@ qla2x00_async_adisc(struct scsi_qla_host *vha, fc_port_t *fcport,
 {
 	srb_t *sp;
 	struct srb_iocb *lio;
-	int rval;
+	int rval = QLA_FUNCTION_FAILED;
+
+	if (!vha->flags.online || (fcport->flags & FCF_ASYNC_SENT))
+		return rval;
 
-	rval = QLA_FUNCTION_FAILED;
 	fcport->flags |= FCF_ASYNC_SENT;
 	sp = qla2x00_get_sp(vha, fcport, GFP_KERNEL);
 	if (!sp)
@@ -644,11 +648,14 @@ static void qla24xx_handle_gnl_done_event(scsi_qla_host_t *vha,
 				break;
 			case DSC_LS_PORT_UNAVAIL:
 			default:
-				if (fcport->loop_id != FC_NO_LOOP_ID)
-					qla2x00_clear_loop_id(fcport);
-
-				fcport->loop_id = loop_id;
-				fcport->fw_login_state = DSC_LS_PORT_UNAVAIL;
+				if (fcport->loop_id == FC_NO_LOOP_ID) {
+					qla2x00_find_new_loop_id(vha, fcport);
+					fcport->fw_login_state =
+					    DSC_LS_PORT_UNAVAIL;
+				}
+				ql_dbg(ql_dbg_disc, vha, 0x20e5,
+				    "%s %d %8phC\n", __func__, __LINE__,
+				    fcport->port_name);
 				qla24xx_fcport_handle_login(vha, fcport);
 				break;
 			}
@@ -931,14 +938,14 @@ int qla24xx_async_gnl(struct scsi_qla_host *vha, fc_port_t *fcport)
 
 	sp->done = qla24xx_async_gnl_sp_done;
 
-	rval = qla2x00_start_sp(sp);
-	if (rval != QLA_SUCCESS)
-		goto done_free_sp;
-
 	ql_dbg(ql_dbg_disc, vha, 0x20da,
 	    "Async-%s - OUT WWPN %8phC hndl %x\n",
 	    sp->name, fcport->port_name, sp->handle);
 
+	rval = qla2x00_start_sp(sp);
+	if (rval != QLA_SUCCESS)
+		goto done_free_sp;
+
 	return rval;
 
 done_free_sp:
@@ -1072,6 +1079,11 @@ qla24xx_async_prli(struct scsi_qla_host *vha, fc_port_t *fcport)
 	if  (fcport->fc4f_nvme)
 		lio->u.logio.flags |= SRB_LOGIN_NVME_PRLI;
 
+	ql_dbg(ql_dbg_disc, vha, 0x211b,
+	    "Async-prli - %8phC hdl=%x, loopid=%x portid=%06x retries=%d %s.\n",
+	    fcport->port_name, sp->handle, fcport->loop_id, fcport->d_id.b24,
+	    fcport->login_retry, fcport->fc4f_nvme ? "nvme" : "fc");
+
 	rval = qla2x00_start_sp(sp);
 	if (rval != QLA_SUCCESS) {
 		fcport->flags |= FCF_LOGIN_NEEDED;
@@ -1079,11 +1091,6 @@ qla24xx_async_prli(struct scsi_qla_host *vha, fc_port_t *fcport)
 		goto done_free_sp;
 	}
 
-	ql_dbg(ql_dbg_disc, vha, 0x211b,
-	    "Async-prli - %8phC hdl=%x, loopid=%x portid=%06x retries=%d %s.\n",
-	    fcport->port_name, sp->handle, fcport->loop_id, fcport->d_id.b24,
-	    fcport->login_retry, fcport->fc4f_nvme ? "nvme" : "fc");
-
 	return rval;
 
 done_free_sp:
@@ -1471,29 +1478,6 @@ int qla24xx_fcport_handle_login(struct scsi_qla_host *vha, fc_port_t *fcport)
 	return 0;
 }
 
-static
-void qla24xx_handle_rscn_event(fc_port_t *fcport, struct event_arg *ea)
-{
-	fcport->rscn_gen++;
-
-	ql_dbg(ql_dbg_disc, fcport->vha, 0x210c,
-	    "%s %8phC DS %d LS %d\n",
-	    __func__, fcport->port_name, fcport->disc_state,
-	    fcport->fw_login_state);
-
-	if (fcport->flags & FCF_ASYNC_SENT)
-		return;
-
-	switch (fcport->disc_state) {
-	case DSC_DELETED:
-	case DSC_LOGIN_COMPLETE:
-		qla24xx_post_gpnid_work(fcport->vha, &ea->id);
-		break;
-	default:
-		break;
-	}
-}
-
 int qla24xx_post_newsess_work(struct scsi_qla_host *vha, port_id_t *id,
     u8 *port_name, u8 *node_name, void *pla, u8 fc4_type)
 {
@@ -1560,8 +1544,6 @@ static void qla_handle_els_plogi_done(scsi_qla_host_t *vha,
 
 void qla2x00_fcport_event_handler(scsi_qla_host_t *vha, struct event_arg *ea)
 {
-	fc_port_t *f, *tf;
-	uint32_t id = 0, mask, rid;
 	fc_port_t *fcport;
 
 	switch (ea->event) {
@@ -1574,10 +1556,6 @@ void qla2x00_fcport_event_handler(scsi_qla_host_t *vha, struct event_arg *ea)
 	case FCME_RSCN:
 		if (test_bit(UNLOADING, &vha->dpc_flags))
 			return;
-		switch (ea->id.b.rsvd_1) {
-		case RSCN_PORT_ADDR:
-#define BIGSCAN 1
-#if defined BIGSCAN & BIGSCAN > 0
 		{
 			unsigned long flags;
 			fcport = qla2x00_find_fcport_by_nportid
@@ -1596,59 +1574,6 @@ void qla2x00_fcport_event_handler(scsi_qla_host_t *vha, struct event_arg *ea)
 			}
 			spin_unlock_irqrestore(&vha->work_lock, flags);
 		}
-#else
-		{
-			int rc;
-			fcport = qla2x00_find_fcport_by_nportid(vha, &ea->id, 1);
-			if (!fcport) {
-				/* cable moved */
-				 rc = qla24xx_post_gpnid_work(vha, &ea->id);
-				 if (rc) {
-					 ql_log(ql_log_warn, vha, 0xd044,
-					     "RSCN GPNID work failed %06x\n",
-					     ea->id.b24);
-				 }
-			} else {
-				ea->fcport = fcport;
-				fcport->scan_needed = 1;
-				qla24xx_handle_rscn_event(fcport, ea);
-			}
-		}
-#endif
-			break;
-		case RSCN_AREA_ADDR:
-		case RSCN_DOM_ADDR:
-			if (ea->id.b.rsvd_1 == RSCN_AREA_ADDR) {
-				mask = 0xffff00;
-				ql_dbg(ql_dbg_async, vha, 0x5044,
-				    "RSCN: Area 0x%06x was affected\n",
-				    ea->id.b24);
-			} else {
-				mask = 0xff0000;
-				ql_dbg(ql_dbg_async, vha, 0x507a,
-				    "RSCN: Domain 0x%06x was affected\n",
-				    ea->id.b24);
-			}
-
-			rid = ea->id.b24 & mask;
-			list_for_each_entry_safe(f, tf, &vha->vp_fcports,
-			    list) {
-				id = f->d_id.b24 & mask;
-				if (rid == id) {
-					ea->fcport = f;
-					qla24xx_handle_rscn_event(f, ea);
-				}
-			}
-			break;
-		case RSCN_FAB_ADDR:
-		default:
-			ql_log(ql_log_warn, vha, 0xd045,
-			    "RSCN: Fabric was affected. Addr format %d\n",
-			    ea->id.b.rsvd_1);
-			qla2x00_mark_all_devices_lost(vha, 1);
-			set_bit(LOOP_RESYNC_NEEDED, &vha->dpc_flags);
-			set_bit(LOCAL_LOOP_UPDATE, &vha->dpc_flags);
-		}
 		break;
 	case FCME_GNL_DONE:
 		qla24xx_handle_gnl_done_event(vha, ea);
@@ -1709,11 +1634,7 @@ void qla_rscn_replay(fc_port_t *fcport)
                ea.event = FCME_RSCN;
                ea.id = fcport->d_id;
                ea.id.b.rsvd_1 = RSCN_PORT_ADDR;
-#if defined BIGSCAN & BIGSCAN > 0
                qla2x00_fcport_event_handler(fcport->vha, &ea);
-#else
-               qla24xx_post_gpnid_work(fcport->vha, &ea.id);
-#endif
 	}
 }
 
@@ -1784,8 +1705,8 @@ qla2x00_async_tm_cmd(fc_port_t *fcport, uint32_t flags, uint32_t lun,
 		lun = (uint16_t)tm_iocb->u.tmf.lun;
 
 		/* Issue Marker IOCB */
-		qla2x00_marker(vha, vha->hw->req_q_map[0],
-		    vha->hw->rsp_q_map[0], fcport->loop_id, lun,
+		qla2x00_marker(vha, vha->hw->base_qpair,
+		    fcport->loop_id, lun,
 		    flags == TCF_LUN_RESET ? MK_SYNC_ID_LUN : MK_SYNC_ID);
 	}
 
@@ -1829,7 +1750,7 @@ qla24xx_async_abort_cmd(srb_t *cmd_sp, bool wait)
 	int rval = QLA_FUNCTION_FAILED;
 
 	sp = qla2xxx_get_qpair_sp(cmd_sp->vha, cmd_sp->qpair, cmd_sp->fcport,
-	    GFP_KERNEL);
+	    GFP_ATOMIC);
 	if (!sp)
 		goto done;
 
@@ -1912,6 +1833,12 @@ qla24xx_handle_prli_done_event(struct scsi_qla_host *vha, struct event_arg *ea)
 
 		ea->fcport->chip_reset = vha->hw->base_qpair->chip_reset;
 		ea->fcport->logout_on_delete = 1;
+		ea->fcport->nvme_prli_service_param = ea->iop[0];
+		if (ea->iop[0] & NVME_PRLI_SP_FIRST_BURST)
+			ea->fcport->nvme_first_burst_size =
+			    (ea->iop[1] & 0xffff) * 512;
+		else
+			ea->fcport->nvme_first_burst_size = 0;
 		qla24xx_post_gpdb_work(vha, ea->fcport, 0);
 		break;
 	default:
@@ -3955,8 +3882,17 @@ qla24xx_config_rings(struct scsi_qla_host *vha)
 		WRT_REG_DWORD(&reg->isp24.rsp_q_in, 0);
 		WRT_REG_DWORD(&reg->isp24.rsp_q_out, 0);
 	}
+
 	qlt_24xx_config_rings(vha);
 
+	/* If the user has configured the speed, set it here */
+	if (ha->set_data_rate) {
+		ql_dbg(ql_dbg_init, vha, 0x00fd,
+		    "Speed set by user : %s Gbps \n",
+		    qla2x00_get_link_speed_str(ha, ha->set_data_rate));
+		icb->firmware_options_3 = (ha->set_data_rate << 13);
+	}
+
 	/* PCI posting */
 	RD_REG_DWORD(&ioreg->hccr);
 }
@@ -4755,6 +4691,16 @@ qla2x00_alloc_fcport(scsi_qla_host_t *vha, gfp_t flags)
 	if (!fcport)
 		return NULL;
 
+	fcport->ct_desc.ct_sns = dma_alloc_coherent(&vha->hw->pdev->dev,
+		sizeof(struct ct_sns_pkt), &fcport->ct_desc.ct_sns_dma,
+		flags);
+	if (!fcport->ct_desc.ct_sns) {
+		ql_log(ql_log_warn, vha, 0xd049,
+		    "Failed to allocate ct_sns request.\n");
+		kfree(fcport);
+		return NULL;
+	}
+
 	/* Setup fcport template structure. */
 	fcport->vha = vha;
 	fcport->port_type = FCT_UNKNOWN;
@@ -4763,13 +4709,11 @@ qla2x00_alloc_fcport(scsi_qla_host_t *vha, gfp_t flags)
 	fcport->supported_classes = FC_COS_UNSPECIFIED;
 	fcport->fp_speed = PORT_SPEED_UNKNOWN;
 
-	fcport->ct_desc.ct_sns = dma_alloc_coherent(&vha->hw->pdev->dev,
-		sizeof(struct ct_sns_pkt), &fcport->ct_desc.ct_sns_dma,
-		flags);
 	fcport->disc_state = DSC_DELETED;
 	fcport->fw_login_state = DSC_LS_PORT_UNAVAIL;
 	fcport->deleted = QLA_SESS_DELETED;
 	fcport->login_retry = vha->hw->login_retry_count;
+	fcport->chip_reset = vha->hw->base_qpair->chip_reset;
 	fcport->logout_on_delete = 1;
 
 	if (!fcport->ct_desc.ct_sns) {
@@ -4778,6 +4722,7 @@ qla2x00_alloc_fcport(scsi_qla_host_t *vha, gfp_t flags)
 		kfree(fcport);
 		fcport = NULL;
 	}
+
 	INIT_WORK(&fcport->del_work, qla24xx_delete_sess_fn);
 	INIT_WORK(&fcport->reg_work, qla_register_fcport_fn);
 	INIT_LIST_HEAD(&fcport->gnl_entry);
@@ -5046,11 +4991,6 @@ qla2x00_configure_local_loop(scsi_qla_host_t *vha)
 		if ((domain & 0xf0) == 0xf0)
 			continue;
 
