Merge branch 'for-next' into for-linus
[safe/jmp/linux-2.6] / drivers / scsi / lpfc / lpfc_scsi.c
index 5f697ac..483fb74 100644 (file)
@@ -1,7 +1,7 @@
 /*******************************************************************
  * This file is part of the Emulex Linux Device Driver for         *
  * Fibre Channel Host Bus Adapters.                                *
- * Copyright (C) 2004-2008 Emulex.  All rights reserved.           *
+ * Copyright (C) 2004-2009 Emulex.  All rights reserved.           *
  * EMULEX and SLI are trademarks of Emulex.                        *
  * www.emulex.com                                                  *
  * Portions Copyright (C) 2004-2005 Christoph Hellwig              *
  * more details, a copy of which can be found in the file COPYING  *
  * included with this package.                                     *
  *******************************************************************/
-
 #include <linux/pci.h>
 #include <linux/interrupt.h>
 #include <linux/delay.h>
+#include <asm/unaligned.h>
 
 #include <scsi/scsi.h>
 #include <scsi/scsi_device.h>
+#include <scsi/scsi_eh.h>
 #include <scsi/scsi_host.h>
 #include <scsi/scsi_tcq.h>
 #include <scsi/scsi_transport_fc.h>
 
 #include "lpfc_version.h"
+#include "lpfc_hw4.h"
 #include "lpfc_hw.h"
 #include "lpfc_sli.h"
+#include "lpfc_sli4.h"
 #include "lpfc_nl.h"
 #include "lpfc_disc.h"
 #include "lpfc_scsi.h"
 #define LPFC_RESET_WAIT  2
 #define LPFC_ABORT_WAIT  2
 
+int _dump_buf_done;
+
+static char *dif_op_str[] = {
+       "SCSI_PROT_NORMAL",
+       "SCSI_PROT_READ_INSERT",
+       "SCSI_PROT_WRITE_STRIP",
+       "SCSI_PROT_READ_STRIP",
+       "SCSI_PROT_WRITE_INSERT",
+       "SCSI_PROT_READ_PASS",
+       "SCSI_PROT_WRITE_PASS",
+};
+static void
+lpfc_release_scsi_buf_s4(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb);
+static void
+lpfc_release_scsi_buf_s3(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb);
+
+static void
+lpfc_debug_save_data(struct lpfc_hba *phba, struct scsi_cmnd *cmnd)
+{
+       void *src, *dst;
+       struct scatterlist *sgde = scsi_sglist(cmnd);
+
+       if (!_dump_buf_data) {
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                       "9050 BLKGRD: ERROR %s _dump_buf_data is NULL\n",
+                               __func__);
+               return;
+       }
+
+
+       if (!sgde) {
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                       "9051 BLKGRD: ERROR: data scatterlist is null\n");
+               return;
+       }
+
+       dst = (void *) _dump_buf_data;
+       while (sgde) {
+               src = sg_virt(sgde);
+               memcpy(dst, src, sgde->length);
+               dst += sgde->length;
+               sgde = sg_next(sgde);
+       }
+}
+
+static void
+lpfc_debug_save_dif(struct lpfc_hba *phba, struct scsi_cmnd *cmnd)
+{
+       void *src, *dst;
+       struct scatterlist *sgde = scsi_prot_sglist(cmnd);
+
+       if (!_dump_buf_dif) {
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                       "9052 BLKGRD: ERROR %s _dump_buf_data is NULL\n",
+                               __func__);
+               return;
+       }
+
+       if (!sgde) {
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                       "9053 BLKGRD: ERROR: prot scatterlist is null\n");
+               return;
+       }
+
+       dst = _dump_buf_dif;
+       while (sgde) {
+               src = sg_virt(sgde);
+               memcpy(dst, src, sgde->length);
+               dst += sgde->length;
+               sgde = sg_next(sgde);
+       }
+}
+
+/**
+ * lpfc_sli4_set_rsp_sgl_last - Set the last bit in the response sge.
+ * @phba: Pointer to HBA object.
+ * @lpfc_cmd: lpfc scsi command object pointer.
+ *
+ * This function is called from the lpfc_prep_task_mgmt_cmd function to
+ * set the last bit in the response sge entry.
+ **/
+static void
+lpfc_sli4_set_rsp_sgl_last(struct lpfc_hba *phba,
+                               struct lpfc_scsi_buf *lpfc_cmd)
+{
+       struct sli4_sge *sgl = (struct sli4_sge *)lpfc_cmd->fcp_bpl;
+       if (sgl) {
+               sgl += 1;
+               sgl->word2 = le32_to_cpu(sgl->word2);
+               bf_set(lpfc_sli4_sge_last, sgl, 1);
+               sgl->word2 = cpu_to_le32(sgl->word2);
+       }
+}
+
 /**
- * lpfc_update_stats: Update statistical data for the command completion.
+ * lpfc_update_stats - Update statistical data for the command completion
  * @phba: Pointer to HBA object.
  * @lpfc_cmd: lpfc scsi command object pointer.
  *
@@ -97,8 +194,7 @@ lpfc_update_stats(struct lpfc_hba *phba, struct  lpfc_scsi_buf *lpfc_cmd)
 }
 
 /**
- * lpfc_send_sdev_queuedepth_change_event: Posts a queuedepth change
- *                   event.
+ * lpfc_send_sdev_queuedepth_change_event - Posts a queuedepth change event
  * @phba: Pointer to HBA context object.
  * @vport: Pointer to vport object.
  * @ndlp: Pointer to FC node associated with the target.
@@ -152,7 +248,37 @@ lpfc_send_sdev_queuedepth_change_event(struct lpfc_hba *phba,
 }
 
 /**
- * lpfc_rampdown_queue_depth: Post RAMP_DOWN_QUEUE event to worker thread.
+ * lpfc_change_queue_depth - Alter scsi device queue depth
+ * @sdev: Pointer the scsi device on which to change the queue depth.
+ * @qdepth: New queue depth to set the sdev to.
+ * @reason: The reason for the queue depth change.
+ *
+ * This function is called by the midlayer and the LLD to alter the queue
+ * depth for a scsi device. This function sets the queue depth to the new
+ * value and sends an event out to log the queue depth change.
+ **/
+int
+lpfc_change_queue_depth(struct scsi_device *sdev, int qdepth, int reason)
+{
+       struct lpfc_vport *vport = (struct lpfc_vport *) sdev->host->hostdata;
+       struct lpfc_hba   *phba = vport->phba;
+       struct lpfc_rport_data *rdata;
+       unsigned long new_queue_depth, old_queue_depth;
+
+       old_queue_depth = sdev->queue_depth;
+       scsi_adjust_queue_depth(sdev, scsi_get_tag_type(sdev), qdepth);
+       new_queue_depth = sdev->queue_depth;
+       rdata = sdev->hostdata;
+       if (rdata)
+               lpfc_send_sdev_queuedepth_change_event(phba, vport,
+                                                      rdata->pnode, sdev->lun,
+                                                      old_queue_depth,
+                                                      new_queue_depth);
+       return sdev->queue_depth;
+}
+
+/**
+ * lpfc_rampdown_queue_depth - Post RAMP_DOWN_QUEUE event to worker thread
  * @phba: The Hba for which this call is being executed.
  *
  * This routine is called when there is resource error in driver or firmware.
@@ -193,7 +319,7 @@ lpfc_rampdown_queue_depth(struct lpfc_hba *phba)
 }
 
 /**
- * lpfc_rampup_queue_depth: Post RAMP_UP_QUEUE event for worker thread.
+ * lpfc_rampup_queue_depth - Post RAMP_UP_QUEUE event for worker thread
  * @phba: The Hba for which this call is being executed.
  *
  * This routine post WORKER_RAMP_UP_QUEUE event for @phba vport. This routine
@@ -205,18 +331,20 @@ lpfc_rampdown_queue_depth(struct lpfc_hba *phba)
  **/
 static inline void
 lpfc_rampup_queue_depth(struct lpfc_vport  *vport,
-                       struct scsi_device *sdev)
+                       uint32_t queue_depth)
 {
        unsigned long flags;
        struct lpfc_hba *phba = vport->phba;
        uint32_t evt_posted;
        atomic_inc(&phba->num_cmd_success);
 
-       if (vport->cfg_lun_queue_depth <= sdev->queue_depth)
+       if (vport->cfg_lun_queue_depth <= queue_depth)
                return;
        spin_lock_irqsave(&phba->hbalock, flags);
-       if (((phba->last_ramp_up_time + QUEUE_RAMP_UP_INTERVAL) > jiffies) ||
-        ((phba->last_rsrc_error_time + QUEUE_RAMP_UP_INTERVAL ) > jiffies)) {
+       if (time_before(jiffies,
+                       phba->last_ramp_up_time + QUEUE_RAMP_UP_INTERVAL) ||
+           time_before(jiffies,
+                       phba->last_rsrc_error_time + QUEUE_RAMP_UP_INTERVAL)) {
                spin_unlock_irqrestore(&phba->hbalock, flags);
                return;
        }
@@ -235,7 +363,7 @@ lpfc_rampup_queue_depth(struct lpfc_vport  *vport,
 }
 