-		/* Bypass if not same domain and area of adapter. */
-		if (area && domain &&
-		    (area != vha->d_id.b.area || domain != vha->d_id.b.domain))
-			continue;
-
 		/* Bypass invalid local loop ID. */
 		if (loop_id > LAST_LOCAL_LOOP_ID)
 			continue;
@@ -6101,11 +6041,6 @@ qla2x00_loop_resync(scsi_qla_host_t *vha)
 {
 	int rval = QLA_SUCCESS;
 	uint32_t wait_time;
-	struct req_que *req;
-	struct rsp_que *rsp;
-
-	req = vha->req;
-	rsp = req->rsp;
 
 	clear_bit(ISP_ABORT_RETRY, &vha->dpc_flags);
 	if (vha->flags.online) {
@@ -6118,8 +6053,8 @@ qla2x00_loop_resync(scsi_qla_host_t *vha)
 					 * Issue a marker after FW becomes
 					 * ready.
 					 */
-					qla2x00_marker(vha, req, rsp, 0, 0,
-						MK_SYNC_ALL);
+					qla2x00_marker(vha, vha->hw->base_qpair,
+					    0, 0, MK_SYNC_ALL);
 					vha->marker_needed = 0;
 				}
 
@@ -6857,8 +6792,6 @@ qla2x00_restart_isp(scsi_qla_host_t *vha)
 {
 	int status = 0;
 	struct qla_hw_data *ha = vha->hw;
-	struct req_que *req = ha->req_q_map[0];
-	struct rsp_que *rsp = ha->rsp_q_map[0];
 
 	/* If firmware needs to be loaded */
 	if (qla2x00_isp_firmware(vha)) {
@@ -6878,7 +6811,7 @@ qla2x00_restart_isp(scsi_qla_host_t *vha)
 		status = qla2x00_fw_ready(vha);
 		if (!status) {
 			/* Issue a marker after FW becomes ready. */
-			qla2x00_marker(vha, req, rsp, 0, 0, MK_SYNC_ALL);
+			qla2x00_marker(vha, ha->base_qpair, 0, 0, MK_SYNC_ALL);
 			set_bit(LOOP_RESYNC_NEEDED, &vha->dpc_flags);
 		}
 
@@ -7933,22 +7866,15 @@ qla24xx_configure_vhba(scsi_qla_host_t *vha)
 	uint16_t mb[MAILBOX_REGISTER_COUNT];
 	struct qla_hw_data *ha = vha->hw;
 	struct scsi_qla_host *base_vha = pci_get_drvdata(ha->pdev);
-	struct req_que *req;
-	struct rsp_que *rsp;
 
 	if (!vha->vp_idx)
 		return -EINVAL;
 
 	rval = qla2x00_fw_ready(base_vha);
-	if (vha->qpair)
-		req = vha->qpair->req;
-	else
-		req = ha->req_q_map[0];
-	rsp = req->rsp;
 
 	if (rval == QLA_SUCCESS) {
 		clear_bit(RESET_MARKER_NEEDED, &vha->dpc_flags);
-		qla2x00_marker(vha, req, rsp, 0, 0, MK_SYNC_ALL);
+		qla2x00_marker(vha, ha->base_qpair, 0, 0, MK_SYNC_ALL);
 	}
 
 	vha->flags.management_server_logged_in = 0;
@@ -8340,8 +8266,6 @@ qla82xx_restart_isp(scsi_qla_host_t *vha)
 {
 	int status, rval;
 	struct qla_hw_data *ha = vha->hw;
-	struct req_que *req = ha->req_q_map[0];
-	struct rsp_que *rsp = ha->rsp_q_map[0];
 	struct scsi_qla_host *vp;
 	unsigned long flags;
 
@@ -8353,7 +8277,7 @@ qla82xx_restart_isp(scsi_qla_host_t *vha)
 		status = qla2x00_fw_ready(vha);
 		if (!status) {
 			/* Issue a marker after FW becomes ready. */
-			qla2x00_marker(vha, req, rsp, 0, 0, MK_SYNC_ALL);
+			qla2x00_marker(vha, ha->base_qpair, 0, 0, MK_SYNC_ALL);
 			vha->flags.online = 1;
 			set_bit(LOOP_RESYNC_NEEDED, &vha->dpc_flags);
 		}
diff --git a/drivers/scsi/qla2xxx/qla_iocb.c b/drivers/scsi/qla2xxx/qla_iocb.c
index 032635321ad6..63f8e3c19841 100644
--- a/drivers/scsi/qla2xxx/qla_iocb.c
+++ b/drivers/scsi/qla2xxx/qla_iocb.c
@@ -336,7 +336,7 @@ qla2x00_start_scsi(srb_t *sp)
 
 	/* Send marker if required */
 	if (vha->marker_needed != 0) {
-		if (qla2x00_marker(vha, req, rsp, 0, 0, MK_SYNC_ALL) !=
+		if (qla2x00_marker(vha, ha->base_qpair, 0, 0, MK_SYNC_ALL) !=
 		    QLA_SUCCESS) {
 			return (QLA_FUNCTION_FAILED);
 		}
@@ -490,8 +490,7 @@ qla2x00_start_iocbs(struct scsi_qla_host *vha, struct req_que *req)
 /**
  * qla2x00_marker() - Send a marker IOCB to the firmware.
  * @vha: HA context
- * @req: request queue
- * @rsp: response queue
+ * @qpair: queue pair pointer
  * @loop_id: loop ID
  * @lun: LUN
  * @type: marker modifier
@@ -501,18 +500,16 @@ qla2x00_start_iocbs(struct scsi_qla_host *vha, struct req_que *req)
  * Returns non-zero if a failure occurred, else zero.
  */
 static int
-__qla2x00_marker(struct scsi_qla_host *vha, struct req_que *req,
-			struct rsp_que *rsp, uint16_t loop_id,
-			uint64_t lun, uint8_t type)
+__qla2x00_marker(struct scsi_qla_host *vha, struct qla_qpair *qpair,
+    uint16_t loop_id, uint64_t lun, uint8_t type)
 {
 	mrk_entry_t *mrk;
 	struct mrk_entry_24xx *mrk24 = NULL;
-
+	struct req_que *req = qpair->req;
 	struct qla_hw_data *ha = vha->hw;
 	scsi_qla_host_t *base_vha = pci_get_drvdata(ha->pdev);
 
-	req = ha->req_q_map[0];
-	mrk = (mrk_entry_t *)qla2x00_alloc_iocbs(vha, NULL);
+	mrk = (mrk_entry_t *)__qla2x00_alloc_iocbs(qpair, NULL);
 	if (mrk == NULL) {
 		ql_log(ql_log_warn, base_vha, 0x3026,
 		    "Failed to allocate Marker IOCB.\n");
@@ -543,16 +540,15 @@ __qla2x00_marker(struct scsi_qla_host *vha, struct req_que *req,
 }
 
 int
-qla2x00_marker(struct scsi_qla_host *vha, struct req_que *req,
-		struct rsp_que *rsp, uint16_t loop_id, uint64_t lun,
-		uint8_t type)
+qla2x00_marker(struct scsi_qla_host *vha, struct qla_qpair *qpair,
+    uint16_t loop_id, uint64_t lun, uint8_t type)
 {
 	int ret;
 	unsigned long flags = 0;
 
-	spin_lock_irqsave(&vha->hw->hardware_lock, flags);
-	ret = __qla2x00_marker(vha, req, rsp, loop_id, lun, type);
-	spin_unlock_irqrestore(&vha->hw->hardware_lock, flags);
+	spin_lock_irqsave(qpair->qp_lock_ptr, flags);
+	ret = __qla2x00_marker(vha, qpair, loop_id, lun, type);
+	spin_unlock_irqrestore(qpair->qp_lock_ptr, flags);
 
 	return (ret);
 }
@@ -567,11 +563,11 @@ qla2x00_marker(struct scsi_qla_host *vha, struct req_que *req,
 int qla2x00_issue_marker(scsi_qla_host_t *vha, int ha_locked)
 {
 	if (ha_locked) {
-		if (__qla2x00_marker(vha, vha->req, vha->req->rsp, 0, 0,
+		if (__qla2x00_marker(vha, vha->hw->base_qpair, 0, 0,
 					MK_SYNC_ALL) != QLA_SUCCESS)
 			return QLA_FUNCTION_FAILED;
 	} else {
-		if (qla2x00_marker(vha, vha->req, vha->req->rsp, 0, 0,
+		if (qla2x00_marker(vha, vha->hw->base_qpair, 0, 0,
 					MK_SYNC_ALL) != QLA_SUCCESS)
 			return QLA_FUNCTION_FAILED;
 	}
@@ -1098,88 +1094,300 @@ qla24xx_walk_and_build_sglist(struct qla_hw_data *ha, srb_t *sp, uint32_t *dsd,
 
 int
 qla24xx_walk_and_build_prot_sglist(struct qla_hw_data *ha, srb_t *sp,
-	uint32_t *dsd, uint16_t tot_dsds, struct qla_tc_param *tc)
+    uint32_t *cur_dsd, uint16_t tot_dsds, struct qla_tgt_cmd *tc)
 {
-	void *next_dsd;
-	uint8_t avail_dsds = 0;
-	uint32_t dsd_list_len;
-	struct dsd_dma *dsd_ptr;
+	struct dsd_dma *dsd_ptr = NULL, *dif_dsd, *nxt_dsd;
 	struct scatterlist *sg, *sgl;
-	int	i;
-	struct scsi_cmnd *cmd;
-	uint32_t *cur_dsd = dsd;
-	uint16_t used_dsds = tot_dsds;
+	struct crc_context *difctx = NULL;
 	struct scsi_qla_host *vha;
+	uint dsd_list_len;
+	uint avail_dsds = 0;
+	uint used_dsds = tot_dsds;
+	bool dif_local_dma_alloc = false;
+	bool direction_to_device = false;
+	int i;
 
 	if (sp) {
-		cmd = GET_CMD_SP(sp);
+		struct scsi_cmnd *cmd = GET_CMD_SP(sp);
 		sgl = scsi_prot_sglist(cmd);
 		vha = sp->vha;
+		difctx = sp->u.scmd.ctx;
+		direction_to_device = cmd->sc_data_direction == DMA_TO_DEVICE;
+		ql_dbg(ql_dbg_tgt + ql_dbg_verbose, vha, 0xe021,
+		  "%s: scsi_cmnd: %p, crc_ctx: %p, sp: %p\n",
+			__func__, cmd, difctx, sp);
 	} else if (tc) {
 		vha = tc->vha;
 		sgl = tc->prot_sg;
+		difctx = tc->ctx;
+		direction_to_device = tc->dma_data_direction == DMA_TO_DEVICE;
 	} else {
 		BUG();
 		return 1;
 	}
 
-	ql_dbg(ql_dbg_tgt, vha, 0xe021,
-		"%s: enter\n", __func__);
-
-	for_each_sg(sgl, sg, tot_dsds, i) {
-		dma_addr_t	sle_dma;
-
-		/* Allocate additional continuation packets? */
-		if (avail_dsds == 0) {
-			avail_dsds = (used_dsds > QLA_DSDS_PER_IOCB) ?
-						QLA_DSDS_PER_IOCB : used_dsds;
-			dsd_list_len = (avail_dsds + 1) * 12;
-			used_dsds -= avail_dsds;
-
-			/* allocate tracking DS */
-			dsd_ptr = kzalloc(sizeof(struct dsd_dma), GFP_ATOMIC);
-			if (!dsd_ptr)
-				return 1;
-
-			/* allocate new list */
-			dsd_ptr->dsd_addr = next_dsd =
-			    dma_pool_alloc(ha->dl_dma_pool, GFP_ATOMIC,
-				&dsd_ptr->dsd_list_dma);
-
-			if (!next_dsd) {
-				/*
-				 * Need to cleanup only this dsd_ptr, rest
-				 * will be done by sp_free_dma()
-				 */
-				kfree(dsd_ptr);
-				return 1;
+	ql_dbg(ql_dbg_tgt + ql_dbg_verbose, vha, 0xe021,
+	    "%s: enter (write=%u)\n", __func__, direction_to_device);
+
+	/* if initiator doing write or target doing read */
+	if (direction_to_device) {
+		for_each_sg(sgl, sg, tot_dsds, i) {
+			dma_addr_t sle_phys = sg_phys(sg);
+
+			/* If SGE addr + len flips bits in upper 32-bits */
+			if (MSD(sle_phys + sg->length) ^ MSD(sle_phys)) {
+				ql_dbg(ql_dbg_tgt + ql_dbg_verbose, vha, 0xe022,
+				    "%s: page boundary crossing (phys=%llx len=%x)\n",
+				    __func__, sle_phys, sg->length);
+
+				if (difctx) {
+					ha->dif_bundle_crossed_pages++;
+					dif_local_dma_alloc = true;
+				} else {
+					ql_dbg(ql_dbg_tgt + ql_dbg_verbose,
+					    vha, 0xe022,
+					    "%s: difctx pointer is NULL\n",
+					    __func__);
+				}
+				break;
 			}
+		}
+		ha->dif_bundle_writes++;
+	} else {
+		ha->dif_bundle_reads++;
+	}
+
+	if (ql2xdifbundlinginternalbuffers)
+		dif_local_dma_alloc = direction_to_device;
+
+	if (dif_local_dma_alloc) {
+		u32 track_difbundl_buf = 0;
+		u32 ldma_sg_len = 0;
+		u8 ldma_needed = 1;
+
+		difctx->no_dif_bundl = 0;
+		difctx->dif_bundl_len = 0;
+
+		/* Track DSD buffers */
+		INIT_LIST_HEAD(&difctx->ldif_dsd_list);
+		/* Track local DMA buffers */
+		INIT_LIST_HEAD(&difctx->ldif_dma_hndl_list);
+
+		for_each_sg(sgl, sg, tot_dsds, i) {
+			u32 sglen = sg_dma_len(sg);
+
+			ql_dbg(ql_dbg_tgt + ql_dbg_verbose, vha, 0xe023,
+			    "%s: sg[%x] (phys=%llx sglen=%x) ldma_sg_len: %x dif_bundl_len: %x ldma_needed: %x\n",
+			    __func__, i, sg_phys(sg), sglen, ldma_sg_len,
+			    difctx->dif_bundl_len, ldma_needed);
+
+			while (sglen) {
+				u32 xfrlen = 0;
+
+				if (ldma_needed) {
+					/*
+					 * Allocate list item to store
+					 * the DMA buffers
+					 */
+					dsd_ptr = kzalloc(sizeof(*dsd_ptr),
+					    GFP_ATOMIC);
+					if (!dsd_ptr) {
+						ql_dbg(ql_dbg_tgt, vha, 0xe024,
+						    "%s: failed alloc dsd_ptr\n",
+						    __func__);
+						return 1;
+					}
+					ha->dif_bundle_kallocs++;
+
+					/* allocate dma buffer */
+					dsd_ptr->dsd_addr = dma_pool_alloc
+						(ha->dif_bundl_pool, GFP_ATOMIC,
+						 &dsd_ptr->dsd_list_dma);
+					if (!dsd_ptr->dsd_addr) {
+						ql_dbg(ql_dbg_tgt, vha, 0xe024,
+						    "%s: failed alloc ->dsd_ptr\n",
+						    __func__);
+						/*
+						 * need to cleanup only this
+						 * dsd_ptr rest will be done
+						 * by sp_free_dma()
+						 */
+						kfree(dsd_ptr);
+						ha->dif_bundle_kallocs--;
+						return 1;
+					}
+					ha->dif_bundle_dma_allocs++;
+					ldma_needed = 0;
+					difctx->no_dif_bundl++;
+					list_add_tail(&dsd_ptr->list,
+					    &difctx->ldif_dma_hndl_list);
+				}
+
+				/* xfrlen is min of dma pool size and sglen */
+				xfrlen = (sglen >
+				   (DIF_BUNDLING_DMA_POOL_SIZE - ldma_sg_len)) ?
+				    DIF_BUNDLING_DMA_POOL_SIZE - ldma_sg_len :
+				    sglen;
+
+				/* replace with local allocated dma buffer */
+				sg_pcopy_to_buffer(sgl, sg_nents(sgl),
+				    dsd_ptr->dsd_addr + ldma_sg_len, xfrlen,
+				    difctx->dif_bundl_len);
+				difctx->dif_bundl_len += xfrlen;
+				sglen -= xfrlen;
+				ldma_sg_len += xfrlen;
+				if (ldma_sg_len == DIF_BUNDLING_DMA_POOL_SIZE ||
+				    sg_is_last(sg)) {
+					ldma_needed = 1;
+					ldma_sg_len = 0;
+				}
+			}
+		}
 