 /**
- * lpfc_ramp_down_queue_handler: WORKER_RAMP_DOWN_QUEUE event handler.
+ * lpfc_ramp_down_queue_handler - WORKER_RAMP_DOWN_QUEUE event handler
  * @phba: The Hba for which this call is being executed.
  *
  * This routine is called to  process WORKER_RAMP_DOWN_QUEUE event for worker
@@ -248,17 +376,16 @@ lpfc_ramp_down_queue_handler(struct lpfc_hba *phba)
        struct lpfc_vport **vports;
        struct Scsi_Host  *shost;
        struct scsi_device *sdev;
-       unsigned long new_queue_depth, old_queue_depth;
+       unsigned long new_queue_depth;
        unsigned long num_rsrc_err, num_cmd_success;
        int i;
-       struct lpfc_rport_data *rdata;
 
        num_rsrc_err = atomic_read(&phba->num_rsrc_err);
        num_cmd_success = atomic_read(&phba->num_cmd_success);
 
        vports = lpfc_create_vport_work_array(phba);
        if (vports != NULL)
-               for(i = 0; i <= phba->max_vpi && vports[i] != NULL; i++) {
+               for (i = 0; i <= phba->max_vports && vports[i] != NULL; i++) {
                        shost = lpfc_shost_from_vport(vports[i]);
                        shost_for_each_device(sdev, shost) {
                                new_queue_depth =
@@ -269,22 +396,8 @@ lpfc_ramp_down_queue_handler(struct lpfc_hba *phba)
                                else
                                        new_queue_depth = sdev->queue_depth -
                                                                new_queue_depth;
-                               old_queue_depth = sdev->queue_depth;
-                               if (sdev->ordered_tags)
-                                       scsi_adjust_queue_depth(sdev,
-                                                       MSG_ORDERED_TAG,
-                                                       new_queue_depth);
-                               else
-                                       scsi_adjust_queue_depth(sdev,
-                                                       MSG_SIMPLE_TAG,
-                                                       new_queue_depth);
-                               rdata = sdev->hostdata;
-                               if (rdata)
-                                       lpfc_send_sdev_queuedepth_change_event(
-                                               phba, vports[i],
-                                               rdata->pnode,
-                                               sdev->lun, old_queue_depth,
-                                               new_queue_depth);
+                               lpfc_change_queue_depth(sdev, new_queue_depth,
+                                                       SCSI_QDEPTH_DEFAULT);
                        }
                }
        lpfc_destroy_vport_work_array(phba, vports);
@@ -293,7 +406,7 @@ lpfc_ramp_down_queue_handler(struct lpfc_hba *phba)
 }
 
 /**
- * lpfc_ramp_up_queue_handler: WORKER_RAMP_UP_QUEUE event handler.
+ * lpfc_ramp_up_queue_handler - WORKER_RAMP_UP_QUEUE event handler
  * @phba: The Hba for which this call is being executed.
  *
  * This routine is called to  process WORKER_RAMP_UP_QUEUE event for worker
@@ -308,32 +421,18 @@ lpfc_ramp_up_queue_handler(struct lpfc_hba *phba)
        struct Scsi_Host  *shost;
        struct scsi_device *sdev;
        int i;
-       struct lpfc_rport_data *rdata;
 
        vports = lpfc_create_vport_work_array(phba);
        if (vports != NULL)
-               for(i = 0; i <= phba->max_vpi && vports[i] != NULL; i++) {
+               for (i = 0; i <= phba->max_vports && vports[i] != NULL; i++) {
                        shost = lpfc_shost_from_vport(vports[i]);
                        shost_for_each_device(sdev, shost) {
                                if (vports[i]->cfg_lun_queue_depth <=
                                    sdev->queue_depth)
                                        continue;
-                               if (sdev->ordered_tags)
-                                       scsi_adjust_queue_depth(sdev,
-                                                       MSG_ORDERED_TAG,
-                                                       sdev->queue_depth+1);
-                               else
-                                       scsi_adjust_queue_depth(sdev,
-                                                       MSG_SIMPLE_TAG,
-                                                       sdev->queue_depth+1);
-                               rdata = sdev->hostdata;
-                               if (rdata)
-                                       lpfc_send_sdev_queuedepth_change_event(
-                                               phba, vports[i],
-                                               rdata->pnode,
-                                               sdev->lun,
-                                               sdev->queue_depth - 1,
-                                               sdev->queue_depth);
+                               lpfc_change_queue_depth(sdev,
+                                                       sdev->queue_depth+1,
+                                                       SCSI_QDEPTH_RAMP_UP);
                        }
                }
        lpfc_destroy_vport_work_array(phba, vports);
@@ -342,7 +441,7 @@ lpfc_ramp_up_queue_handler(struct lpfc_hba *phba)
 }
 
 /**
- * lpfc_scsi_dev_block: set all scsi hosts to block state.
+ * lpfc_scsi_dev_block - set all scsi hosts to block state
  * @phba: Pointer to HBA context object.
  *
  * This function walks vport list and set each SCSI host to block state
@@ -360,7 +459,7 @@ lpfc_scsi_dev_block(struct lpfc_hba *phba)
 
        vports = lpfc_create_vport_work_array(phba);
        if (vports != NULL)
-               for (i = 0; i <= phba->max_vpi && vports[i] != NULL; i++) {
+               for (i = 0; i <= phba->max_vports && vports[i] != NULL; i++) {
                        shost = lpfc_shost_from_vport(vports[i]);
                        shost_for_each_device(sdev, shost) {
                                rport = starget_to_rport(scsi_target(sdev));
@@ -371,22 +470,23 @@ lpfc_scsi_dev_block(struct lpfc_hba *phba)
 }
 
 /**
- * lpfc_new_scsi_buf: Scsi buffer allocator.
+ * lpfc_new_scsi_buf_s3 - Scsi buffer allocator for HBA with SLI3 IF spec
  * @vport: The virtual port for which this call being executed.
+ * @num_to_allocate: The requested number of buffers to allocate.
  *
- * This routine allocates a scsi buffer, which contains all the necessary
- * information needed to initiate a SCSI I/O.  The non-DMAable buffer region
- * contains information to build the IOCB.  The DMAable region contains
- * memory for the FCP CMND, FCP RSP, and the initial BPL.  In addition to
- * allocating memory, the FCP CMND and FCP RSP BDEs are setup in the BPL
- * and the BPL BDE is setup in the IOCB.
+ * This routine allocates a scsi buffer for device with SLI-3 interface spec,
+ * the scsi buffer contains all the necessary information needed to initiate
+ * a SCSI I/O. The non-DMAable buffer region contains information to build
+ * the IOCB. The DMAable region contains memory for the FCP CMND, FCP RSP,
+ * and the initial BPL. In addition to allocating memory, the FCP CMND and
+ * FCP RSP BDEs are setup in the BPL and the BPL BDE is setup in the IOCB.
  *
  * Return codes:
- *   NULL - Error
- *   Pointer to lpfc_scsi_buf data structure - Success
+ *   int - number of scsi buffers that were allocated.
+ *   0 = failure, less than num_to_alloc is a partial failure.
  **/
-static struct lpfc_scsi_buf *
-lpfc_new_scsi_buf(struct lpfc_vport *vport)
+static int
+lpfc_new_scsi_buf_s3(struct lpfc_vport *vport, int num_to_alloc)
 {
        struct lpfc_hba *phba = vport->phba;
        struct lpfc_scsi_buf *psb;
@@ -396,106 +496,403 @@ lpfc_new_scsi_buf(struct lpfc_vport *vport)
        dma_addr_t pdma_phys_fcp_rsp;
        dma_addr_t pdma_phys_bpl;
        uint16_t iotag;
+       int bcnt;
 
-       psb = kzalloc(sizeof(struct lpfc_scsi_buf), GFP_KERNEL);
-       if (!psb)
-               return NULL;
+       for (bcnt = 0; bcnt < num_to_alloc; bcnt++) {
+               psb = kzalloc(sizeof(struct lpfc_scsi_buf), GFP_KERNEL);
+               if (!psb)
+                       break;
+
+               /*
+                * Get memory from the pci pool to map the virt space to pci
+                * bus space for an I/O.  The DMA buffer includes space for the
+                * struct fcp_cmnd, struct fcp_rsp and the number of bde's
+                * necessary to support the sg_tablesize.
+                */
+               psb->data = pci_pool_alloc(phba->lpfc_scsi_dma_buf_pool,
+                                       GFP_KERNEL, &psb->dma_handle);
+               if (!psb->data) {
+                       kfree(psb);
+                       break;
+               }
+
+               /* Initialize virtual ptrs to dma_buf region. */
+               memset(psb->data, 0, phba->cfg_sg_dma_buf_size);
+
+               /* Allocate iotag for psb->cur_iocbq. */
+               iotag = lpfc_sli_next_iotag(phba, &psb->cur_iocbq);
+               if (iotag == 0) {
+                       pci_pool_free(phba->lpfc_scsi_dma_buf_pool,
+                                       psb->data, psb->dma_handle);
+                       kfree(psb);
+                       break;
+               }
+               psb->cur_iocbq.iocb_flag |= LPFC_IO_FCP;
+
+               psb->fcp_cmnd = psb->data;
+               psb->fcp_rsp = psb->data + sizeof(struct fcp_cmnd);
+               psb->fcp_bpl = psb->data + sizeof(struct fcp_cmnd) +
+                       sizeof(struct fcp_rsp);
+
+               /* Initialize local short-hand pointers. */
+               bpl = psb->fcp_bpl;
+               pdma_phys_fcp_cmd = psb->dma_handle;
+               pdma_phys_fcp_rsp = psb->dma_handle + sizeof(struct fcp_cmnd);
+               pdma_phys_bpl = psb->dma_handle + sizeof(struct fcp_cmnd) +
+                       sizeof(struct fcp_rsp);
+
+               /*
+                * The first two bdes are the FCP_CMD and FCP_RSP. The balance
+                * are sg list bdes.  Initialize the first two and leave the
+                * rest for queuecommand.
+                */
+               bpl[0].addrHigh = le32_to_cpu(putPaddrHigh(pdma_phys_fcp_cmd));
+               bpl[0].addrLow = le32_to_cpu(putPaddrLow(pdma_phys_fcp_cmd));
+               bpl[0].tus.f.bdeSize = sizeof(struct fcp_cmnd);
+               bpl[0].tus.f.bdeFlags = BUFF_TYPE_BDE_64;
+               bpl[0].tus.w = le32_to_cpu(bpl[0].tus.w);
+
+               /* Setup the physical region for the FCP RSP */
+               bpl[1].addrHigh = le32_to_cpu(putPaddrHigh(pdma_phys_fcp_rsp));
+               bpl[1].addrLow = le32_to_cpu(putPaddrLow(pdma_phys_fcp_rsp));
+               bpl[1].tus.f.bdeSize = sizeof(struct fcp_rsp);
+               bpl[1].tus.f.bdeFlags = BUFF_TYPE_BDE_64;
+               bpl[1].tus.w = le32_to_cpu(bpl[1].tus.w);
+
+               /*
+                * Since the IOCB for the FCP I/O is built into this
+                * lpfc_scsi_buf, initialize it with all known data now.
+                */
+               iocb = &psb->cur_iocbq.iocb;
+               iocb->un.fcpi64.bdl.ulpIoTag32 = 0;
+               if ((phba->sli_rev == 3) &&
+                               !(phba->sli3_options & LPFC_SLI3_BG_ENABLED)) {
+                       /* fill in immediate fcp command BDE */
+                       iocb->un.fcpi64.bdl.bdeFlags = BUFF_TYPE_BDE_IMMED;
+                       iocb->un.fcpi64.bdl.bdeSize = sizeof(struct fcp_cmnd);
+                       iocb->un.fcpi64.bdl.addrLow = offsetof(IOCB_t,
+                                       unsli3.fcp_ext.icd);
+                       iocb->un.fcpi64.bdl.addrHigh = 0;
+                       iocb->ulpBdeCount = 0;
+                       iocb->ulpLe = 0;
+                       /* fill in responce BDE */
+                       iocb->unsli3.fcp_ext.rbde.tus.f.bdeFlags =
+                                                       BUFF_TYPE_BDE_64;
+                       iocb->unsli3.fcp_ext.rbde.tus.f.bdeSize =
+                               sizeof(struct fcp_rsp);
+                       iocb->unsli3.fcp_ext.rbde.addrLow =
+                               putPaddrLow(pdma_phys_fcp_rsp);
+                       iocb->unsli3.fcp_ext.rbde.addrHigh =
+                               putPaddrHigh(pdma_phys_fcp_rsp);
+               } else {
+                       iocb->un.fcpi64.bdl.bdeFlags = BUFF_TYPE_BLP_64;
+                       iocb->un.fcpi64.bdl.bdeSize =
+                                       (2 * sizeof(struct ulp_bde64));
+                       iocb->un.fcpi64.bdl.addrLow =
+                                       putPaddrLow(pdma_phys_bpl);
+                       iocb->un.fcpi64.bdl.addrHigh =
+                                       putPaddrHigh(pdma_phys_bpl);
+                       iocb->ulpBdeCount = 1;
+                       iocb->ulpLe = 1;
+               }
+               iocb->ulpClass = CLASS3;
+               psb->status = IOSTAT_SUCCESS;
+               /* Put it back into the SCSI buffer list */
+               lpfc_release_scsi_buf_s3(phba, psb);
 
-       /*
-        * Get memory from the pci pool to map the virt space to pci bus space
-        * for an I/O.  The DMA buffer includes space for the struct fcp_cmnd,
-        * struct fcp_rsp and the number of bde's necessary to support the
-        * sg_tablesize.
-        */
-       psb->data = pci_pool_alloc(phba->lpfc_scsi_dma_buf_pool, GFP_KERNEL,
-                                                       &psb->dma_handle);
-       if (!psb->data) {
-               kfree(psb);
-               return NULL;
        }
 
-       /* Initialize virtual ptrs to dma_buf region. */
-       memset(psb->data, 0, phba->cfg_sg_dma_buf_size);
+       return bcnt;
+}
 
-       /* Allocate iotag for psb->cur_iocbq. */
-       iotag = lpfc_sli_next_iotag(phba, &psb->cur_iocbq);
-       if (iotag == 0) {
-               pci_pool_free(phba->lpfc_scsi_dma_buf_pool,
-                             psb->data, psb->dma_handle);
-               kfree (psb);
-               return NULL;
+/**
+ * lpfc_sli4_fcp_xri_aborted - Fast-path process of fcp xri abort
+ * @phba: pointer to lpfc hba data structure.
+ * @axri: pointer to the fcp xri abort wcqe structure.
+ *
+ * This routine is invoked by the worker thread to process a SLI4 fast-path
+ * FCP aborted xri.
+ **/
+void
+lpfc_sli4_fcp_xri_aborted(struct lpfc_hba *phba,
+                         struct sli4_wcqe_xri_aborted *axri)
+{
+       uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
+       struct lpfc_scsi_buf *psb, *next_psb;
+       unsigned long iflag = 0;
+
+       spin_lock_irqsave(&phba->sli4_hba.abts_scsi_buf_list_lock, iflag);
+       list_for_each_entry_safe(psb, next_psb,
+               &phba->sli4_hba.lpfc_abts_scsi_buf_list, list) {
+               if (psb->cur_iocbq.sli4_xritag == xri) {
+                       list_del(&psb->list);
+                       psb->exch_busy = 0;
+                       psb->status = IOSTAT_SUCCESS;
+                       spin_unlock_irqrestore(
+                               &phba->sli4_hba.abts_scsi_buf_list_lock,
+                               iflag);
+                       lpfc_release_scsi_buf_s4(phba, psb);
+                       return;
+               }
        }
-       psb->cur_iocbq.iocb_flag |= LPFC_IO_FCP;
+       spin_unlock_irqrestore(&phba->sli4_hba.abts_scsi_buf_list_lock,
+                               iflag);
+}
 
-       psb->fcp_cmnd = psb->data;
-       psb->fcp_rsp = psb->data + sizeof(struct fcp_cmnd);
-       psb->fcp_bpl = psb->data + sizeof(struct fcp_cmnd) +
-                                                       sizeof(struct fcp_rsp);
+/**
+ * lpfc_sli4_repost_scsi_sgl_list - Repsot the Scsi buffers sgl pages as block
+ * @phba: pointer to lpfc hba data structure.
+ *
+ * This routine walks the list of scsi buffers that have been allocated and
+ * repost them to the HBA by using SGL block post. This is needed after a
+ * pci_function_reset/warm_start or start. The lpfc_hba_down_post_s4 routine
+ * is responsible for moving all scsi buffers on the lpfc_abts_scsi_sgl_list
+ * to the lpfc_scsi_buf_list. If the repost fails, reject all scsi buffers.
+ *
+ * Returns: 0 = success, non-zero failure.
+ **/
+int
+lpfc_sli4_repost_scsi_sgl_list(struct lpfc_hba *phba)
+{
+       struct lpfc_scsi_buf *psb;
+       int index, status, bcnt = 0, rcnt = 0, rc = 0;
+       LIST_HEAD(sblist);
+
+       for (index = 0; index < phba->sli4_hba.scsi_xri_cnt; index++) {
+               psb = phba->sli4_hba.lpfc_scsi_psb_array[index];
+               if (psb) {
+                       /* Remove from SCSI buffer list */
+                       list_del(&psb->list);
+                       /* Add it to a local SCSI buffer list */
+                       list_add_tail(&psb->list, &sblist);
+                       if (++rcnt == LPFC_NEMBED_MBOX_SGL_CNT) {
+                               bcnt = rcnt;
+                               rcnt = 0;
+                       }
+               } else
+                       /* A hole present in the XRI array, need to skip */
+                       bcnt = rcnt;
 
-       /* Initialize local short-hand pointers. */
-       bpl = psb->fcp_bpl;
-       pdma_phys_fcp_cmd = psb->dma_handle;
-       pdma_phys_fcp_rsp = psb->dma_handle + sizeof(struct fcp_cmnd);
-       pdma_phys_bpl = psb->dma_handle + sizeof(struct fcp_cmnd) +
-                       sizeof(struct fcp_rsp);
+               if (index == phba->sli4_hba.scsi_xri_cnt - 1)
+                       /* End of XRI array for SCSI buffer, complete */
+                       bcnt = rcnt;
 
-       /*
-        * The first two bdes are the FCP_CMD and FCP_RSP.  The balance are sg
-        * list bdes.  Initialize the first two and leave the rest for
-        * queuecommand.
-        */
-       bpl[0].addrHigh = le32_to_cpu(putPaddrHigh(pdma_phys_fcp_cmd));
-       bpl[0].addrLow = le32_to_cpu(putPaddrLow(pdma_phys_fcp_cmd));
-       bpl[0].tus.f.bdeSize = sizeof(struct fcp_cmnd);
-       bpl[0].tus.f.bdeFlags = BUFF_TYPE_BDE_64;
-       bpl[0].tus.w = le32_to_cpu(bpl[0].tus.w);
-
-       /* Setup the physical region for the FCP RSP */
-       bpl[1].addrHigh = le32_to_cpu(putPaddrHigh(pdma_phys_fcp_rsp));
-       bpl[1].addrLow = le32_to_cpu(putPaddrLow(pdma_phys_fcp_rsp));
-       bpl[1].tus.f.bdeSize = sizeof(struct fcp_rsp);
-       bpl[1].tus.f.bdeFlags = BUFF_TYPE_BDE_64;
-       bpl[1].tus.w = le32_to_cpu(bpl[1].tus.w);
+               /* Continue until collect up to a nembed page worth of sgls */
+               if (bcnt == 0)
+                       continue;
+               /* Now, post the SCSI buffer list sgls as a block */
+               status = lpfc_sli4_post_scsi_sgl_block(phba, &sblist, bcnt);
+               /* Reset SCSI buffer count for next round of posting */
+               bcnt = 0;
+               while (!list_empty(&sblist)) {
+                       list_remove_head(&sblist, psb, struct lpfc_scsi_buf,
+                                        list);
+                       if (status) {
+                               /* Put this back on the abort scsi list */
+                               psb->exch_busy = 1;
+                               rc++;
+                       } else {
+                               psb->exch_busy = 0;
+                               psb->status = IOSTAT_SUCCESS;
+                       }
+                       /* Put it back into the SCSI buffer list */
+                       lpfc_release_scsi_buf_s4(phba, psb);
+               }
+       }
+       return rc;
+}
 