-			if (sp) {
-				list_add_tail(&dsd_ptr->list,
-				    &((struct crc_context *)
-					    sp->u.scmd.ctx)->dsd_list);
+		track_difbundl_buf = used_dsds = difctx->no_dif_bundl;
+		ql_dbg(ql_dbg_tgt + ql_dbg_verbose, vha, 0xe025,
+		    "dif_bundl_len=%x, no_dif_bundl=%x track_difbundl_buf: %x\n",
+		    difctx->dif_bundl_len, difctx->no_dif_bundl,
+		    track_difbundl_buf);
 
-				sp->flags |= SRB_CRC_CTX_DSD_VALID;
-			} else {
-				list_add_tail(&dsd_ptr->list,
-				    &(tc->ctx->dsd_list));
-				*tc->ctx_dsd_alloced = 1;
+		if (sp)
+			sp->flags |= SRB_DIF_BUNDL_DMA_VALID;
+		else
+			tc->prot_flags = DIF_BUNDL_DMA_VALID;
+
+		list_for_each_entry_safe(dif_dsd, nxt_dsd,
+		    &difctx->ldif_dma_hndl_list, list) {
+			u32 sglen = (difctx->dif_bundl_len >
+			    DIF_BUNDLING_DMA_POOL_SIZE) ?
+			    DIF_BUNDLING_DMA_POOL_SIZE : difctx->dif_bundl_len;
+
+			BUG_ON(track_difbundl_buf == 0);
+
+			/* Allocate additional continuation packets? */
+			if (avail_dsds == 0) {
+				ql_dbg(ql_dbg_tgt + ql_dbg_verbose, vha,
+				    0xe024,
+				    "%s: adding continuation iocb's\n",
+				    __func__);
+				avail_dsds = (used_dsds > QLA_DSDS_PER_IOCB) ?
+				    QLA_DSDS_PER_IOCB : used_dsds;
+				dsd_list_len = (avail_dsds + 1) * 12;
+				used_dsds -= avail_dsds;
+
+				/* allocate tracking DS */
+				dsd_ptr = kzalloc(sizeof(*dsd_ptr), GFP_ATOMIC);
+				if (!dsd_ptr) {
+					ql_dbg(ql_dbg_tgt, vha, 0xe026,
+					    "%s: failed alloc dsd_ptr\n",
+					    __func__);
+					return 1;
+				}
+				ha->dif_bundle_kallocs++;
+
+				difctx->no_ldif_dsd++;
+				/* allocate new list */
+				dsd_ptr->dsd_addr =
+				    dma_pool_alloc(ha->dl_dma_pool, GFP_ATOMIC,
+					&dsd_ptr->dsd_list_dma);
+				if (!dsd_ptr->dsd_addr) {
+					ql_dbg(ql_dbg_tgt, vha, 0xe026,
+					    "%s: failed alloc ->dsd_addr\n",
+					    __func__);
+					/*
+					 * need to cleanup only this dsd_ptr
+					 *  rest will be done by sp_free_dma()
+					 */
+					kfree(dsd_ptr);
+					ha->dif_bundle_kallocs--;
+					return 1;
+				}
+				ha->dif_bundle_dma_allocs++;
+
+				if (sp) {
+					list_add_tail(&dsd_ptr->list,
+					    &difctx->ldif_dsd_list);
+					sp->flags |= SRB_CRC_CTX_DSD_VALID;
+				} else {
+					list_add_tail(&dsd_ptr->list,
+					    &difctx->ldif_dsd_list);
+					tc->ctx_dsd_alloced = 1;
+				}
+
+				/* add new list to cmd iocb or last list */
+				*cur_dsd++ =
+				    cpu_to_le32(LSD(dsd_ptr->dsd_list_dma));
+				*cur_dsd++ =
+				    cpu_to_le32(MSD(dsd_ptr->dsd_list_dma));
+				*cur_dsd++ = dsd_list_len;
+				cur_dsd = dsd_ptr->dsd_addr;
 			}
-
-			/* add new list to cmd iocb or last list */
-			*cur_dsd++ = cpu_to_le32(LSD(dsd_ptr->dsd_list_dma));
-			*cur_dsd++ = cpu_to_le32(MSD(dsd_ptr->dsd_list_dma));
-			*cur_dsd++ = dsd_list_len;
-			cur_dsd = (uint32_t *)next_dsd;
+			*cur_dsd++ = cpu_to_le32(LSD(dif_dsd->dsd_list_dma));
+			*cur_dsd++ = cpu_to_le32(MSD(dif_dsd->dsd_list_dma));
+			*cur_dsd++ = cpu_to_le32(sglen);
+			avail_dsds--;
+			difctx->dif_bundl_len -= sglen;
+			track_difbundl_buf--;
 		}
-		sle_dma = sg_dma_address(sg);
-
-		*cur_dsd++ = cpu_to_le32(LSD(sle_dma));
-		*cur_dsd++ = cpu_to_le32(MSD(sle_dma));
-		*cur_dsd++ = cpu_to_le32(sg_dma_len(sg));
 
-		avail_dsds--;
+		ql_dbg(ql_dbg_tgt + ql_dbg_verbose, vha, 0xe026,
+		    "%s: no_ldif_dsd:%x, no_dif_bundl:%x\n", __func__,
+			difctx->no_ldif_dsd, difctx->no_dif_bundl);
+	} else {
+		for_each_sg(sgl, sg, tot_dsds, i) {
+			dma_addr_t sle_dma;
+
+			/* Allocate additional continuation packets? */
+			if (avail_dsds == 0) {
+				avail_dsds = (used_dsds > QLA_DSDS_PER_IOCB) ?
+				    QLA_DSDS_PER_IOCB : used_dsds;
+				dsd_list_len = (avail_dsds + 1) * 12;
+				used_dsds -= avail_dsds;
+
+				/* allocate tracking DS */
+				dsd_ptr = kzalloc(sizeof(*dsd_ptr), GFP_ATOMIC);
+				if (!dsd_ptr) {
+					ql_dbg(ql_dbg_tgt + ql_dbg_verbose,
+					    vha, 0xe027,
+					    "%s: failed alloc dsd_dma...\n",
+					    __func__);
+					return 1;
+				}
+
+				/* allocate new list */
+				dsd_ptr->dsd_addr =
+				    dma_pool_alloc(ha->dl_dma_pool, GFP_ATOMIC,
+					&dsd_ptr->dsd_list_dma);
+				if (!dsd_ptr->dsd_addr) {
+					/* need to cleanup only this dsd_ptr */
+					/* rest will be done by sp_free_dma() */
+					kfree(dsd_ptr);
+					return 1;
+				}
+
+				if (sp) {
+					list_add_tail(&dsd_ptr->list,
+					    &difctx->dsd_list);
+					sp->flags |= SRB_CRC_CTX_DSD_VALID;
+				} else {
+					list_add_tail(&dsd_ptr->list,
+					    &difctx->dsd_list);
+					tc->ctx_dsd_alloced = 1;
+				}
+
+				/* add new list to cmd iocb or last list */
+				*cur_dsd++ =
+				    cpu_to_le32(LSD(dsd_ptr->dsd_list_dma));
+				*cur_dsd++ =
+				    cpu_to_le32(MSD(dsd_ptr->dsd_list_dma));
+				*cur_dsd++ = dsd_list_len;
+				cur_dsd = dsd_ptr->dsd_addr;
+			}
+			sle_dma = sg_dma_address(sg);
+			*cur_dsd++ = cpu_to_le32(LSD(sle_dma));
+			*cur_dsd++ = cpu_to_le32(MSD(sle_dma));
+			*cur_dsd++ = cpu_to_le32(sg_dma_len(sg));
+			avail_dsds--;
+		}
 	}
 	/* Null termination */
 	*cur_dsd++ = 0;
@@ -1187,7 +1395,6 @@ qla24xx_walk_and_build_prot_sglist(struct qla_hw_data *ha, srb_t *sp,
 	*cur_dsd++ = 0;
 	return 0;
 }
-
 /**
  * qla24xx_build_scsi_crc_2_iocbs() - Build IOCB command utilizing Command
  *							Type 6 IOCB types.
@@ -1416,21 +1623,19 @@ qla24xx_start_scsi(srb_t *sp)
 	uint16_t	req_cnt;
 	uint16_t	tot_dsds;
 	struct req_que *req = NULL;
-	struct rsp_que *rsp = NULL;
 	struct scsi_cmnd *cmd = GET_CMD_SP(sp);
 	struct scsi_qla_host *vha = sp->vha;
 	struct qla_hw_data *ha = vha->hw;
 
 	/* Setup device pointers. */
 	req = vha->req;
-	rsp = req->rsp;
 
 	/* So we know we haven't pci_map'ed anything yet */
 	tot_dsds = 0;
 
 	/* Send marker if required */
 	if (vha->marker_needed != 0) {
-		if (qla2x00_marker(vha, req, rsp, 0, 0, MK_SYNC_ALL) !=
+		if (qla2x00_marker(vha, ha->base_qpair, 0, 0, MK_SYNC_ALL) !=
 		    QLA_SUCCESS)
 			return QLA_FUNCTION_FAILED;
 		vha->marker_needed = 0;
@@ -1583,7 +1788,7 @@ qla24xx_dif_start_scsi(srb_t *sp)
 
 	/* Send marker if required */
 	if (vha->marker_needed != 0) {
-		if (qla2x00_marker(vha, req, rsp, 0, 0, MK_SYNC_ALL) !=
+		if (qla2x00_marker(vha, ha->base_qpair, 0, 0, MK_SYNC_ALL) !=
 		    QLA_SUCCESS)
 			return QLA_FUNCTION_FAILED;
 		vha->marker_needed = 0;
@@ -1754,7 +1959,6 @@ qla2xxx_start_scsi_mq(srb_t *sp)
 	uint16_t	req_cnt;
 	uint16_t	tot_dsds;
 	struct req_que *req = NULL;
-	struct rsp_que *rsp = NULL;
 	struct scsi_cmnd *cmd = GET_CMD_SP(sp);
 	struct scsi_qla_host *vha = sp->fcport->vha;
 	struct qla_hw_data *ha = vha->hw;
@@ -1764,7 +1968,6 @@ qla2xxx_start_scsi_mq(srb_t *sp)
 	spin_lock_irqsave(&qpair->qp_lock, flags);
 
 	/* Setup qpair pointers */
-	rsp = qpair->rsp;
 	req = qpair->req;
 
 	/* So we know we haven't pci_map'ed anything yet */
@@ -1772,7 +1975,7 @@ qla2xxx_start_scsi_mq(srb_t *sp)
 
 	/* Send marker if required */
 	if (vha->marker_needed != 0) {
-		if (__qla2x00_marker(vha, req, rsp, 0, 0, MK_SYNC_ALL) !=
+		if (__qla2x00_marker(vha, qpair, 0, 0, MK_SYNC_ALL) !=
 		    QLA_SUCCESS) {
 			spin_unlock_irqrestore(&qpair->qp_lock, flags);
 			return QLA_FUNCTION_FAILED;
@@ -1940,7 +2143,7 @@ qla2xxx_dif_start_scsi_mq(srb_t *sp)
 