-       /*
-        * Since the IOCB for the FCP I/O is built into this lpfc_scsi_buf,
-        * initialize it with all known data now.
-        */
-       iocb = &psb->cur_iocbq.iocb;
-       iocb->un.fcpi64.bdl.ulpIoTag32 = 0;
-       if (phba->sli_rev == 3) {
-               /* fill in immediate fcp command BDE */
-               iocb->un.fcpi64.bdl.bdeFlags = BUFF_TYPE_BDE_IMMED;
+/**
+ * lpfc_new_scsi_buf_s4 - Scsi buffer allocator for HBA with SLI4 IF spec
+ * @vport: The virtual port for which this call being executed.
+ * @num_to_allocate: The requested number of buffers to allocate.
+ *
+ * This routine allocates a scsi buffer for device with SLI-4 interface spec,
+ * the scsi buffer contains all the necessary information needed to initiate
+ * a SCSI I/O.
+ *
+ * Return codes:
+ *   int - number of scsi buffers that were allocated.
+ *   0 = failure, less than num_to_alloc is a partial failure.
+ **/
+static int
+lpfc_new_scsi_buf_s4(struct lpfc_vport *vport, int num_to_alloc)
+{
+       struct lpfc_hba *phba = vport->phba;
+       struct lpfc_scsi_buf *psb;
+       struct sli4_sge *sgl;
+       IOCB_t *iocb;
+       dma_addr_t pdma_phys_fcp_cmd;
+       dma_addr_t pdma_phys_fcp_rsp;
+       dma_addr_t pdma_phys_bpl, pdma_phys_bpl1;
+       uint16_t iotag, last_xritag = NO_XRI;
+       int status = 0, index;
+       int bcnt;
+       int non_sequential_xri = 0;
+       int rc = 0;
+       LIST_HEAD(sblist);
+
+       for (bcnt = 0; bcnt < num_to_alloc; bcnt++) {
+               psb = kzalloc(sizeof(struct lpfc_scsi_buf), GFP_KERNEL);
+               if (!psb)
+                       break;
+
+               /*
+                * Get memory from the pci pool to map the virt space to pci bus
+                * space for an I/O.  The DMA buffer includes space for the
+                * struct fcp_cmnd, struct fcp_rsp and the number of bde's
+                * necessary to support the sg_tablesize.
+                */
+               psb->data = pci_pool_alloc(phba->lpfc_scsi_dma_buf_pool,
+                                               GFP_KERNEL, &psb->dma_handle);
+               if (!psb->data) {
+                       kfree(psb);
+                       break;
+               }
+
+               /* Initialize virtual ptrs to dma_buf region. */
+               memset(psb->data, 0, phba->cfg_sg_dma_buf_size);
+
+               /* Allocate iotag for psb->cur_iocbq. */
+               iotag = lpfc_sli_next_iotag(phba, &psb->cur_iocbq);
+               if (iotag == 0) {
+                       kfree(psb);
+                       break;
+               }
+
+               psb->cur_iocbq.sli4_xritag = lpfc_sli4_next_xritag(phba);
+               if (psb->cur_iocbq.sli4_xritag == NO_XRI) {
+                       pci_pool_free(phba->lpfc_scsi_dma_buf_pool,
+                             psb->data, psb->dma_handle);
+                       kfree(psb);
+                       break;
+               }
+               if (last_xritag != NO_XRI
+                       && psb->cur_iocbq.sli4_xritag != (last_xritag+1)) {
+                       non_sequential_xri = 1;
+               } else
+                       list_add_tail(&psb->list, &sblist);
+               last_xritag = psb->cur_iocbq.sli4_xritag;
+
+               index = phba->sli4_hba.scsi_xri_cnt++;
+               psb->cur_iocbq.iocb_flag |= LPFC_IO_FCP;
+
+               psb->fcp_bpl = psb->data;
+               psb->fcp_cmnd = (psb->data + phba->cfg_sg_dma_buf_size)
+                       - (sizeof(struct fcp_cmnd) + sizeof(struct fcp_rsp));
+               psb->fcp_rsp = (struct fcp_rsp *)((uint8_t *)psb->fcp_cmnd +
+                                       sizeof(struct fcp_cmnd));
+
+               /* Initialize local short-hand pointers. */
+               sgl = (struct sli4_sge *)psb->fcp_bpl;
+               pdma_phys_bpl = psb->dma_handle;
+               pdma_phys_fcp_cmd =
+                       (psb->dma_handle + phba->cfg_sg_dma_buf_size)
+                        - (sizeof(struct fcp_cmnd) + sizeof(struct fcp_rsp));
+               pdma_phys_fcp_rsp = pdma_phys_fcp_cmd + sizeof(struct fcp_cmnd);
+
+               /*
+                * The first two bdes are the FCP_CMD and FCP_RSP.  The balance
+                * are sg list bdes.  Initialize the first two and leave the
+                * rest for queuecommand.
+                */
+               sgl->addr_hi = cpu_to_le32(putPaddrHigh(pdma_phys_fcp_cmd));
+               sgl->addr_lo = cpu_to_le32(putPaddrLow(pdma_phys_fcp_cmd));
+               bf_set(lpfc_sli4_sge_last, sgl, 0);
+               sgl->word2 = cpu_to_le32(sgl->word2);
+               sgl->sge_len = cpu_to_le32(sizeof(struct fcp_cmnd));
+               sgl++;
+
+               /* Setup the physical region for the FCP RSP */
+               sgl->addr_hi = cpu_to_le32(putPaddrHigh(pdma_phys_fcp_rsp));
+               sgl->addr_lo = cpu_to_le32(putPaddrLow(pdma_phys_fcp_rsp));
+               bf_set(lpfc_sli4_sge_last, sgl, 1);
+               sgl->word2 = cpu_to_le32(sgl->word2);
+               sgl->sge_len = cpu_to_le32(sizeof(struct fcp_rsp));
+
+               /*
+                * Since the IOCB for the FCP I/O is built into this
+                * lpfc_scsi_buf, initialize it with all known data now.
+                */
+               iocb = &psb->cur_iocbq.iocb;
+               iocb->un.fcpi64.bdl.ulpIoTag32 = 0;
+               iocb->un.fcpi64.bdl.bdeFlags = BUFF_TYPE_BDE_64;
+               /* setting the BLP size to 2 * sizeof BDE may not be correct.
+                * We are setting the bpl to point to out sgl. An sgl's
+                * entries are 16 bytes, a bpl entries are 12 bytes.
+                */
                iocb->un.fcpi64.bdl.bdeSize = sizeof(struct fcp_cmnd);
-               iocb->un.fcpi64.bdl.addrLow = offsetof(IOCB_t,
-                                                      unsli3.fcp_ext.icd);
-               iocb->un.fcpi64.bdl.addrHigh = 0;
-               iocb->ulpBdeCount = 0;
-               iocb->ulpLe = 0;
-               /* fill in responce BDE */
-               iocb->unsli3.fcp_ext.rbde.tus.f.bdeFlags = BUFF_TYPE_BDE_64;
-               iocb->unsli3.fcp_ext.rbde.tus.f.bdeSize =
-                                               sizeof(struct fcp_rsp);
-               iocb->unsli3.fcp_ext.rbde.addrLow =
-                                               putPaddrLow(pdma_phys_fcp_rsp);
-               iocb->unsli3.fcp_ext.rbde.addrHigh =
-                                               putPaddrHigh(pdma_phys_fcp_rsp);
-       } else {
-               iocb->un.fcpi64.bdl.bdeFlags = BUFF_TYPE_BLP_64;
-               iocb->un.fcpi64.bdl.bdeSize = (2 * sizeof(struct ulp_bde64));
-               iocb->un.fcpi64.bdl.addrLow = putPaddrLow(pdma_phys_bpl);
-               iocb->un.fcpi64.bdl.addrHigh = putPaddrHigh(pdma_phys_bpl);
+               iocb->un.fcpi64.bdl.addrLow = putPaddrLow(pdma_phys_fcp_cmd);
+               iocb->un.fcpi64.bdl.addrHigh = putPaddrHigh(pdma_phys_fcp_cmd);
                iocb->ulpBdeCount = 1;
                iocb->ulpLe = 1;
+               iocb->ulpClass = CLASS3;
+               if (phba->cfg_sg_dma_buf_size > SGL_PAGE_SIZE)
+                       pdma_phys_bpl1 = pdma_phys_bpl + SGL_PAGE_SIZE;
+               else
+                       pdma_phys_bpl1 = 0;
+               psb->dma_phys_bpl = pdma_phys_bpl;
+               phba->sli4_hba.lpfc_scsi_psb_array[index] = psb;
+               if (non_sequential_xri) {
+                       status = lpfc_sli4_post_sgl(phba, pdma_phys_bpl,
+                                               pdma_phys_bpl1,
+                                               psb->cur_iocbq.sli4_xritag);
+                       if (status) {
+                               /* Put this back on the abort scsi list */
+                               psb->exch_busy = 1;
+                               rc++;
+                       } else {
+                               psb->exch_busy = 0;
+                               psb->status = IOSTAT_SUCCESS;
+                       }
+                       /* Put it back into the SCSI buffer list */
+                       lpfc_release_scsi_buf_s4(phba, psb);
+                       break;
+               }
+       }
+       if (bcnt) {
+               status = lpfc_sli4_post_scsi_sgl_block(phba, &sblist, bcnt);
+               /* Reset SCSI buffer count for next round of posting */
+               while (!list_empty(&sblist)) {
+                       list_remove_head(&sblist, psb, struct lpfc_scsi_buf,
+                                list);
+                       if (status) {
+                               /* Put this back on the abort scsi list */
+                               psb->exch_busy = 1;
+                               rc++;
+                       } else {
+                               psb->exch_busy = 0;
+                               psb->status = IOSTAT_SUCCESS;
+                       }
+                       /* Put it back into the SCSI buffer list */
+                       lpfc_release_scsi_buf_s4(phba, psb);
+               }
        }
-       iocb->ulpClass = CLASS3;
 
-       return psb;
+       return bcnt + non_sequential_xri - rc;
 }
 
 /**
- * lpfc_get_scsi_buf: Get a scsi buffer from lpfc_scsi_buf_list list of Hba.
- * @phba: The Hba for which this call is being executed.
+ * lpfc_new_scsi_buf - Wrapper funciton for scsi buffer allocator
+ * @vport: The virtual port for which this call being executed.
+ * @num_to_allocate: The requested number of buffers to allocate.
+ *
+ * This routine wraps the actual SCSI buffer allocator function pointer from
+ * the lpfc_hba struct.
+ *
+ * Return codes:
+ *   int - number of scsi buffers that were allocated.
+ *   0 = failure, less than num_to_alloc is a partial failure.
+ **/
+static inline int
+lpfc_new_scsi_buf(struct lpfc_vport *vport, int num_to_alloc)
+{
+       return vport->phba->lpfc_new_scsi_buf(vport, num_to_alloc);
+}
+
+/**
+ * lpfc_get_scsi_buf - Get a scsi buffer from lpfc_scsi_buf_list of the HBA
+ * @phba: The HBA for which this call is being executed.
  *
  * This routine removes a scsi buffer from head of @phba lpfc_scsi_buf_list list
  * and returns to caller.
@@ -516,13 +913,14 @@ lpfc_get_scsi_buf(struct lpfc_hba * phba)
        if (lpfc_cmd) {
                lpfc_cmd->seg_cnt = 0;
                lpfc_cmd->nonsg_phys = 0;
+               lpfc_cmd->prot_seg_cnt = 0;
        }
        spin_unlock_irqrestore(&phba->scsi_buf_list_lock, iflag);
        return  lpfc_cmd;
 }
 
 /**
- * lpfc_release_scsi_buf: Return a scsi buffer back to hba lpfc_scsi_buf_list list.
+ * lpfc_release_scsi_buf - Return a scsi buffer back to hba scsi buf list
  * @phba: The Hba for which this call is being executed.
  * @psb: The scsi buffer which is being released.
  *
@@ -530,7 +928,7 @@ lpfc_get_scsi_buf(struct lpfc_hba * phba)
  * lpfc_scsi_buf_list list.
  **/
 static void
-lpfc_release_scsi_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
+lpfc_release_scsi_buf_s3(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
 {
        unsigned long iflag = 0;
 
@@ -541,21 +939,68 @@ lpfc_release_scsi_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
 }
 
 /**
- * lpfc_scsi_prep_dma_buf: Routine to do DMA mapping for scsi buffer.
+ * lpfc_release_scsi_buf_s4: Return a scsi buffer back to hba scsi buf list.
+ * @phba: The Hba for which this call is being executed.
+ * @psb: The scsi buffer which is being released.
+ *
+ * This routine releases @psb scsi buffer by adding it to tail of @phba
+ * lpfc_scsi_buf_list list. For SLI4 XRI's are tied to the scsi buffer
+ * and cannot be reused for at least RA_TOV amount of time if it was
+ * aborted.
+ **/
+static void
+lpfc_release_scsi_buf_s4(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
+{
+       unsigned long iflag = 0;
+
+       if (psb->exch_busy) {
+               spin_lock_irqsave(&phba->sli4_hba.abts_scsi_buf_list_lock,
+                                       iflag);
+               psb->pCmd = NULL;
+               list_add_tail(&psb->list,
+                       &phba->sli4_hba.lpfc_abts_scsi_buf_list);
+               spin_unlock_irqrestore(&phba->sli4_hba.abts_scsi_buf_list_lock,
+                                       iflag);
+       } else {
+
+               spin_lock_irqsave(&phba->scsi_buf_list_lock, iflag);
+               psb->pCmd = NULL;
+               list_add_tail(&psb->list, &phba->lpfc_scsi_buf_list);
+               spin_unlock_irqrestore(&phba->scsi_buf_list_lock, iflag);
+       }
+}
+
+/**
+ * lpfc_release_scsi_buf: Return a scsi buffer back to hba scsi buf list.
+ * @phba: The Hba for which this call is being executed.
+ * @psb: The scsi buffer which is being released.
+ *
+ * This routine releases @psb scsi buffer by adding it to tail of @phba
+ * lpfc_scsi_buf_list list.
+ **/
+static void
+lpfc_release_scsi_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
+{
+
+       phba->lpfc_release_scsi_buf(phba, psb);
+}
+
+/**
+ * lpfc_scsi_prep_dma_buf_s3 - DMA mapping for scsi buffer to SLI3 IF spec
  * @phba: The Hba for which this call is being executed.
  * @lpfc_cmd: The scsi buffer which is going to be mapped.
  *
  * This routine does the pci dma mapping for scatter-gather list of scsi cmnd
- * field of @lpfc_cmd. This routine scans through sg elements and format the
- * bdea. This routine also initializes all IOCB fields which are dependent on
- * scsi command request buffer.
+ * field of @lpfc_cmd for device with SLI-3 interface spec. This routine scans
+ * through sg elements and format the bdea. This routine also initializes all
+ * IOCB fields which are dependent on scsi command request buffer.
  *
  * Return codes:
  *   1 - Error
  *   0 - Success
  **/
 static int
-lpfc_scsi_prep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
+lpfc_scsi_prep_dma_buf_s3(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
 {
        struct scsi_cmnd *scsi_cmnd = lpfc_cmd->pCmd;
        struct scatterlist *sgel = NULL;
@@ -589,8 +1034,9 @@ lpfc_scsi_prep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
 
                lpfc_cmd->seg_cnt = nseg;
                if (lpfc_cmd->seg_cnt > phba->cfg_sg_seg_cnt) {
-                       printk(KERN_ERR "%s: Too many sg segments from "
-                              "dma_map_sg.  Config %d, seg_cnt %d",
+                       lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                               "9064 BLKGRD: %s: Too many sg segments from "
+                              "dma_map_sg.  Config %d, seg_cnt %d\n",
                               __func__, phba->cfg_sg_seg_cnt,
                               lpfc_cmd->seg_cnt);
                        scsi_dma_unmap(scsi_cmnd);
@@ -609,6 +1055,7 @@ lpfc_scsi_prep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
                scsi_for_each_sg(scsi_cmnd, sgel, nseg, num_bde) {
                        physaddr = sg_dma_address(sgel);
                        if (phba->sli_rev == 3 &&
+                           !(phba->sli3_options & LPFC_SLI3_BG_ENABLED) &&
                            nseg <= LPFC_EXT_DATA_BDE_COUNT) {
                                data_bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
                                data_bde->tus.f.bdeSize = sg_dma_len(sgel);
@@ -634,7 +1081,8 @@ lpfc_scsi_prep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
         * explicitly reinitialized and for SLI-3 the extended bde count is
         * explicitly reinitialized since all iocb memory resources are reused.
         */
-       if (phba->sli_rev == 3) {
+       if (phba->sli_rev == 3 &&
+           !(phba->sli3_options & LPFC_SLI3_BG_ENABLED)) {
                if (num_bde > LPFC_EXT_DATA_BDE_COUNT) {
                        /*
                         * The extended IOCB format can only fit 3 BDE or a BPL.
@@ -661,37 +1109,849 @@ lpfc_scsi_prep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
                        ((num_bde + 2) * sizeof(struct ulp_bde64));
        }
        fcp_cmnd->fcpDl = cpu_to_be32(scsi_bufflen(scsi_cmnd));
+
+       /*
+        * Due to difference in data length between DIF/non-DIF paths,
+        * we need to set word 4 of IOCB here
+        */
+       iocb_cmd->un.fcpi.fcpi_parm = scsi_bufflen(scsi_cmnd);
        return 0;
 }
 
+/*
+ * Given a scsi cmnd, determine the BlockGuard profile to be used
+ * with the cmd
+ */
+static int
+lpfc_sc_to_sli_prof(struct lpfc_hba *phba, struct scsi_cmnd *sc)
+{
+       uint8_t guard_type = scsi_host_get_guard(sc->device->host);
+       uint8_t ret_prof = LPFC_PROF_INVALID;
+
+       if (guard_type == SHOST_DIX_GUARD_IP) {
+               switch (scsi_get_prot_op(sc)) {
+               case SCSI_PROT_READ_INSERT:
+               case SCSI_PROT_WRITE_STRIP:
+                       ret_prof = LPFC_PROF_AST2;
+                       break;
+
+               case SCSI_PROT_READ_STRIP:
+               case SCSI_PROT_WRITE_INSERT:
+                       ret_prof = LPFC_PROF_A1;
+                       break;
+
+               case SCSI_PROT_READ_PASS:
+               case SCSI_PROT_WRITE_PASS:
+                       ret_prof = LPFC_PROF_AST1;
+                       break;
+
+               case SCSI_PROT_NORMAL:
+               default:
+                       lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                               "9063 BLKGRD:Bad op/guard:%d/%d combination\n",
+                                       scsi_get_prot_op(sc), guard_type);
+                       break;
+
+               }
+       } else if (guard_type == SHOST_DIX_GUARD_CRC) {
+               switch (scsi_get_prot_op(sc)) {
+               case SCSI_PROT_READ_STRIP:
+               case SCSI_PROT_WRITE_INSERT:
+                       ret_prof = LPFC_PROF_A1;
+                       break;
+
+               case SCSI_PROT_READ_PASS:
+               case SCSI_PROT_WRITE_PASS:
+                       ret_prof = LPFC_PROF_C1;
+                       break;
+
+               case SCSI_PROT_READ_INSERT:
+               case SCSI_PROT_WRITE_STRIP:
+               case SCSI_PROT_NORMAL:
+               default:
+                       lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                               "9075 BLKGRD: Bad op/guard:%d/%d combination\n",
+                                       scsi_get_prot_op(sc), guard_type);
+                       break;
+               }
+       } else {
+               /* unsupported format */
+               BUG();
+       }
+
+       return ret_prof;
+}
+
+struct scsi_dif_tuple {
+       __be16 guard_tag;       /* Checksum */
+       __be16 app_tag;         /* Opaque storage */
+       __be32 ref_tag;         /* Target LBA or indirect LBA */
+};
+
+static inline unsigned
+lpfc_cmd_blksize(struct scsi_cmnd *sc)
+{
+       return sc->device->sector_size;
+}
+
 /**
- * lpfc_send_scsi_error_event: Posts an event when there is SCSI error.
- * @phba: Pointer to hba context object.
- * @vport: Pointer to vport object.
- * @lpfc_cmd: Pointer to lpfc scsi command which reported the error.
- * @rsp_iocb: Pointer to response iocb object which reported error.
+ * lpfc_get_cmd_dif_parms - Extract DIF parameters from SCSI command
+ * @sc:             in: SCSI command
+ * @apptagmask:     out: app tag mask
+ * @apptagval:      out: app tag value
+ * @reftag:         out: ref tag (reference tag)
+ *
+ * Description:
+ *   Extract DIF parameters from the command if possible.  Otherwise,
+ *   use default parameters.
  *
- * This function posts an event when there is a SCSI command reporting
- * error from the scsi device.
  **/
-static void
-lpfc_send_scsi_error_event(struct lpfc_hba *phba, struct lpfc_vport *vport,
-               struct lpfc_scsi_buf *lpfc_cmd, struct lpfc_iocbq *rsp_iocb) {
-       struct scsi_cmnd *cmnd = lpfc_cmd->pCmd;
-       struct fcp_rsp *fcprsp = lpfc_cmd->fcp_rsp;
-       uint32_t resp_info = fcprsp->rspStatus2;
-       uint32_t scsi_status = fcprsp->rspStatus3;
-       uint32_t fcpi_parm = rsp_iocb->iocb.un.fcpi.fcpi_parm;
-       struct lpfc_fast_path_event *fast_path_evt = NULL;
-       struct lpfc_nodelist *pnode = lpfc_cmd->rdata->pnode;
-       unsigned long flags;
+static inline void
+lpfc_get_cmd_dif_parms(struct scsi_cmnd *sc, uint16_t *apptagmask,
+               uint16_t *apptagval, uint32_t *reftag)
+{
+       struct  scsi_dif_tuple *spt;
+       unsigned char op = scsi_get_prot_op(sc);
+       unsigned int protcnt = scsi_prot_sg_count(sc);
+       static int cnt;
 