 	/* Send marker if required */
 	if (vha->marker_needed != 0) {
-		if (__qla2x00_marker(vha, req, rsp, 0, 0, MK_SYNC_ALL) !=
+		if (__qla2x00_marker(vha, qpair, 0, 0, MK_SYNC_ALL) !=
 		    QLA_SUCCESS) {
 			spin_unlock_irqrestore(&qpair->qp_lock, flags);
 			return QLA_FUNCTION_FAILED;
@@ -2208,8 +2411,11 @@ qla24xx_prli_iocb(srb_t *sp, struct logio_entry_24xx *logio)
 
 	logio->entry_type = LOGINOUT_PORT_IOCB_TYPE;
 	logio->control_flags = cpu_to_le16(LCF_COMMAND_PRLI);
-	if (lio->u.logio.flags & SRB_LOGIN_NVME_PRLI)
+	if (lio->u.logio.flags & SRB_LOGIN_NVME_PRLI) {
 		logio->control_flags |= LCF_NVME_PRLI;
+		if (sp->vha->flags.nvme_first_burst)
+			logio->io_parameter[0] = NVME_PRLI_SP_FIRST_BURST;
+	}
 
 	logio->nport_handle = cpu_to_le16(sp->fcport->loop_id);
 	logio->port_id[0] = sp->fcport->d_id.b.al_pa;
@@ -2991,8 +3197,8 @@ qla82xx_start_scsi(srb_t *sp)
 
 	/* Send marker if required */
 	if (vha->marker_needed != 0) {
-		if (qla2x00_marker(vha, req,
-			rsp, 0, 0, MK_SYNC_ALL) != QLA_SUCCESS) {
+		if (qla2x00_marker(vha, ha->base_qpair,
+			0, 0, MK_SYNC_ALL) != QLA_SUCCESS) {
 			ql_log(ql_log_warn, vha, 0x300c,
 			    "qla2x00_marker failed for cmd=%p.\n", cmd);
 			return QLA_FUNCTION_FAILED;
@@ -3434,23 +3640,22 @@ qla24xx_prlo_iocb(srb_t *sp, struct logio_entry_24xx *logio)
 int
 qla2x00_start_sp(srb_t *sp)
 {
-	int rval;
+	int rval = QLA_SUCCESS;
 	scsi_qla_host_t *vha = sp->vha;
 	struct qla_hw_data *ha = vha->hw;
 	struct qla_qpair *qp = sp->qpair;
 	void *pkt;
 	unsigned long flags;
 
-	rval = QLA_FUNCTION_FAILED;
 	spin_lock_irqsave(qp->qp_lock_ptr, flags);
 	pkt = __qla2x00_alloc_iocbs(sp->qpair, sp);
 	if (!pkt) {
+		rval = EAGAIN;
 		ql_log(ql_log_warn, vha, 0x700c,
 		    "qla2x00_alloc_iocbs failed.\n");
 		goto done;
 	}
 
-	rval = QLA_SUCCESS;
 	switch (sp->type) {
 	case SRB_LOGIN_CMD:
 		IS_FWI2_CAPABLE(ha) ?
@@ -3646,8 +3851,8 @@ qla2x00_start_bidir(srb_t *sp, struct scsi_qla_host *vha, uint32_t tot_dsds)
 
 	/* Send marker if required */
 	if (vha->marker_needed != 0) {
-		if (qla2x00_marker(vha, req,
-			rsp, 0, 0, MK_SYNC_ALL) != QLA_SUCCESS)
+		if (qla2x00_marker(vha, ha->base_qpair,
+			0, 0, MK_SYNC_ALL) != QLA_SUCCESS)
 			return EXT_STATUS_MAILBOX;
 		vha->marker_needed = 0;
 	}
diff --git a/drivers/scsi/qla2xxx/qla_isr.c b/drivers/scsi/qla2xxx/qla_isr.c
index 8507c43b918c..69bbea9239cc 100644
--- a/drivers/scsi/qla2xxx/qla_isr.c
+++ b/drivers/scsi/qla2xxx/qla_isr.c
@@ -834,7 +834,8 @@ skip_rio:
 			 * Restore for Physical Port only
 			 */
 			if (!vha->vp_idx) {
-				if (ha->flags.fawwpn_enabled) {
+				if (ha->flags.fawwpn_enabled &&
+				    (ha->current_topology == ISP_CFG_F)) {
 					void *wwpn = ha->init_cb->port_name;
 					memcpy(vha->port_name, wwpn, WWN_SIZE);
 					fc_host_port_name(vha->host) =
@@ -1714,6 +1715,15 @@ qla24xx_logio_entry(scsi_qla_host_t *vha, struct req_que *req,
 
 		vha->hw->exch_starvation = 0;
 		data[0] = MBS_COMMAND_COMPLETE;
+
+		if (sp->type == SRB_PRLI_CMD) {
+			lio->u.logio.iop[0] =
+			    le32_to_cpu(logio->io_parameter[0]);
+			lio->u.logio.iop[1] =
+			    le32_to_cpu(logio->io_parameter[1]);
+			goto logio_done;
+		}
+
 		if (sp->type != SRB_LOGIN_CMD)
 			goto logio_done;
 
@@ -2725,6 +2735,17 @@ check_scsi_status:
 			    cp->device->vendor);
 		break;
 
+	case CS_DMA:
+		ql_log(ql_log_info, fcport->vha, 0x3022,
+		    "CS_DMA error: 0x%x-0x%x (0x%x) nexus=%ld:%d:%llu portid=%06x oxid=0x%x cdb=%10phN len=0x%x rsp_info=0x%x resid=0x%x fw_resid=0x%x sp=%p cp=%p.\n",
+		    comp_status, scsi_status, res, vha->host_no,
+		    cp->device->id, cp->device->lun, fcport->d_id.b24,
+		    ox_id, cp->cmnd, scsi_bufflen(cp), rsp_info_len,
+		    resid_len, fw_resid_len, sp, cp);
+		ql_dump_buffer(ql_dbg_tgt + ql_dbg_verbose, vha, 0xe0ee,
+		    pkt, sizeof(*sts24));
+		res = DID_ERROR << 16;
+		break;
 	default:
 		res = DID_ERROR << 16;
 		break;
@@ -3410,7 +3431,7 @@ qla24xx_enable_msix(struct qla_hw_data *ha, struct rsp_que *rsp)
 		min_vecs++;
 	}
 
-	if (USER_CTRL_IRQ(ha)) {
+	if (USER_CTRL_IRQ(ha) || !ha->mqiobase) {
 		/* user wants to control IRQ setting for target mode */
 		ret = pci_alloc_irq_vectors(ha->pdev, min_vecs,
 		    ha->msix_count, PCI_IRQ_MSIX);
diff --git a/drivers/scsi/qla2xxx/qla_mbx.c b/drivers/scsi/qla2xxx/qla_mbx.c
index 191b6b7c8747..5400696e1f6b 100644
--- a/drivers/scsi/qla2xxx/qla_mbx.c
+++ b/drivers/scsi/qla2xxx/qla_mbx.c
@@ -1109,7 +1109,12 @@ qla2x00_get_fw_version(scsi_qla_host_t *vha)
 		 * FW supports nvme and driver load parameter requested nvme.
 		 * BIT 26 of fw_attributes indicates NVMe support.
 		 */
-		if ((ha->fw_attributes_h & 0x400) && ql2xnvmeenable) {
+		if ((ha->fw_attributes_h &
+		    (FW_ATTR_H_NVME | FW_ATTR_H_NVME_UPDATED)) &&
+			ql2xnvmeenable) {
+			if (ha->fw_attributes_h & FW_ATTR_H_NVME_FBURST)
+				vha->flags.nvme_first_burst = 1;
+
 			vha->flags.nvme_enabled = 1;
 			ql_log(ql_log_info, vha, 0xd302,
 			    "%s: FC-NVMe is Enabled (0x%x)\n",
@@ -1508,16 +1513,12 @@ qla2x00_abort_target(struct fc_port *fcport, uint64_t l, int tag)
 	mbx_cmd_t  mc;
 	mbx_cmd_t  *mcp = &mc;
 	scsi_qla_host_t *vha;
-	struct req_que *req;
-	struct rsp_que *rsp;
 
 	vha = fcport->vha;
 
 	ql_dbg(ql_dbg_mbx + ql_dbg_verbose, vha, 0x103e,
 	    "Entered %s.\n", __func__);
 
-	req = vha->hw->req_q_map[0];
-	rsp = req->rsp;
 	mcp->mb[0] = MBC_ABORT_TARGET;
 	mcp->out_mb = MBX_9|MBX_2|MBX_1|MBX_0;
 	if (HAS_EXTENDED_IDS(vha->hw)) {
@@ -1540,7 +1541,7 @@ qla2x00_abort_target(struct fc_port *fcport, uint64_t l, int tag)
 	}
 
 	/* Issue marker IOCB. */
-	rval2 = qla2x00_marker(vha, req, rsp, fcport->loop_id, 0,
+	rval2 = qla2x00_marker(vha, vha->hw->base_qpair, fcport->loop_id, 0,
 							MK_SYNC_ID);
 	if (rval2 != QLA_SUCCESS) {
 		ql_dbg(ql_dbg_mbx, vha, 0x1040,
@@ -1560,16 +1561,12 @@ qla2x00_lun_reset(struct fc_port *fcport, uint64_t l, int tag)
 	mbx_cmd_t  mc;
 	mbx_cmd_t  *mcp = &mc;
 	scsi_qla_host_t *vha;
-	struct req_que *req;
-	struct rsp_que *rsp;
 
 	vha = fcport->vha;
 
 	ql_dbg(ql_dbg_mbx + ql_dbg_verbose, vha, 0x1042,
 	    "Entered %s.\n", __func__);
 
-	req = vha->hw->req_q_map[0];
-	rsp = req->rsp;
 	mcp->mb[0] = MBC_LUN_RESET;
 	mcp->out_mb = MBX_9|MBX_3|MBX_2|MBX_1|MBX_0;
 	if (HAS_EXTENDED_IDS(vha->hw))
@@ -1589,7 +1586,7 @@ qla2x00_lun_reset(struct fc_port *fcport, uint64_t l, int tag)
 	}
 
 	/* Issue marker IOCB. */
-	rval2 = qla2x00_marker(vha, req, rsp, fcport->loop_id, l,
+	rval2 = qla2x00_marker(vha, vha->hw->base_qpair, fcport->loop_id, l,
 								MK_SYNC_ID_LUN);
 	if (rval2 != QLA_SUCCESS) {
 		ql_dbg(ql_dbg_mbx, vha, 0x1044,
@@ -2244,10 +2241,7 @@ qla2x00_lip_reset(scsi_qla_host_t *vha)
 		mcp->out_mb = MBX_2|MBX_1|MBX_0;
 	} else if (IS_FWI2_CAPABLE(vha->hw)) {
 		mcp->mb[0] = MBC_LIP_FULL_LOGIN;
-		if (N2N_TOPO(vha->hw))
-			mcp->mb[1] = BIT_4; /* re-init */
-		else
-			mcp->mb[1] = BIT_6; /* LIP */
+		mcp->mb[1] = BIT_4;
 		mcp->mb[2] = 0;
 		mcp->mb[3] = vha->hw->loop_reset_delay;
 		mcp->out_mb = MBX_3|MBX_2|MBX_1|MBX_0;
@@ -2757,7 +2751,7 @@ qla2x00_full_login_lip(scsi_qla_host_t *vha)
 	    "Entered %s.\n", __func__);
 
 	mcp->mb[0] = MBC_LIP_FULL_LOGIN;
-	mcp->mb[1] = IS_FWI2_CAPABLE(vha->hw) ? BIT_3 : 0;
+	mcp->mb[1] = IS_FWI2_CAPABLE(vha->hw) ? BIT_4 : 0;
 	mcp->mb[2] = 0;
 	mcp->mb[3] = 0;
 	mcp->out_mb = MBX_3|MBX_2|MBX_1|MBX_0;
@@ -3184,7 +3178,6 @@ __qla24xx_issue_tmf(char *name, uint32_t type, struct fc_port *fcport,
 	scsi_qla_host_t *vha;
 	struct qla_hw_data *ha;
 	struct req_que *req;
-	struct rsp_que *rsp;
 	struct qla_qpair *qpair;
 
 	vha = fcport->vha;
@@ -3197,10 +3190,7 @@ __qla24xx_issue_tmf(char *name, uint32_t type, struct fc_port *fcport,
 	if (vha->vp_idx && vha->qpair) {
 		/* NPIV port */
 		qpair = vha->qpair;
-		rsp = qpair->rsp;
 		req = qpair->req;
-	} else {
-		rsp = req->rsp;
 	}
 
 	tsk = dma_pool_zalloc(ha->s_dma_pool, GFP_KERNEL, &tsk_dma);
@@ -3257,7 +3247,7 @@ __qla24xx_issue_tmf(char *name, uint32_t type, struct fc_port *fcport,
 	}
 
 	/* Issue marker IOCB. */
-	rval2 = qla2x00_marker(vha, req, rsp, fcport->loop_id, l,
+	rval2 = qla2x00_marker(vha, ha->base_qpair, fcport->loop_id, l,
 	    type == TCF_LUN_RESET ? MK_SYNC_ID_LUN: MK_SYNC_ID);
 	if (rval2 != QLA_SUCCESS) {
 		ql_dbg(ql_dbg_mbx, vha, 0x1099,
@@ -5248,6 +5238,66 @@ qla81xx_write_mpi_register(scsi_qla_host_t *vha, uint16_t *mb)
 	return rval;
 }
 