-       /* If there is queuefull or busy condition send a scsi event */
-       if ((cmnd->result == SAM_STAT_TASK_SET_FULL) ||
-               (cmnd->result == SAM_STAT_BUSY)) {
-               fast_path_evt = lpfc_alloc_fast_evt(phba);
-               if (!fast_path_evt)
-                       return;
+       if (protcnt && (op == SCSI_PROT_WRITE_STRIP ||
+                               op == SCSI_PROT_WRITE_PASS)) {
+
+               cnt++;
+               spt = page_address(sg_page(scsi_prot_sglist(sc))) +
+                       scsi_prot_sglist(sc)[0].offset;
+               *apptagmask = 0;
+               *apptagval = 0;
+               *reftag = cpu_to_be32(spt->ref_tag);
+
+       } else {
+               /* SBC defines ref tag to be lower 32bits of LBA */
+               *reftag = (uint32_t) (0xffffffff & scsi_get_lba(sc));
+               *apptagmask = 0;
+               *apptagval = 0;
+       }
+}
+
+/*
+ * This function sets up buffer list for protection groups of
+ * type LPFC_PG_TYPE_NO_DIF
+ *
+ * This is usually used when the HBA is instructed to generate
+ * DIFs and insert them into data stream (or strip DIF from
+ * incoming data stream)
+ *
+ * The buffer list consists of just one protection group described
+ * below:
+ *                                +-------------------------+
+ *   start of prot group  -->     |          PDE_1          |
+ *                                +-------------------------+
+ *                                |         Data BDE        |
+ *                                +-------------------------+
+ *                                |more Data BDE's ... (opt)|
+ *                                +-------------------------+
+ *
+ * @sc: pointer to scsi command we're working on
+ * @bpl: pointer to buffer list for protection groups
+ * @datacnt: number of segments of data that have been dma mapped
+ *
+ * Note: Data s/g buffers have been dma mapped
+ */
+static int
+lpfc_bg_setup_bpl(struct lpfc_hba *phba, struct scsi_cmnd *sc,
+               struct ulp_bde64 *bpl, int datasegcnt)
+{
+       struct scatterlist *sgde = NULL; /* s/g data entry */
+       struct lpfc_pde *pde1 = NULL;
+       dma_addr_t physaddr;
+       int i = 0, num_bde = 0;
+       int datadir = sc->sc_data_direction;
+       int prof = LPFC_PROF_INVALID;
+       unsigned blksize;
+       uint32_t reftag;
+       uint16_t apptagmask, apptagval;
+
+       pde1 = (struct lpfc_pde *) bpl;
+       prof = lpfc_sc_to_sli_prof(phba, sc);
+
+       if (prof == LPFC_PROF_INVALID)
+               goto out;
+
+       /* extract some info from the scsi command for PDE1*/
+       blksize = lpfc_cmd_blksize(sc);
+       lpfc_get_cmd_dif_parms(sc, &apptagmask, &apptagval, &reftag);
+
+       /* setup PDE1 with what we have */
+       lpfc_pde_set_bg_parms(pde1, LPFC_PDE1_DESCRIPTOR, prof, blksize,
+                       BG_EC_STOP_ERR);
+       lpfc_pde_set_dif_parms(pde1, apptagmask, apptagval, reftag);
+
+       num_bde++;
+       bpl++;
+
+       /* assumption: caller has already run dma_map_sg on command data */
+       scsi_for_each_sg(sc, sgde, datasegcnt, i) {
+               physaddr = sg_dma_address(sgde);
+               bpl->addrLow = le32_to_cpu(putPaddrLow(physaddr));
+               bpl->addrHigh = le32_to_cpu(putPaddrHigh(physaddr));
+               bpl->tus.f.bdeSize = sg_dma_len(sgde);
+               if (datadir == DMA_TO_DEVICE)
+                       bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
+               else
+                       bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
+               bpl->tus.w = le32_to_cpu(bpl->tus.w);
+               bpl++;
+               num_bde++;
+       }
+
+out:
+       return num_bde;
+}
+
+/*
+ * This function sets up buffer list for protection groups of
+ * type LPFC_PG_TYPE_DIF_BUF
+ *
+ * This is usually used when DIFs are in their own buffers,
+ * separate from the data. The HBA can then by instructed
+ * to place the DIFs in the outgoing stream.  For read operations,
+ * The HBA could extract the DIFs and place it in DIF buffers.
+ *
+ * The buffer list for this type consists of one or more of the
+ * protection groups described below:
+ *                                    +-------------------------+
+ *   start of first prot group  -->   |          PDE_1          |
+ *                                    +-------------------------+
+ *                                    |      PDE_3 (Prot BDE)   |
+ *                                    +-------------------------+
+ *                                    |        Data BDE         |
+ *                                    +-------------------------+
+ *                                    |more Data BDE's ... (opt)|
+ *                                    +-------------------------+
+ *   start of new  prot group  -->    |          PDE_1          |
+ *                                    +-------------------------+
+ *                                    |          ...            |
+ *                                    +-------------------------+
+ *
+ * @sc: pointer to scsi command we're working on
+ * @bpl: pointer to buffer list for protection groups
+ * @datacnt: number of segments of data that have been dma mapped
+ * @protcnt: number of segment of protection data that have been dma mapped
+ *
+ * Note: It is assumed that both data and protection s/g buffers have been
+ *       mapped for DMA
+ */
+static int
+lpfc_bg_setup_bpl_prot(struct lpfc_hba *phba, struct scsi_cmnd *sc,
+               struct ulp_bde64 *bpl, int datacnt, int protcnt)
+{
+       struct scatterlist *sgde = NULL; /* s/g data entry */
+       struct scatterlist *sgpe = NULL; /* s/g prot entry */
+       struct lpfc_pde *pde1 = NULL;
+       struct ulp_bde64 *prot_bde = NULL;
+       dma_addr_t dataphysaddr, protphysaddr;
+       unsigned short curr_data = 0, curr_prot = 0;
+       unsigned int split_offset, protgroup_len;
+       unsigned int protgrp_blks, protgrp_bytes;
+       unsigned int remainder, subtotal;
+       int prof = LPFC_PROF_INVALID;
+       int datadir = sc->sc_data_direction;
+       unsigned char pgdone = 0, alldone = 0;
+       unsigned blksize;
+       uint32_t reftag;
+       uint16_t apptagmask, apptagval;
+       int num_bde = 0;
+
+       sgpe = scsi_prot_sglist(sc);
+       sgde = scsi_sglist(sc);
+
+       if (!sgpe || !sgde) {
+               lpfc_printf_log(phba, KERN_ERR, LOG_FCP,
+                               "9020 Invalid s/g entry: data=0x%p prot=0x%p\n",
+                               sgpe, sgde);
+               return 0;
+       }
+
+       prof = lpfc_sc_to_sli_prof(phba, sc);
+       if (prof == LPFC_PROF_INVALID)
+               goto out;
+
+       /* extract some info from the scsi command for PDE1*/
+       blksize = lpfc_cmd_blksize(sc);
+       lpfc_get_cmd_dif_parms(sc, &apptagmask, &apptagval, &reftag);
+
+       split_offset = 0;
+       do {
+               /* setup the first PDE_1 */
+               pde1 = (struct lpfc_pde *) bpl;
+
+               lpfc_pde_set_bg_parms(pde1, LPFC_PDE1_DESCRIPTOR, prof, blksize,
+                               BG_EC_STOP_ERR);
+               lpfc_pde_set_dif_parms(pde1, apptagmask, apptagval, reftag);
+
+               num_bde++;
+               bpl++;
+
+               /* setup the first BDE that points to protection buffer */
+               prot_bde = (struct ulp_bde64 *) bpl;
+               protphysaddr = sg_dma_address(sgpe);
+               prot_bde->addrLow = le32_to_cpu(putPaddrLow(protphysaddr));
+               prot_bde->addrHigh = le32_to_cpu(putPaddrHigh(protphysaddr));
+               protgroup_len = sg_dma_len(sgpe);
+
+
+               /* must be integer multiple of the DIF block length */
+               BUG_ON(protgroup_len % 8);
+
+               protgrp_blks = protgroup_len / 8;
+               protgrp_bytes = protgrp_blks * blksize;
+
+               prot_bde->tus.f.bdeSize = protgroup_len;
+               if (datadir == DMA_TO_DEVICE)
+                       prot_bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
+               else
+                       prot_bde->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
+               prot_bde->tus.w = le32_to_cpu(bpl->tus.w);
+
+               curr_prot++;
+               num_bde++;
+
+               /* setup BDE's for data blocks associated with DIF data */
+               pgdone = 0;
+               subtotal = 0; /* total bytes processed for current prot grp */
+               while (!pgdone) {
+                       if (!sgde) {
+                               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                                       "9065 BLKGRD:%s Invalid data segment\n",
+                                               __func__);
+                               return 0;
+                       }
+                       bpl++;
+                       dataphysaddr = sg_dma_address(sgde) + split_offset;
+                       bpl->addrLow = le32_to_cpu(putPaddrLow(dataphysaddr));
+                       bpl->addrHigh = le32_to_cpu(putPaddrHigh(dataphysaddr));
+
+                       remainder = sg_dma_len(sgde) - split_offset;
+