+/* Set the specified data rate */
+int
+qla2x00_set_data_rate(scsi_qla_host_t *vha, uint16_t mode)
+{
+	int rval;
+	mbx_cmd_t mc;
+	mbx_cmd_t *mcp = &mc;
+	struct qla_hw_data *ha = vha->hw;
+	uint16_t val;
+
+	ql_dbg(ql_dbg_mbx + ql_dbg_verbose, vha, 0x1106,
+	    "Entered %s speed:0x%x mode:0x%x.\n", __func__, ha->set_data_rate,
+	    mode);
+
+	if (!IS_FWI2_CAPABLE(ha))
+		return QLA_FUNCTION_FAILED;
+
+	memset(mcp, 0, sizeof(*mcp));
+	switch (ha->set_data_rate) {
+	case PORT_SPEED_AUTO:
+	case PORT_SPEED_4GB:
+	case PORT_SPEED_8GB:
+	case PORT_SPEED_16GB:
+	case PORT_SPEED_32GB:
+		val = ha->set_data_rate;
+		break;
+	default:
+		ql_log(ql_log_warn, vha, 0x1199,
+		    "Unrecognized speed setting:%d. Setting Autoneg\n",
+		    ha->set_data_rate);
+		val = ha->set_data_rate = PORT_SPEED_AUTO;
+		break;
+	}
+
+	mcp->mb[0] = MBC_DATA_RATE;
+	mcp->mb[1] = mode;
+	mcp->mb[2] = val;
+
+	mcp->out_mb = MBX_2|MBX_1|MBX_0;
+	mcp->in_mb = MBX_2|MBX_1|MBX_0;
+	if (IS_QLA83XX(ha) || IS_QLA27XX(ha))
+		mcp->in_mb |= MBX_4|MBX_3;
+	mcp->tov = MBX_TOV_SECONDS;
+	mcp->flags = 0;
+	rval = qla2x00_mailbox_command(vha, mcp);
+	if (rval != QLA_SUCCESS) {
+		ql_dbg(ql_dbg_mbx, vha, 0x1107,
+		    "Failed=%x mb[0]=%x.\n", rval, mcp->mb[0]);
+	} else {
+		if (mcp->mb[1] != 0x7)
+			ql_dbg(ql_dbg_mbx, vha, 0x1179,
+				"Speed set:0x%x\n", mcp->mb[1]);
+
+		ql_dbg(ql_dbg_mbx + ql_dbg_verbose, vha, 0x1108,
+		    "Done %s.\n", __func__);
+	}
+
+	return rval;
+}
+
 int
 qla2x00_get_data_rate(scsi_qla_host_t *vha)
 {
@@ -5263,7 +5313,7 @@ qla2x00_get_data_rate(scsi_qla_host_t *vha)
 		return QLA_FUNCTION_FAILED;
 
 	mcp->mb[0] = MBC_DATA_RATE;
-	mcp->mb[1] = 0;
+	mcp->mb[1] = QLA_GET_DATA_RATE;
 	mcp->out_mb = MBX_1|MBX_0;
 	mcp->in_mb = MBX_2|MBX_1|MBX_0;
 	if (IS_QLA83XX(ha) || IS_QLA27XX(ha))
@@ -6268,8 +6318,6 @@ int __qla24xx_parse_gpdb(struct scsi_qla_host *vha, fc_port_t *fcport,
 	fcport->d_id.b.rsvd_1 = 0;
 
 	if (fcport->fc4f_nvme) {
-		fcport->nvme_prli_service_param =
-		    pd->prli_nvme_svc_param_word_3;
 		fcport->port_type = FCT_NVME;
 	} else {
 		/* If not target must be initiator or unknown type. */
diff --git a/drivers/scsi/qla2xxx/qla_nvme.c b/drivers/scsi/qla2xxx/qla_nvme.c
index 39d892bbd219..41c85da3ab32 100644
--- a/drivers/scsi/qla2xxx/qla_nvme.c
+++ b/drivers/scsi/qla2xxx/qla_nvme.c
@@ -185,6 +185,14 @@ static void qla_nvme_abort_work(struct work_struct *work)
 	struct qla_hw_data *ha = fcport->vha->hw;
 	int rval;
 
+	if (fcport)
+		ql_dbg(ql_dbg_io, fcport->vha, 0xffff,
+		    "%s called for sp=%p, hndl=%x on fcport=%p deleted=%d\n",
+		    __func__, sp, sp->handle, fcport, fcport->deleted);
+
+	if (!ha->flags.fw_started && (fcport && fcport->deleted))
+		return;
+
 	rval = ha->isp_ops->abort_command(sp);
 
 	ql_dbg(ql_dbg_io, fcport->vha, 0x212b,
@@ -358,17 +366,24 @@ static inline int qla2x00_start_nvme_mq(srb_t *sp)
 
 	/* No data transfer how do we check buffer len == 0?? */
 	if (fd->io_dir == NVMEFC_FCP_READ) {
-		cmd_pkt->control_flags =
-		    cpu_to_le16(CF_READ_DATA | CF_NVME_ENABLE);
+		cmd_pkt->control_flags = CF_READ_DATA;
 		vha->qla_stats.input_bytes += fd->payload_length;
 		vha->qla_stats.input_requests++;
 	} else if (fd->io_dir == NVMEFC_FCP_WRITE) {
-		cmd_pkt->control_flags =
-		    cpu_to_le16(CF_WRITE_DATA | CF_NVME_ENABLE);
+		cmd_pkt->control_flags = CF_WRITE_DATA;
+		if ((vha->flags.nvme_first_burst) &&
+		    (sp->fcport->nvme_prli_service_param &
+			NVME_PRLI_SP_FIRST_BURST)) {
+			if ((fd->payload_length <=
+			    sp->fcport->nvme_first_burst_size) ||
+				(sp->fcport->nvme_first_burst_size == 0))
+				cmd_pkt->control_flags |=
+				    CF_NVME_FIRST_BURST_ENABLE;
+		}
 		vha->qla_stats.output_bytes += fd->payload_length;
 		vha->qla_stats.output_requests++;
 	} else if (fd->io_dir == 0) {
-		cmd_pkt->control_flags = cpu_to_le16(CF_NVME_ENABLE);
+		cmd_pkt->control_flags = 0;
 	}
 
 	/* Set NPORT-ID */
@@ -600,6 +615,7 @@ static void qla_nvme_unregister_remote_port(struct work_struct *work)
 	struct fc_port *fcport = container_of(work, struct fc_port,
 	    nvme_del_work);
 	struct qla_nvme_rport *qla_rport, *trport;
+	scsi_qla_host_t *base_vha;
 
 	if (!IS_ENABLED(CONFIG_NVME_FC))
 		return;
@@ -607,6 +623,15 @@ static void qla_nvme_unregister_remote_port(struct work_struct *work)
 	ql_log(ql_log_warn, NULL, 0x2112,
 	    "%s: unregister remoteport on %p\n",__func__, fcport);
 
+	base_vha = pci_get_drvdata(fcport->vha->hw->pdev);
+	if (test_bit(PFLG_DRIVER_REMOVING, &base_vha->pci_flags)) {
+		ql_dbg(ql_dbg_disc, fcport->vha, 0x2114,
+		    "%s: Notify FC-NVMe transport, set devloss=0\n",
+		    __func__);
+
+		nvme_fc_set_remoteport_devloss(fcport->nvme_remote_port, 0);
+	}
+
 	list_for_each_entry_safe(qla_rport, trport,
 	    &fcport->vha->nvme_rport_list, list) {
 		if (qla_rport->fcport == fcport) {
@@ -623,23 +648,11 @@ static void qla_nvme_unregister_remote_port(struct work_struct *work)
 
 void qla_nvme_delete(struct scsi_qla_host *vha)
 {
-	struct qla_nvme_rport *qla_rport, *trport;
-	fc_port_t *fcport;
 	int nv_ret;
 
 	if (!IS_ENABLED(CONFIG_NVME_FC))
 		return;
 
-	list_for_each_entry_safe(qla_rport, trport,
-	    &vha->nvme_rport_list, list) {
-		fcport = qla_rport->fcport;
-
-		ql_log(ql_log_info, fcport->vha, 0x2114, "%s: fcport=%p\n",
-		    __func__, fcport);
-
-		nvme_fc_set_remoteport_devloss(fcport->nvme_remote_port, 0);
-	}
-
 	if (vha->nvme_local_port) {
 		init_completion(&vha->nvme_del_done);
 		ql_log(ql_log_info, vha, 0x2116,
diff --git a/drivers/scsi/qla2xxx/qla_nvme.h b/drivers/scsi/qla2xxx/qla_nvme.h
index 4941d107fb1c..da8dad5ad693 100644
--- a/drivers/scsi/qla2xxx/qla_nvme.h
+++ b/drivers/scsi/qla2xxx/qla_nvme.h
@@ -57,7 +57,7 @@ struct cmd_nvme {
 	uint64_t rsvd;
 
 	uint16_t control_flags;         /* Control Flags */
-#define CF_NVME_ENABLE                  BIT_9
+#define CF_NVME_FIRST_BURST_ENABLE	BIT_11
 #define CF_DIF_SEG_DESCR_ENABLE         BIT_3
 #define CF_DATA_SEG_DESCR_ENABLE        BIT_2
 #define CF_READ_DATA                    BIT_1
diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
index c6ef83d0d99b..677f82fdf56f 100644
--- a/drivers/scsi/qla2xxx/qla_os.c
+++ b/drivers/scsi/qla2xxx/qla_os.c
@@ -285,6 +285,27 @@ MODULE_PARM_DESC(qla2xuseresexchforels,
 		 "Reserve 1/2 of emergency exchanges for ELS.\n"
 		 " 0 (default): disabled");
 
+int ql2xprotmask;
+module_param(ql2xprotmask, int, 0644);
+MODULE_PARM_DESC(ql2xprotmask,
+		 "Override DIF/DIX protection capabilities mask\n"
+		 "Default is 0 which sets protection mask based on "
+		 "capabilities reported by HBA firmware.\n");
+
+int ql2xprotguard;
+module_param(ql2xprotguard, int, 0644);
+MODULE_PARM_DESC(ql2xprotguard, "Override choice of DIX checksum\n"
+		 "  0 -- Let HBA firmware decide\n"
+		 "  1 -- Force T10 CRC\n"
+		 "  2 -- Force IP checksum\n");
+
+int ql2xdifbundlinginternalbuffers;
+module_param(ql2xdifbundlinginternalbuffers, int, 0644);
+MODULE_PARM_DESC(ql2xdifbundlinginternalbuffers,
+    "Force using internal buffers for DIF information\n"
+    "0 (Default). Based on check.\n"
+    "1 Force using internal buffers\n");
+
 /*
  * SCSI host template entry points
  */
@@ -804,7 +825,44 @@ qla2xxx_qpair_sp_free_dma(void *ptr)
 		ha->gbl_dsd_inuse -= ctx1->dsd_use_cnt;
 		ha->gbl_dsd_avail += ctx1->dsd_use_cnt;
 		mempool_free(ctx1, ha->ctx_mempool);
+		sp->flags &= ~SRB_FCP_CMND_DMA_VALID;
+	}
+	if (sp->flags & SRB_DIF_BUNDL_DMA_VALID) {
+		struct crc_context *difctx = sp->u.scmd.ctx;
+		struct dsd_dma *dif_dsd, *nxt_dsd;
+
+		list_for_each_entry_safe(dif_dsd, nxt_dsd,
+		    &difctx->ldif_dma_hndl_list, list) {
+			list_del(&dif_dsd->list);
+			dma_pool_free(ha->dif_bundl_pool, dif_dsd->dsd_addr,
+			    dif_dsd->dsd_list_dma);
+			kfree(dif_dsd);
+			difctx->no_dif_bundl--;
+		}
+
+		list_for_each_entry_safe(dif_dsd, nxt_dsd,
+		    &difctx->ldif_dsd_list, list) {
+			list_del(&dif_dsd->list);
+			dma_pool_free(ha->dl_dma_pool, dif_dsd->dsd_addr,
+			    dif_dsd->dsd_list_dma);
+			kfree(dif_dsd);
+			difctx->no_ldif_dsd--;
+		}
+
+		if (difctx->no_ldif_dsd) {
+			ql_dbg(ql_dbg_tgt+ql_dbg_verbose, sp->vha, 0xe022,
+			    "%s: difctx->no_ldif_dsd=%x\n",
+			    __func__, difctx->no_ldif_dsd);
+		}
+
+		if (difctx->no_dif_bundl) {
+			ql_dbg(ql_dbg_tgt+ql_dbg_verbose, sp->vha, 0xe022,
+			    "%s: difctx->no_dif_bundl=%x\n",
+			    __func__, difctx->no_dif_bundl);
+		}
+		sp->flags &= ~SRB_DIF_BUNDL_DMA_VALID;
 	}
+
 end:
 	CMD_SP(cmd) = NULL;
 	qla2xxx_rel_qpair_sp(sp->qpair, sp);
@@ -3342,13 +3400,16 @@ skip_dpc:
 			    "Registering for DIF/DIX type 1 and 3 protection.\n");
 			if (ql2xenabledif == 1)
 				prot = SHOST_DIX_TYPE0_PROTECTION;
-			scsi_host_set_prot(host,
-			    prot | SHOST_DIF_TYPE1_PROTECTION
-			    | SHOST_DIF_TYPE2_PROTECTION
-			    | SHOST_DIF_TYPE3_PROTECTION
-			    | SHOST_DIX_TYPE1_PROTECTION
-			    | SHOST_DIX_TYPE2_PROTECTION
-			    | SHOST_DIX_TYPE3_PROTECTION);
+			if (ql2xprotmask)
+				scsi_host_set_prot(host, ql2xprotmask);
+			else
+				scsi_host_set_prot(host,
+				    prot | SHOST_DIF_TYPE1_PROTECTION
+				    | SHOST_DIF_TYPE2_PROTECTION
+				    | SHOST_DIF_TYPE3_PROTECTION
+				    | SHOST_DIX_TYPE1_PROTECTION
+				    | SHOST_DIX_TYPE2_PROTECTION
+				    | SHOST_DIX_TYPE3_PROTECTION);
 
 			guard = SHOST_DIX_GUARD_CRC;
 
@@ -3356,7 +3417,10 @@ skip_dpc:
 			    (ql2xenabledif > 1 || IS_PI_DIFB_DIX0_CAPABLE(ha)))
 				guard |= SHOST_DIX_GUARD_IP;
 