+                       if ((subtotal + remainder) <= protgrp_bytes) {
+                               /* we can use this whole buffer */
+                               bpl->tus.f.bdeSize = remainder;
+                               split_offset = 0;
+
+                               if ((subtotal + remainder) == protgrp_bytes)
+                                       pgdone = 1;
+                       } else {
+                               /* must split this buffer with next prot grp */
+                               bpl->tus.f.bdeSize = protgrp_bytes - subtotal;
+                               split_offset += bpl->tus.f.bdeSize;
+                       }
+
+                       subtotal += bpl->tus.f.bdeSize;
+
+                       if (datadir == DMA_TO_DEVICE)
+                               bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64;
+                       else
+                               bpl->tus.f.bdeFlags = BUFF_TYPE_BDE_64I;
+                       bpl->tus.w = le32_to_cpu(bpl->tus.w);
+
+                       num_bde++;
+                       curr_data++;
+
+                       if (split_offset)
+                               break;
+
+                       /* Move to the next s/g segment if possible */
+                       sgde = sg_next(sgde);
+               }
+
+               /* are we done ? */
+               if (curr_prot == protcnt) {
+                       alldone = 1;
+               } else if (curr_prot < protcnt) {
+                       /* advance to next prot buffer */
+                       sgpe = sg_next(sgpe);
+                       bpl++;
+
+                       /* update the reference tag */
+                       reftag += protgrp_blks;
+               } else {
+                       /* if we're here, we have a bug */
+                       lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                               "9054 BLKGRD: bug in %s\n", __func__);
+               }
+
+       } while (!alldone);
+
+out:
+
+
+       return num_bde;
+}
+/*
+ * Given a SCSI command that supports DIF, determine composition of protection
+ * groups involved in setting up buffer lists
+ *
+ * Returns:
+ *                           for DIF (for both read and write)
+ * */
+static int
+lpfc_prot_group_type(struct lpfc_hba *phba, struct scsi_cmnd *sc)
+{
+       int ret = LPFC_PG_TYPE_INVALID;
+       unsigned char op = scsi_get_prot_op(sc);
+
+       switch (op) {
+       case SCSI_PROT_READ_STRIP:
+       case SCSI_PROT_WRITE_INSERT:
+               ret = LPFC_PG_TYPE_NO_DIF;
+               break;
+       case SCSI_PROT_READ_INSERT:
+       case SCSI_PROT_WRITE_STRIP:
+       case SCSI_PROT_READ_PASS:
+       case SCSI_PROT_WRITE_PASS:
+               ret = LPFC_PG_TYPE_DIF_BUF;
+               break;
+       default:
+               lpfc_printf_log(phba, KERN_ERR, LOG_FCP,
+                               "9021 Unsupported protection op:%d\n", op);
+               break;
+       }
+
+       return ret;
+}
+
+/*
+ * This is the protection/DIF aware version of
+ * lpfc_scsi_prep_dma_buf(). It may be a good idea to combine the
+ * two functions eventually, but for now, it's here
+ */
+static int
+lpfc_bg_scsi_prep_dma_buf(struct lpfc_hba *phba,
+               struct lpfc_scsi_buf *lpfc_cmd)
+{
+       struct scsi_cmnd *scsi_cmnd = lpfc_cmd->pCmd;
+       struct fcp_cmnd *fcp_cmnd = lpfc_cmd->fcp_cmnd;
+       struct ulp_bde64 *bpl = lpfc_cmd->fcp_bpl;
+       IOCB_t *iocb_cmd = &lpfc_cmd->cur_iocbq.iocb;
+       uint32_t num_bde = 0;
+       int datasegcnt, protsegcnt, datadir = scsi_cmnd->sc_data_direction;
+       int prot_group_type = 0;
+       int diflen, fcpdl;
+       unsigned blksize;
+
+       /*
+        * Start the lpfc command prep by bumping the bpl beyond fcp_cmnd
+        *  fcp_rsp regions to the first data bde entry
+        */
+       bpl += 2;
+       if (scsi_sg_count(scsi_cmnd)) {
+               /*
+                * The driver stores the segment count returned from pci_map_sg
+                * because this a count of dma-mappings used to map the use_sg
+                * pages.  They are not guaranteed to be the same for those
+                * architectures that implement an IOMMU.
+                */
+               datasegcnt = dma_map_sg(&phba->pcidev->dev,
+                                       scsi_sglist(scsi_cmnd),
+                                       scsi_sg_count(scsi_cmnd), datadir);
+               if (unlikely(!datasegcnt))
+                       return 1;
+
+               lpfc_cmd->seg_cnt = datasegcnt;
+               if (lpfc_cmd->seg_cnt > phba->cfg_sg_seg_cnt) {
+                       lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                                       "9067 BLKGRD: %s: Too many sg segments"
+                                       " from dma_map_sg.  Config %d, seg_cnt"
+                                       " %d\n",
+                                       __func__, phba->cfg_sg_seg_cnt,
+                                       lpfc_cmd->seg_cnt);
+                       scsi_dma_unmap(scsi_cmnd);
+                       return 1;
+               }
+
+               prot_group_type = lpfc_prot_group_type(phba, scsi_cmnd);
+
+               switch (prot_group_type) {
+               case LPFC_PG_TYPE_NO_DIF:
+                       num_bde = lpfc_bg_setup_bpl(phba, scsi_cmnd, bpl,
+                                       datasegcnt);
+                       /* we should have 2 or more entries in buffer list */
+                       if (num_bde < 2)
+                               goto err;
+                       break;
+               case LPFC_PG_TYPE_DIF_BUF:{
+                       /*
+                        * This type indicates that protection buffers are
+                        * passed to the driver, so that needs to be prepared
+                        * for DMA
+                        */
+                       protsegcnt = dma_map_sg(&phba->pcidev->dev,
+                                       scsi_prot_sglist(scsi_cmnd),
+                                       scsi_prot_sg_count(scsi_cmnd), datadir);
+                       if (unlikely(!protsegcnt)) {
+                               scsi_dma_unmap(scsi_cmnd);
+                               return 1;
+                       }
+
+                       lpfc_cmd->prot_seg_cnt = protsegcnt;
+                       if (lpfc_cmd->prot_seg_cnt
+                           > phba->cfg_prot_sg_seg_cnt) {
+                               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                                       "9068 BLKGRD: %s: Too many prot sg "
+                                       "segments from dma_map_sg.  Config %d,"
+                                               "prot_seg_cnt %d\n", __func__,
+                                               phba->cfg_prot_sg_seg_cnt,
+                                               lpfc_cmd->prot_seg_cnt);
+                               dma_unmap_sg(&phba->pcidev->dev,
+                                            scsi_prot_sglist(scsi_cmnd),
+                                            scsi_prot_sg_count(scsi_cmnd),
+                                            datadir);
+                               scsi_dma_unmap(scsi_cmnd);
+                               return 1;
+                       }
+
+                       num_bde = lpfc_bg_setup_bpl_prot(phba, scsi_cmnd, bpl,
+                                       datasegcnt, protsegcnt);
+                       /* we should have 3 or more entries in buffer list */
+                       if (num_bde < 3)
+                               goto err;
+                       break;
+               }
+               case LPFC_PG_TYPE_INVALID:
+               default:
+                       lpfc_printf_log(phba, KERN_ERR, LOG_FCP,
+                                       "9022 Unexpected protection group %i\n",
+                                       prot_group_type);
+                       return 1;
+               }
+       }
+
+       /*
+        * Finish initializing those IOCB fields that are dependent on the
+        * scsi_cmnd request_buffer.  Note that the bdeSize is explicitly
+        * reinitialized since all iocb memory resources are used many times
+        * for transmit, receive, and continuation bpl's.
+        */
+       iocb_cmd->un.fcpi64.bdl.bdeSize = (2 * sizeof(struct ulp_bde64));
+       iocb_cmd->un.fcpi64.bdl.bdeSize += (num_bde * sizeof(struct ulp_bde64));
+       iocb_cmd->ulpBdeCount = 1;
+       iocb_cmd->ulpLe = 1;
+
+       fcpdl = scsi_bufflen(scsi_cmnd);
+
+       if (scsi_get_prot_type(scsi_cmnd) == SCSI_PROT_DIF_TYPE1) {
+               /*
+                * We are in DIF Type 1 mode
+                * Every data block has a 8 byte DIF (trailer)
+                * attached to it.  Must ajust FCP data length
+                */
+               blksize = lpfc_cmd_blksize(scsi_cmnd);
+               diflen = (fcpdl / blksize) * 8;
+               fcpdl += diflen;
+       }
+       fcp_cmnd->fcpDl = be32_to_cpu(fcpdl);
+
+       /*
+        * Due to difference in data length between DIF/non-DIF paths,
+        * we need to set word 4 of IOCB here