-			scsi_host_set_guard(host, guard);
+			if (ql2xprotguard)
+				scsi_host_set_guard(host, ql2xprotguard);
+			else
+				scsi_host_set_guard(host, guard);
 		} else
 			base_vha->flags.difdix_supported = 0;
 	}
@@ -3997,9 +4061,86 @@ qla2x00_mem_alloc(struct qla_hw_data *ha, uint16_t req_len, uint16_t rsp_len,
 			    "Failed to allocate memory for fcp_cmnd_dma_pool.\n");
 			goto fail_dl_dma_pool;
 		}
+
+		if (ql2xenabledif) {
+			u64 bufsize = DIF_BUNDLING_DMA_POOL_SIZE;
+			struct dsd_dma *dsd, *nxt;
+			uint i;
+			/* Creata a DMA pool of buffers for DIF bundling */
+			ha->dif_bundl_pool = dma_pool_create(name,
+			    &ha->pdev->dev, DIF_BUNDLING_DMA_POOL_SIZE, 8, 0);
+			if (!ha->dif_bundl_pool) {
+				ql_dbg_pci(ql_dbg_init, ha->pdev, 0x0024,
+				    "%s: failed create dif_bundl_pool\n",
+				    __func__);
+				goto fail_dif_bundl_dma_pool;
+			}
+
+			INIT_LIST_HEAD(&ha->pool.good.head);
+			INIT_LIST_HEAD(&ha->pool.unusable.head);
+			ha->pool.good.count = 0;
+			ha->pool.unusable.count = 0;
+			for (i = 0; i < 128; i++) {
+				dsd = kzalloc(sizeof(*dsd), GFP_ATOMIC);
+				if (!dsd) {
+					ql_dbg_pci(ql_dbg_init, ha->pdev,
+					    0xe0ee, "%s: failed alloc dsd\n",
+					    __func__);
+					return 1;
+				}
+				ha->dif_bundle_kallocs++;
+
+				dsd->dsd_addr = dma_pool_alloc(
+				    ha->dif_bundl_pool, GFP_ATOMIC,
+				    &dsd->dsd_list_dma);
+				if (!dsd->dsd_addr) {
+					ql_dbg_pci(ql_dbg_init, ha->pdev,
+					    0xe0ee,
+					    "%s: failed alloc ->dsd_addr\n",
+					    __func__);
+					kfree(dsd);
+					ha->dif_bundle_kallocs--;
+					continue;
+				}
+				ha->dif_bundle_dma_allocs++;
+
+				/*
+				 * if DMA buffer crosses 4G boundary,
+				 * put it on bad list
+				 */
+				if (MSD(dsd->dsd_list_dma) ^
+				    MSD(dsd->dsd_list_dma + bufsize)) {
+					list_add_tail(&dsd->list,
+					    &ha->pool.unusable.head);
+					ha->pool.unusable.count++;
+				} else {
+					list_add_tail(&dsd->list,
+					    &ha->pool.good.head);
+					ha->pool.good.count++;
+				}
+			}
+
+			/* return the good ones back to the pool */
+			list_for_each_entry_safe(dsd, nxt,
+			    &ha->pool.good.head, list) {
+				list_del(&dsd->list);
+				dma_pool_free(ha->dif_bundl_pool,
+				    dsd->dsd_addr, dsd->dsd_list_dma);
+				ha->dif_bundle_dma_allocs--;
+				kfree(dsd);
+				ha->dif_bundle_kallocs--;
+			}
+
+			ql_dbg_pci(ql_dbg_init, ha->pdev, 0x0024,
+			    "%s: dif dma pool (good=%u unusable=%u)\n",
+			    __func__, ha->pool.good.count,
+			    ha->pool.unusable.count);
+		}
+
 		ql_dbg_pci(ql_dbg_init, ha->pdev, 0x0025,
-		    "dl_dma_pool=%p fcp_cmnd_dma_pool=%p.\n",
-		    ha->dl_dma_pool, ha->fcp_cmnd_dma_pool);
+		    "dl_dma_pool=%p fcp_cmnd_dma_pool=%p dif_bundl_pool=%p.\n",
+		    ha->dl_dma_pool, ha->fcp_cmnd_dma_pool,
+		    ha->dif_bundl_pool);
 	}
 
 	/* Allocate memory for SNS commands */
@@ -4164,6 +4305,24 @@ fail_free_ms_iocb:
 		dma_free_coherent(&ha->pdev->dev, sizeof(struct sns_cmd_pkt),
 		    ha->sns_cmd, ha->sns_cmd_dma);
 fail_dma_pool:
+	if (ql2xenabledif) {
+		struct dsd_dma *dsd, *nxt;
+
+		list_for_each_entry_safe(dsd, nxt, &ha->pool.unusable.head,
+		    list) {
+			list_del(&dsd->list);
+			dma_pool_free(ha->dif_bundl_pool, dsd->dsd_addr,
+			    dsd->dsd_list_dma);
+			ha->dif_bundle_dma_allocs--;
+			kfree(dsd);
+			ha->dif_bundle_kallocs--;
+			ha->pool.unusable.count--;
+		}
+		dma_pool_destroy(ha->dif_bundl_pool);
+		ha->dif_bundl_pool = NULL;
+	}
+
+fail_dif_bundl_dma_pool:
 	if (IS_QLA82XX(ha) || ql2xenabledif) {
 		dma_pool_destroy(ha->fcp_cmnd_dma_pool);
 		ha->fcp_cmnd_dma_pool = NULL;
@@ -4544,6 +4703,32 @@ qla2x00_mem_free(struct qla_hw_data *ha)
 
 	mempool_destroy(ha->ctx_mempool);
 
+	if (ql2xenabledif) {
+		struct dsd_dma *dsd, *nxt;
+
+		list_for_each_entry_safe(dsd, nxt, &ha->pool.unusable.head,
+					 list) {
+			list_del(&dsd->list);
+			dma_pool_free(ha->dif_bundl_pool, dsd->dsd_addr,
+				      dsd->dsd_list_dma);
+			ha->dif_bundle_dma_allocs--;
+			kfree(dsd);
+			ha->dif_bundle_kallocs--;
+			ha->pool.unusable.count--;
+		}
+		list_for_each_entry_safe(dsd, nxt, &ha->pool.good.head, list) {
+			list_del(&dsd->list);
+			dma_pool_free(ha->dif_bundl_pool, dsd->dsd_addr,
+				      dsd->dsd_list_dma);
+			ha->dif_bundle_dma_allocs--;
+			kfree(dsd);
+			ha->dif_bundle_kallocs--;
+		}
+	}
+
+	if (ha->dif_bundl_pool)
+		dma_pool_destroy(ha->dif_bundl_pool);
+
 	qlt_mem_free(ha);
 
 	if (ha->init_cb)
@@ -5019,14 +5204,14 @@ qla2x00_do_work(struct scsi_qla_host *vha)
 	struct qla_work_evt *e, *tmp;
 	unsigned long flags;
 	LIST_HEAD(work);
+	int rc;
 
 	spin_lock_irqsave(&vha->work_lock, flags);
 	list_splice_init(&vha->work_list, &work);
 	spin_unlock_irqrestore(&vha->work_lock, flags);
 
 	list_for_each_entry_safe(e, tmp, &work, list) {
-		list_del_init(&e->list);
-
+		rc = QLA_SUCCESS;
 		switch (e->type) {
 		case QLA_EVT_AEN:
 			fc_host_post_event(vha->host, fc_get_event_number(),
@@ -5040,7 +5225,7 @@ qla2x00_do_work(struct scsi_qla_host *vha)
 			    e->u.logio.data);
 			break;
 		case QLA_EVT_ASYNC_LOGOUT:
-			qla2x00_async_logout(vha, e->u.logio.fcport);
+			rc = qla2x00_async_logout(vha, e->u.logio.fcport);
 			break;
 		case QLA_EVT_ASYNC_LOGOUT_DONE:
 			qla2x00_async_logout_done(vha, e->u.logio.fcport,
@@ -5085,7 +5270,7 @@ qla2x00_do_work(struct scsi_qla_host *vha)
 			qla24xx_do_nack_work(vha, e);
 			break;
 		case QLA_EVT_ASYNC_PRLO:
-			qla2x00_async_prlo(vha, e->u.logio.fcport);
+			rc = qla2x00_async_prlo(vha, e->u.logio.fcport);
 			break;
 		case QLA_EVT_ASYNC_PRLO_DONE:
 			qla2x00_async_prlo_done(vha, e->u.logio.fcport,
@@ -5118,6 +5303,15 @@ qla2x00_do_work(struct scsi_qla_host *vha)
 			    e->u.fcport.fcport, false);
 			break;
 		}
+
+		if (rc == EAGAIN) {
+			/* put 'work' at head of 'vha->work_list' */
+			spin_lock_irqsave(&vha->work_lock, flags);
+			list_splice(&work, &vha->work_list);
+			spin_unlock_irqrestore(&vha->work_lock, flags);
+			break;
+		}
+		list_del_init(&e->list);
 		if (e->flags & QLA_EVT_FLAG_FREE)
 			kfree(e);
 
@@ -6930,13 +7124,64 @@ qla2xxx_pci_resume(struct pci_dev *pdev)
 	ha->flags.eeh_busy = 0;
 }
 
+static void
+qla_pci_reset_prepare(struct pci_dev *pdev)
+{
+	scsi_qla_host_t *base_vha = pci_get_drvdata(pdev);
+	struct qla_hw_data *ha = base_vha->hw;
+	struct qla_qpair *qpair;
+
+	ql_log(ql_log_warn, base_vha, 0xffff,
+	    "%s.\n", __func__);
+
+	/*
+	 * PCI FLR/function reset is about to reset the
+	 * slot. Stop the chip to stop all DMA access.
+	 * It is assumed that pci_reset_done will be called
+	 * after FLR to resume Chip operation.
+	 */
+	ha->flags.eeh_busy = 1;
+	mutex_lock(&ha->mq_lock);
+	list_for_each_entry(qpair, &base_vha->qp_list, qp_list_elem)
+		qpair->online = 0;
+	mutex_unlock(&ha->mq_lock);
+
+	set_bit(ABORT_ISP_ACTIVE, &base_vha->dpc_flags);
+	qla2x00_abort_isp_cleanup(base_vha);
+	qla2x00_abort_all_cmds(base_vha, DID_RESET << 16);
+}
+
+static void
+qla_pci_reset_done(struct pci_dev *pdev)
+{
+	scsi_qla_host_t *base_vha = pci_get_drvdata(pdev);
+	struct qla_hw_data *ha = base_vha->hw;
+	struct qla_qpair *qpair;
+
+	ql_log(ql_log_warn, base_vha, 0xffff,
+	    "%s.\n", __func__);
+
+	/*
+	 * FLR just completed by PCI layer. Resume adapter
+	 */
+	ha->flags.eeh_busy = 0;
+	mutex_lock(&ha->mq_lock);
+	list_for_each_entry(qpair, &base_vha->qp_list, qp_list_elem)
+		qpair->online = 1;
+	mutex_unlock(&ha->mq_lock);
+
+	base_vha->flags.online = 1;
+	ha->isp_ops->abort_isp(base_vha);
+	clear_bit(ABORT_ISP_ACTIVE, &base_vha->dpc_flags);
+}
+
 static int qla2xxx_map_queues(struct Scsi_Host *shost)
 {
 	int rc;
 	scsi_qla_host_t *vha = (scsi_qla_host_t *)shost->hostdata;
 	struct blk_mq_queue_map *qmap = &shost->tag_set.map[0];
 
-	if (USER_CTRL_IRQ(vha->hw))
+	if (USER_CTRL_IRQ(vha->hw) || !vha->hw->mqiobase)
 		rc = blk_mq_map_queues(qmap);
 	else
 		rc = blk_mq_pci_map_queues(qmap, vha->hw->pdev, vha->irq_offset);
@@ -6948,6 +7193,8 @@ static const struct pci_error_handlers qla2xxx_err_handler = {
 	.mmio_enabled = qla2xxx_pci_mmio_enabled,
 	.slot_reset = qla2xxx_pci_slot_reset,
 	.resume = qla2xxx_pci_resume,
+	.reset_prepare = qla_pci_reset_prepare,
+	.reset_done = qla_pci_reset_done,
 };
 
 static struct pci_device_id qla2xxx_pci_tbl[] = {
diff --git a/drivers/scsi/qla2xxx/qla_target.c b/drivers/scsi/qla2xxx/qla_target.c
index 510337eac106..582d1663f971 100644
--- a/drivers/scsi/qla2xxx/qla_target.c
+++ b/drivers/scsi/qla2xxx/qla_target.c
@@ -660,14 +660,14 @@ int qla24xx_async_notify_ack(scsi_qla_host_t *vha, fc_port_t *fcport,
 	sp->u.iocb_cmd.u.nack.ntfy = ntfy;
 	sp->done = qla2x00_async_nack_sp_done;
 
-	rval = qla2x00_start_sp(sp);
-	if (rval != QLA_SUCCESS)
-		goto done_free_sp;
-
 	ql_dbg(ql_dbg_disc, vha, 0x20f4,
 	    "Async-%s %8phC hndl %x %s\n",
 	    sp->name, fcport->port_name, sp->handle, c);
 
+	rval = qla2x00_start_sp(sp);
+	if (rval != QLA_SUCCESS)
+		goto done_free_sp;
+
 	return rval;
 
 done_free_sp:
@@ -684,6 +684,9 @@ void qla24xx_do_nack_work(struct scsi_qla_host *vha, struct qla_work_evt *e)
 
 	switch (e->u.nack.type) {
 	case SRB_NACK_PRLI:
+		t = e->u.nack.fcport;
+		flush_work(&t->del_work);
+		flush_work(&t->free_work);
 		mutex_lock(&vha->vha_tgt.tgt_mutex);
 		t = qlt_create_sess(vha, e->u.nack.fcport, 0);
 		mutex_unlock(&vha->vha_tgt.tgt_mutex);
@@ -3230,7 +3233,7 @@ qlt_build_ctio_crc2_pkt(struct qla_qpair *qpair, struct qla_tgt_prm *prm)
 
 		cur_dsd = (uint32_t *) &crc_ctx_pkt->u.bundling.dif_address;
 		if (qla24xx_walk_and_build_prot_sglist(ha, NULL, cur_dsd,
-			prm->prot_seg_cnt, &tc))
+			prm->prot_seg_cnt, cmd))
 			goto crc_queuing_error;
 	}
 	return QLA_SUCCESS;
@@ -3257,13 +3260,10 @@ int qlt_xmit_response(struct qla_tgt_cmd *cmd, int xmit_type,
 	unsigned long flags = 0;
 	int res;
 
-	if (cmd->sess && cmd->sess->deleted) {
+	if (!qpair->fw_started || (cmd->reset_count != qpair->chip_reset) ||
+	    (cmd->sess && cmd->sess->deleted)) {
 		cmd->state = QLA_TGT_STATE_PROCESSED;
-		if (cmd->sess->logout_completed)
-			/* no need to terminate. FW already freed exchange. */
-			qlt_abort_cmd_on_host_reset(cmd->vha, cmd);
-		else
-			qlt_send_term_exchange(qpair, cmd, &cmd->atio, 0, 0);
+		qlt_abort_cmd_on_host_reset(cmd->vha, cmd);
 		return 0;
 	}
 
@@ -6343,7 +6343,7 @@ static void qlt_tmr_work(struct qla_tgt *tgt,
 	struct atio_from_isp *a = &prm->tm_iocb2;
 	struct scsi_qla_host *vha = tgt->vha;
 	struct qla_hw_data *ha = vha->hw;
-	struct fc_port *sess = NULL;
+	struct fc_port *sess;
 	unsigned long flags;
 	uint8_t *s_id = NULL; /* to hide compiler warnings */
 	int rc;
@@ -6369,7 +6369,6 @@ static void qlt_tmr_work(struct qla_tgt *tgt,
 			goto out_term2;
 	} else {
 		if (sess->deleted) {
-			sess = NULL;
 			goto out_term2;
 		}
 
@@ -6377,7 +6376,6 @@ static void qlt_tmr_work(struct qla_tgt *tgt,
 			ql_dbg(ql_dbg_tgt_tmr, vha, 0xf020,
 			    "%s: kref_get fail %8phC\n",
 			     __func__, sess->port_name);
-			sess = NULL;
 			goto out_term2;
 		}
 	}
@@ -6396,8 +6394,6 @@ static void qlt_tmr_work(struct qla_tgt *tgt,
 	return;
 
 out_term2:
-	if (sess)
-		ha->tgt.tgt_ops->put_sess(sess);
 	spin_unlock_irqrestore(&ha->tgt.sess_lock, flags);
 out_term:
 	qlt_send_term_exchange(ha->base_qpair, NULL, &prm->tm_iocb2, 1, 0);
diff --git a/drivers/scsi/qla2xxx/qla_target.h b/drivers/scsi/qla2xxx/qla_target.h
index 577e1786a3f1..f3de75000a08 100644
--- a/drivers/scsi/qla2xxx/qla_target.h
+++ b/drivers/scsi/qla2xxx/qla_target.h
@@ -928,6 +928,8 @@ struct qla_tgt_cmd {
 	uint64_t	lba;
 	uint16_t	a_guard, e_guard, a_app_tag, e_app_tag;
 	uint32_t	a_ref_tag, e_ref_tag;
+#define DIF_BUNDL_DMA_VALID 1
+	uint16_t prot_flags;
 
 	uint64_t jiffies_at_alloc;
 	uint64_t jiffies_at_free;
diff --git a/drivers/scsi/qla2xxx/qla_tmpl.c b/drivers/scsi/qla2xxx/qla_tmpl.c
index 0ccd06f11f12..9e52500caff0 100644
--- a/drivers/scsi/qla2xxx/qla_tmpl.c
+++ b/drivers/scsi/qla2xxx/qla_tmpl.c
@@ -221,7 +221,13 @@ qla27xx_skip_entry(struct qla27xx_fwdt_entry *ent, void *buf)
 		ent->hdr.driver_flags |= DRIVER_FLAG_SKIP_ENTRY;
 }
 
-static int
+static inline struct qla27xx_fwdt_entry *
+qla27xx_next_entry(struct qla27xx_fwdt_entry *ent)
+{
+	return (void *)ent + ent->hdr.size;
+}
+
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t0(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -229,10 +235,10 @@ qla27xx_fwdt_entry_t0(struct scsi_qla_host *vha,
 	    "%s: nop [%lx]\n", __func__, *len);
 	qla27xx_skip_entry(ent, buf);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t255(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -241,10 +247,10 @@ qla27xx_fwdt_entry_t255(struct scsi_qla_host *vha,
 	qla27xx_skip_entry(ent, buf);
 
 	/* terminate */
-	return true;
+	return NULL;
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t256(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -255,10 +261,10 @@ qla27xx_fwdt_entry_t256(struct scsi_qla_host *vha,
 	qla27xx_read_window(reg, ent->t256.base_addr, ent->t256.pci_offset,
 	    ent->t256.reg_count, ent->t256.reg_width, buf, len);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t257(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -269,10 +275,10 @@ qla27xx_fwdt_entry_t257(struct scsi_qla_host *vha,
 	qla27xx_write_reg(reg, IOBASE_ADDR, ent->t257.base_addr, buf);
 	qla27xx_write_reg(reg, ent->t257.pci_offset, ent->t257.write_data, buf);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t258(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -284,10 +290,10 @@ qla27xx_fwdt_entry_t258(struct scsi_qla_host *vha,
 	qla27xx_read_window(reg, ent->t258.base_addr, ent->t258.pci_offset,
 	    ent->t258.reg_count, ent->t258.reg_width, buf, len);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t259(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -299,10 +305,10 @@ qla27xx_fwdt_entry_t259(struct scsi_qla_host *vha,
 	qla27xx_write_reg(reg, ent->t259.banksel_offset, ent->t259.bank, buf);
 	qla27xx_write_reg(reg, ent->t259.pci_offset, ent->t259.write_data, buf);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t260(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -313,10 +319,10 @@ qla27xx_fwdt_entry_t260(struct scsi_qla_host *vha,
 	qla27xx_insert32(ent->t260.pci_offset, buf, len);
 	qla27xx_read_reg(reg, ent->t260.pci_offset, buf, len);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t261(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -326,10 +332,10 @@ qla27xx_fwdt_entry_t261(struct scsi_qla_host *vha,
 	    "%s: wrpci [%lx]\n", __func__, *len);
 	qla27xx_write_reg(reg, ent->t261.pci_offset, ent->t261.write_data, buf);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t262(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -362,6 +368,11 @@ qla27xx_fwdt_entry_t262(struct scsi_qla_host *vha,
 			ent->t262.start_addr = start;
 			ent->t262.end_addr = end;
 		}
+	} else if (ent->t262.ram_area == T262_RAM_AREA_MISC) {
+		if (buf) {
+			ent->t262.start_addr = start;
+			ent->t262.end_addr = end;
+		}
 	} else {
 		ql_dbg(ql_dbg_misc, vha, 0xd022,
 		    "%s: unknown area %x\n", __func__, ent->t262.ram_area);
@@ -384,10 +395,10 @@ qla27xx_fwdt_entry_t262(struct scsi_qla_host *vha,
 	}
 	*len += dwords * sizeof(uint32_t);
 done:
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t263(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -450,10 +461,10 @@ qla27xx_fwdt_entry_t263(struct scsi_qla_host *vha,
 			qla27xx_skip_entry(ent, buf);
 	}
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t264(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -478,10 +489,10 @@ qla27xx_fwdt_entry_t264(struct scsi_qla_host *vha,
 		qla27xx_skip_entry(ent, buf);
 	}
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t265(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -492,10 +503,10 @@ qla27xx_fwdt_entry_t265(struct scsi_qla_host *vha,
 	if (buf)
 		qla24xx_pause_risc(reg, vha->hw);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t266(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -504,10 +515,10 @@ qla27xx_fwdt_entry_t266(struct scsi_qla_host *vha,
 	if (buf)
 		qla24xx_soft_reset(vha->hw);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t267(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -517,10 +528,10 @@ qla27xx_fwdt_entry_t267(struct scsi_qla_host *vha,
 	    "%s: dis intr [%lx]\n", __func__, *len);
 	qla27xx_write_reg(reg, ent->t267.pci_offset, ent->t267.data, buf);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t268(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -587,10 +598,10 @@ qla27xx_fwdt_entry_t268(struct scsi_qla_host *vha,
 		break;
 	}
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t269(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -604,10 +615,10 @@ qla27xx_fwdt_entry_t269(struct scsi_qla_host *vha,
 	if (buf)
 		ent->t269.scratch_size = 5 * sizeof(uint32_t);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t270(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -625,10 +636,10 @@ qla27xx_fwdt_entry_t270(struct scsi_qla_host *vha,
 		addr += sizeof(uint32_t);
 	}
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t271(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -642,10 +653,10 @@ qla27xx_fwdt_entry_t271(struct scsi_qla_host *vha,
 	qla27xx_write_reg(reg, 0xc4, data, buf);
 	qla27xx_write_reg(reg, 0xc0, addr, buf);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t272(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -662,10 +673,10 @@ qla27xx_fwdt_entry_t272(struct scsi_qla_host *vha,
 	}
 	*len += dwords * sizeof(uint32_t);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t273(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -685,10 +696,10 @@ qla27xx_fwdt_entry_t273(struct scsi_qla_host *vha,
 		addr += sizeof(uint32_t);
 	}
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t274(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -746,10 +757,10 @@ qla27xx_fwdt_entry_t274(struct scsi_qla_host *vha,
 			qla27xx_skip_entry(ent, buf);
 	}
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_t275(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
@@ -763,7 +774,7 @@ qla27xx_fwdt_entry_t275(struct scsi_qla_host *vha,
 		qla27xx_skip_entry(ent, buf);
 		goto done;
 	}
-	if (offset + ent->t275.length > ent->hdr.entry_size) {
+	if (offset + ent->t275.length > ent->hdr.size) {
 		ql_dbg(ql_dbg_misc, vha, 0xd030,
 		    "%s: buffer overflow\n", __func__);
 		qla27xx_skip_entry(ent, buf);
@@ -772,59 +783,103 @@ qla27xx_fwdt_entry_t275(struct scsi_qla_host *vha,
 
 	qla27xx_insertbuf(ent->t275.buffer, ent->t275.length, buf, len);
 done:
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-static int
+static struct qla27xx_fwdt_entry *
+qla27xx_fwdt_entry_t276(struct scsi_qla_host *vha,
+    struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
+{
+	uint type = vha->hw->pdev->device >> 4 & 0xf;
+	uint func = vha->hw->port_no & 0x3;
+
+	ql_dbg(ql_dbg_misc + ql_dbg_verbose, vha, 0xd214,
+	    "%s: cond [%lx]\n", __func__, *len);
+
+	if (type != ent->t276.cond1 || func != ent->t276.cond2) {
+		ent = qla27xx_next_entry(ent);
+		qla27xx_skip_entry(ent, buf);
+	}
+
+	return qla27xx_next_entry(ent);
+}
+
+static struct qla27xx_fwdt_entry *
+qla27xx_fwdt_entry_t277(struct scsi_qla_host *vha,
+    struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
+{
+	struct device_reg_24xx __iomem *reg = qla27xx_isp_reg(vha);
+
+	ql_dbg(ql_dbg_misc + ql_dbg_verbose, vha, 0xd215,
+	    "%s: rdpep [%lx]\n", __func__, *len);
+	qla27xx_insert32(ent->t277.wr_cmd_data, buf, len);
+	qla27xx_write_reg(reg, ent->t277.cmd_addr, ent->t277.wr_cmd_data, buf);
+	qla27xx_read_reg(reg, ent->t277.data_addr, buf, len);
+
+	return qla27xx_next_entry(ent);
+}
+
+static struct qla27xx_fwdt_entry *
+qla27xx_fwdt_entry_t278(struct scsi_qla_host *vha,
+    struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
+{
+	struct device_reg_24xx __iomem *reg = qla27xx_isp_reg(vha);
+
+	ql_dbg(ql_dbg_misc + ql_dbg_verbose, vha, 0xd216,
+	    "%s: wrpep [%lx]\n", __func__, *len);
+	qla27xx_write_reg(reg, ent->t278.data_addr, ent->t278.wr_data, buf);
+	qla27xx_write_reg(reg, ent->t278.cmd_addr, ent->t278.wr_cmd_data, buf);
+
+	return qla27xx_next_entry(ent);
+}
+
+static struct qla27xx_fwdt_entry *
 qla27xx_fwdt_entry_other(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_entry *ent, void *buf, ulong *len)
 {
 	ql_dbg(ql_dbg_misc, vha, 0xd2ff,
-	    "%s: type %x [%lx]\n", __func__, ent->hdr.entry_type, *len);
+	    "%s: type %x [%lx]\n", __func__, ent->hdr.type, *len);
 	qla27xx_skip_entry(ent, buf);
 
-	return false;
+	return qla27xx_next_entry(ent);
 }
 
-struct qla27xx_fwdt_entry_call {
+static struct {
 	uint type;
-	int (*call)(
-	    struct scsi_qla_host *,
-	    struct qla27xx_fwdt_entry *,
-	    void *,
-	    ulong *);
-};
-
-static struct qla27xx_fwdt_entry_call ql27xx_fwdt_entry_call_list[] = {
-	{ ENTRY_TYPE_NOP		, qla27xx_fwdt_entry_t0    } ,
-	{ ENTRY_TYPE_TMP_END		, qla27xx_fwdt_entry_t255  } ,
-	{ ENTRY_TYPE_RD_IOB_T1		, qla27xx_fwdt_entry_t256  } ,
-	{ ENTRY_TYPE_WR_IOB_T1		, qla27xx_fwdt_entry_t257  } ,
-	{ ENTRY_TYPE_RD_IOB_T2		, qla27xx_fwdt_entry_t258  } ,
-	{ ENTRY_TYPE_WR_IOB_T2		, qla27xx_fwdt_entry_t259  } ,
-	{ ENTRY_TYPE_RD_PCI		, qla27xx_fwdt_entry_t260  } ,
-	{ ENTRY_TYPE_WR_PCI		, qla27xx_fwdt_entry_t261  } ,
-	{ ENTRY_TYPE_RD_RAM		, qla27xx_fwdt_entry_t262  } ,
-	{ ENTRY_TYPE_GET_QUEUE		, qla27xx_fwdt_entry_t263  } ,
-	{ ENTRY_TYPE_GET_FCE		, qla27xx_fwdt_entry_t264  } ,
-	{ ENTRY_TYPE_PSE_RISC		, qla27xx_fwdt_entry_t265  } ,
-	{ ENTRY_TYPE_RST_RISC		, qla27xx_fwdt_entry_t266  } ,
-	{ ENTRY_TYPE_DIS_INTR		, qla27xx_fwdt_entry_t267  } ,
-	{ ENTRY_TYPE_GET_HBUF		, qla27xx_fwdt_entry_t268  } ,
-	{ ENTRY_TYPE_SCRATCH		, qla27xx_fwdt_entry_t269  } ,
-	{ ENTRY_TYPE_RDREMREG		, qla27xx_fwdt_entry_t270  } ,
-	{ ENTRY_TYPE_WRREMREG		, qla27xx_fwdt_entry_t271  } ,
-	{ ENTRY_TYPE_RDREMRAM		, qla27xx_fwdt_entry_t272  } ,
-	{ ENTRY_TYPE_PCICFG		, qla27xx_fwdt_entry_t273  } ,
-	{ ENTRY_TYPE_GET_SHADOW		, qla27xx_fwdt_entry_t274  } ,
-	{ ENTRY_TYPE_WRITE_BUF		, qla27xx_fwdt_entry_t275  } ,
-	{ -1				, qla27xx_fwdt_entry_other }
+	typeof(qla27xx_fwdt_entry_other)(*call);
+} qla27xx_fwdt_entry_call[] = {
+	{ ENTRY_TYPE_NOP,		qla27xx_fwdt_entry_t0    },
+	{ ENTRY_TYPE_TMP_END,		qla27xx_fwdt_entry_t255  },
+	{ ENTRY_TYPE_RD_IOB_T1,		qla27xx_fwdt_entry_t256  },
+	{ ENTRY_TYPE_WR_IOB_T1,		qla27xx_fwdt_entry_t257  },
+	{ ENTRY_TYPE_RD_IOB_T2,		qla27xx_fwdt_entry_t258  },
+	{ ENTRY_TYPE_WR_IOB_T2,		qla27xx_fwdt_entry_t259  },
+	{ ENTRY_TYPE_RD_PCI,		qla27xx_fwdt_entry_t260  },
+	{ ENTRY_TYPE_WR_PCI,		qla27xx_fwdt_entry_t261  },
+	{ ENTRY_TYPE_RD_RAM,		qla27xx_fwdt_entry_t262  },
+	{ ENTRY_TYPE_GET_QUEUE,		qla27xx_fwdt_entry_t263  },
+	{ ENTRY_TYPE_GET_FCE,		qla27xx_fwdt_entry_t264  },
+	{ ENTRY_TYPE_PSE_RISC,		qla27xx_fwdt_entry_t265  },
+	{ ENTRY_TYPE_RST_RISC,		qla27xx_fwdt_entry_t266  },
+	{ ENTRY_TYPE_DIS_INTR,		qla27xx_fwdt_entry_t267  },
+	{ ENTRY_TYPE_GET_HBUF,		qla27xx_fwdt_entry_t268  },
+	{ ENTRY_TYPE_SCRATCH,		qla27xx_fwdt_entry_t269  },
+	{ ENTRY_TYPE_RDREMREG,		qla27xx_fwdt_entry_t270  },
+	{ ENTRY_TYPE_WRREMREG,		qla27xx_fwdt_entry_t271  },
+	{ ENTRY_TYPE_RDREMRAM,		qla27xx_fwdt_entry_t272  },
+	{ ENTRY_TYPE_PCICFG,		qla27xx_fwdt_entry_t273  },
+	{ ENTRY_TYPE_GET_SHADOW,	qla27xx_fwdt_entry_t274  },
+	{ ENTRY_TYPE_WRITE_BUF,		qla27xx_fwdt_entry_t275  },
+	{ ENTRY_TYPE_CONDITIONAL,	qla27xx_fwdt_entry_t276  },
+	{ ENTRY_TYPE_RDPEPREG,		qla27xx_fwdt_entry_t277  },
+	{ ENTRY_TYPE_WRPEPREG,		qla27xx_fwdt_entry_t278  },
+	{ -1,				qla27xx_fwdt_entry_other }
 };
 
-static inline int (*qla27xx_find_entry(uint type))
-	(struct scsi_qla_host *, struct qla27xx_fwdt_entry *, void *, ulong *)
+static inline
+typeof(qla27xx_fwdt_entry_call->call)(qla27xx_find_entry(uint type))
 {
-	struct qla27xx_fwdt_entry_call *list = ql27xx_fwdt_entry_call_list;
+	typeof(*qla27xx_fwdt_entry_call) *list = qla27xx_fwdt_entry_call;
 
 	while (list->type < type)
 		list++;
@@ -834,14 +889,6 @@ static inline int (*qla27xx_find_entry(uint type))
 	return qla27xx_fwdt_entry_other;
 }
 
-static inline void *
-qla27xx_next_entry(void *p)
-{
-	struct qla27xx_fwdt_entry *ent = p;
-
-	return p + ent->hdr.entry_size;
-}
-
 static void
 qla27xx_walk_template(struct scsi_qla_host *vha,
 	struct qla27xx_fwdt_template *tmp, void *buf, ulong *len)
@@ -852,18 +899,16 @@ qla27xx_walk_template(struct scsi_qla_host *vha,
 	ql_dbg(ql_dbg_misc, vha, 0xd01a,
 	    "%s: entry count %lx\n", __func__, count);
 	while (count--) {
-		if (buf && *len >= vha->hw->fw_dump_len)
+		ent = qla27xx_find_entry(ent->hdr.type)(vha, ent, buf, len);
+		if (!ent)
 			break;
-		if (qla27xx_find_entry(ent->hdr.entry_type)(vha, ent, buf, len))
-			break;
-		ent = qla27xx_next_entry(ent);
 	}
 
 	if (count)
 		ql_dbg(ql_dbg_misc, vha, 0xd018,
 		    "%s: entry residual count (%lx)\n", __func__, count);
 
-	if (ent->hdr.entry_type != ENTRY_TYPE_TMP_END)
+	if (ent)
 		ql_dbg(ql_dbg_misc, vha, 0xd019,
 		    "%s: missing end entry (%lx)\n", __func__, count);
 
diff --git a/drivers/scsi/qla2xxx/qla_tmpl.h b/drivers/scsi/qla2xxx/qla_tmpl.h
index 141c1c5e73f4..5c2c2a8a19c4 100644
--- a/drivers/scsi/qla2xxx/qla_tmpl.h
+++ b/drivers/scsi/qla2xxx/qla_tmpl.h
@@ -54,6 +54,9 @@ struct __packed qla27xx_fwdt_template {
 #define ENTRY_TYPE_PCICFG		273
 #define ENTRY_TYPE_GET_SHADOW		274
 #define ENTRY_TYPE_WRITE_BUF		275
+#define ENTRY_TYPE_CONDITIONAL		276
+#define ENTRY_TYPE_RDPEPREG		277
+#define ENTRY_TYPE_WRPEPREG		278
 
 #define CAPTURE_FLAG_PHYS_ONLY		BIT_0
 #define CAPTURE_FLAG_PHYS_VIRT		BIT_1
@@ -62,8 +65,8 @@ struct __packed qla27xx_fwdt_template {
 
 struct __packed qla27xx_fwdt_entry {
 	struct __packed {
-		uint32_t entry_type;
-		uint32_t entry_size;
+		uint32_t type;
+		uint32_t size;
 		uint32_t reserved_1;
 
 		uint8_t  capture_flags;
@@ -199,6 +202,24 @@ struct __packed qla27xx_fwdt_entry {
 			uint32_t length;
 			uint8_t  buffer[];
 		} t275;
+
+		struct __packed {
+			uint32_t cond1;
+			uint32_t cond2;
+		} t276;
+
+		struct __packed {
+			uint32_t cmd_addr;
+			uint32_t wr_cmd_data;
+			uint32_t data_addr;
+		} t277;
+
+		struct __packed {
+			uint32_t cmd_addr;
+			uint32_t wr_cmd_data;
+			uint32_t data_addr;
+			uint32_t wr_data;
+		} t278;
 	};
 };
 
@@ -206,6 +227,7 @@ struct __packed qla27xx_fwdt_entry {
 #define T262_RAM_AREA_EXTERNAL_RAM	2
 #define T262_RAM_AREA_SHARED_RAM	3
 #define T262_RAM_AREA_DDR_RAM		4
+#define T262_RAM_AREA_MISC		5
 
 #define T263_QUEUE_TYPE_REQ		1
 #define T263_QUEUE_TYPE_RSP		2
diff --git a/drivers/scsi/qla2xxx/qla_version.h b/drivers/scsi/qla2xxx/qla_version.h
index ca7945cb959b..0690dac24081 100644
--- a/drivers/scsi/qla2xxx/qla_version.h
+++ b/drivers/scsi/qla2xxx/qla_version.h
@@ -7,7 +7,7 @@
 /*
  * Driver version
  */
-#define QLA2XXX_VERSION      "10.00.00.12-k"
+#define QLA2XXX_VERSION      "10.00.00.14-k"
 
 #define QLA_DRIVER_MAJOR_VER	10
 #define QLA_DRIVER_MINOR_VER	0
diff --git a/drivers/scsi/qla2xxx/tcm_qla2xxx.c b/drivers/scsi/qla2xxx/tcm_qla2xxx.c
index 283e6b80abb5..8a3075d17c63 100644
--- a/drivers/scsi/qla2xxx/tcm_qla2xxx.c
+++ b/drivers/scsi/qla2xxx/tcm_qla2xxx.c
@@ -420,26 +420,6 @@ static int tcm_qla2xxx_write_pending(struct se_cmd *se_cmd)
 	return qlt_rdy_to_xfer(cmd);
 }
 
-static int tcm_qla2xxx_write_pending_status(struct se_cmd *se_cmd)
-{
-	unsigned long flags;
-	/*
-	 * Check for WRITE_PENDING status to determine if we need to wait for
-	 * CTIO aborts to be posted via hardware in tcm_qla2xxx_handle_data().
-	 */
-	spin_lock_irqsave(&se_cmd->t_state_lock, flags);
-	if (se_cmd->t_state == TRANSPORT_WRITE_PENDING ||
-	    se_cmd->t_state == TRANSPORT_COMPLETE_QF_WP) {
-		spin_unlock_irqrestore(&se_cmd->t_state_lock, flags);
-		wait_for_completion_timeout(&se_cmd->t_transport_stop_comp,
-						50);
-		return 0;
-	}
-	spin_unlock_irqrestore(&se_cmd->t_state_lock, flags);
-
-	return 0;
-}
-
 static void tcm_qla2xxx_set_default_node_attrs(struct se_node_acl *nacl)
 {
 	return;
@@ -537,15 +517,6 @@ static void tcm_qla2xxx_handle_data_work(struct work_struct *work)
 
 	cmd->qpair->tgt_counters.qla_core_ret_ctio++;
 	if (!cmd->write_data_transferred) {
-		/*
-		 * Check if se_cmd has already been aborted via LUN_RESET, and
-		 * waiting upon completion in tcm_qla2xxx_write_pending_status()
-		 */
-		if (cmd->se_cmd.transport_state & CMD_T_ABORTED) {
-			complete(&cmd->se_cmd.t_transport_stop_comp);
-			return;
-		}
-
 		switch (cmd->dif_err_code) {
 		case DIF_ERR_GRD:
 			cmd->se_cmd.pi_err =
@@ -1902,7 +1873,6 @@ static const struct target_core_fabric_ops tcm_qla2xxx_ops = {
 	.sess_get_index			= tcm_qla2xxx_sess_get_index,
 	.sess_get_initiator_sid		= NULL,
 	.write_pending			= tcm_qla2xxx_write_pending,
-	.write_pending_status		= tcm_qla2xxx_write_pending_status,
 	.set_default_node_attributes	= tcm_qla2xxx_set_default_node_attrs,
 	.get_cmd_state			= tcm_qla2xxx_get_cmd_state,
 	.queue_data_in			= tcm_qla2xxx_queue_data_in,
@@ -1943,7 +1913,6 @@ static const struct target_core_fabric_ops tcm_qla2xxx_npiv_ops = {
 	.sess_get_index			= tcm_qla2xxx_sess_get_index,
 	.sess_get_initiator_sid		= NULL,
 	.write_pending			= tcm_qla2xxx_write_pending,
-	.write_pending_status		= tcm_qla2xxx_write_pending_status,
 	.set_default_node_attributes	= tcm_qla2xxx_set_default_node_attrs,
 	.get_cmd_state			= tcm_qla2xxx_get_cmd_state,
 	.queue_data_in			= tcm_qla2xxx_queue_data_in,