+        */
+       iocb_cmd->un.fcpi.fcpi_parm = fcpdl;
+
+       return 0;
+err:
+       lpfc_printf_log(phba, KERN_ERR, LOG_FCP,
+                       "9023 Could not setup all needed BDE's"
+                       "prot_group_type=%d, num_bde=%d\n",
+                       prot_group_type, num_bde);
+       return 1;
+}
+
+/*
+ * This function checks for BlockGuard errors detected by
+ * the HBA.  In case of errors, the ASC/ASCQ fields in the
+ * sense buffer will be set accordingly, paired with
+ * ILLEGAL_REQUEST to signal to the kernel that the HBA
+ * detected corruption.
+ *
+ * Returns:
+ *  0 - No error found
+ *  1 - BlockGuard error found
+ * -1 - Internal error (bad profile, ...etc)
+ */
+static int
+lpfc_parse_bg_err(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd,
+                       struct lpfc_iocbq *pIocbOut)
+{
+       struct scsi_cmnd *cmd = lpfc_cmd->pCmd;
+       struct sli3_bg_fields *bgf = &pIocbOut->iocb.unsli3.sli3_bg;
+       int ret = 0;
+       uint32_t bghm = bgf->bghm;
+       uint32_t bgstat = bgf->bgstat;
+       uint64_t failing_sector = 0;
+
+       lpfc_printf_log(phba, KERN_ERR, LOG_BG, "9069 BLKGRD: BG ERROR in cmd"
+                       " 0x%x lba 0x%llx blk cnt 0x%x "
+                       "bgstat=0x%x bghm=0x%x\n",
+                       cmd->cmnd[0], (unsigned long long)scsi_get_lba(cmd),
+                       blk_rq_sectors(cmd->request), bgstat, bghm);
+
+       spin_lock(&_dump_buf_lock);
+       if (!_dump_buf_done) {
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,  "9070 BLKGRD: Saving"
+                       " Data for %u blocks to debugfs\n",
+                               (cmd->cmnd[7] << 8 | cmd->cmnd[8]));
+               lpfc_debug_save_data(phba, cmd);
+
+               /* If we have a prot sgl, save the DIF buffer */
+               if (lpfc_prot_group_type(phba, cmd) ==
+                               LPFC_PG_TYPE_DIF_BUF) {
+                       lpfc_printf_log(phba, KERN_ERR, LOG_BG, "9071 BLKGRD: "
+                               "Saving DIF for %u blocks to debugfs\n",
+                               (cmd->cmnd[7] << 8 | cmd->cmnd[8]));
+                       lpfc_debug_save_dif(phba, cmd);
+               }
+
+               _dump_buf_done = 1;
+       }
+       spin_unlock(&_dump_buf_lock);
+
+       if (lpfc_bgs_get_invalid_prof(bgstat)) {
+               cmd->result = ScsiResult(DID_ERROR, 0);
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG, "9072 BLKGRD: Invalid"
+                       " BlockGuard profile. bgstat:0x%x\n",
+                       bgstat);
+               ret = (-1);
+               goto out;
+       }
+
+       if (lpfc_bgs_get_uninit_dif_block(bgstat)) {
+               cmd->result = ScsiResult(DID_ERROR, 0);
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG, "9073 BLKGRD: "
+                               "Invalid BlockGuard DIF Block. bgstat:0x%x\n",
+                               bgstat);
+               ret = (-1);
+               goto out;
+       }
+
+       if (lpfc_bgs_get_guard_err(bgstat)) {
+               ret = 1;
+
+               scsi_build_sense_buffer(1, cmd->sense_buffer, ILLEGAL_REQUEST,
+                               0x10, 0x1);
+               cmd->result = DRIVER_SENSE << 24
+                       | ScsiResult(DID_ABORT, SAM_STAT_CHECK_CONDITION);
+               phba->bg_guard_err_cnt++;
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                       "9055 BLKGRD: guard_tag error\n");
+       }
+
+       if (lpfc_bgs_get_reftag_err(bgstat)) {
+               ret = 1;
+
+               scsi_build_sense_buffer(1, cmd->sense_buffer, ILLEGAL_REQUEST,
+                               0x10, 0x3);
+               cmd->result = DRIVER_SENSE << 24
+                       | ScsiResult(DID_ABORT, SAM_STAT_CHECK_CONDITION);
+
+               phba->bg_reftag_err_cnt++;
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                       "9056 BLKGRD: ref_tag error\n");
+       }
+
+       if (lpfc_bgs_get_apptag_err(bgstat)) {
+               ret = 1;
+
+               scsi_build_sense_buffer(1, cmd->sense_buffer, ILLEGAL_REQUEST,
+                               0x10, 0x2);
+               cmd->result = DRIVER_SENSE << 24
+                       | ScsiResult(DID_ABORT, SAM_STAT_CHECK_CONDITION);
+
+               phba->bg_apptag_err_cnt++;
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                       "9061 BLKGRD: app_tag error\n");
+       }
+
+       if (lpfc_bgs_get_hi_water_mark_present(bgstat)) {
+               /*
+                * setup sense data descriptor 0 per SPC-4 as an information
+                * field, and put the failing LBA in it
+                */
+               cmd->sense_buffer[8] = 0;     /* Information */
+               cmd->sense_buffer[9] = 0xa;   /* Add. length */
+               bghm /= cmd->device->sector_size;
+
+               failing_sector = scsi_get_lba(cmd);
+               failing_sector += bghm;
+
+               put_unaligned_be64(failing_sector, &cmd->sense_buffer[10]);
+       }
+
+       if (!ret) {
+               /* No error was reported - problem in FW? */
+               cmd->result = ScsiResult(DID_ERROR, 0);
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                       "9057 BLKGRD: no errors reported!\n");
+       }
+
+out:
+       return ret;
+}
+
+/**
+ * lpfc_scsi_prep_dma_buf_s4 - DMA mapping for scsi buffer to SLI4 IF spec
+ * @phba: The Hba for which this call is being executed.
+ * @lpfc_cmd: The scsi buffer which is going to be mapped.
+ *
+ * This routine does the pci dma mapping for scatter-gather list of scsi cmnd
+ * field of @lpfc_cmd for device with SLI-4 interface spec.
+ *
+ * Return codes:
+ *     1 - Error
+ *     0 - Success
+ **/
+static int
+lpfc_scsi_prep_dma_buf_s4(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
+{
+       struct scsi_cmnd *scsi_cmnd = lpfc_cmd->pCmd;
+       struct scatterlist *sgel = NULL;
+       struct fcp_cmnd *fcp_cmnd = lpfc_cmd->fcp_cmnd;
+       struct sli4_sge *sgl = (struct sli4_sge *)lpfc_cmd->fcp_bpl;
+       IOCB_t *iocb_cmd = &lpfc_cmd->cur_iocbq.iocb;
+       dma_addr_t physaddr;
+       uint32_t num_bde = 0;
+       uint32_t dma_len;
+       uint32_t dma_offset = 0;
+       int nseg;
+
+       /*
+        * There are three possibilities here - use scatter-gather segment, use
+        * the single mapping, or neither.  Start the lpfc command prep by
+        * bumping the bpl beyond the fcp_cmnd and fcp_rsp regions to the first
+        * data bde entry.
+        */
+       if (scsi_sg_count(scsi_cmnd)) {
+               /*
+                * The driver stores the segment count returned from pci_map_sg
+                * because this a count of dma-mappings used to map the use_sg
+                * pages.  They are not guaranteed to be the same for those
+                * architectures that implement an IOMMU.
+                */
+
+               nseg = scsi_dma_map(scsi_cmnd);
+               if (unlikely(!nseg))
+                       return 1;
+               sgl += 1;
+               /* clear the last flag in the fcp_rsp map entry */
+               sgl->word2 = le32_to_cpu(sgl->word2);
+               bf_set(lpfc_sli4_sge_last, sgl, 0);
+               sgl->word2 = cpu_to_le32(sgl->word2);
+               sgl += 1;
+
+               lpfc_cmd->seg_cnt = nseg;
+               if (lpfc_cmd->seg_cnt > phba->cfg_sg_seg_cnt) {
+                       lpfc_printf_log(phba, KERN_ERR, LOG_BG, "9074 BLKGRD:"
+                               " %s: Too many sg segments from "
+                               "dma_map_sg.  Config %d, seg_cnt %d\n",
+                               __func__, phba->cfg_sg_seg_cnt,
+                              lpfc_cmd->seg_cnt);
+                       scsi_dma_unmap(scsi_cmnd);
+                       return 1;
+               }
+
+               /*
+                * The driver established a maximum scatter-gather segment count
+                * during probe that limits the number of sg elements in any
+                * single scsi command.  Just run through the seg_cnt and format
+                * the sge's.
+                * When using SLI-3 the driver will try to fit all the BDEs into
+                * the IOCB. If it can't then the BDEs get added to a BPL as it
+                * does for SLI-2 mode.
+                */
+               scsi_for_each_sg(scsi_cmnd, sgel, nseg, num_bde) {
+                       physaddr = sg_dma_address(sgel);
+                       dma_len = sg_dma_len(sgel);
+                       sgl->addr_lo = cpu_to_le32(putPaddrLow(physaddr));
+                       sgl->addr_hi = cpu_to_le32(putPaddrHigh(physaddr));
+                       if ((num_bde + 1) == nseg)
+                               bf_set(lpfc_sli4_sge_last, sgl, 1);
+                       else
+                               bf_set(lpfc_sli4_sge_last, sgl, 0);
+                       bf_set(lpfc_sli4_sge_offset, sgl, dma_offset);
+                       sgl->word2 = cpu_to_le32(sgl->word2);
+                       sgl->sge_len = cpu_to_le32(dma_len);
+                       dma_offset += dma_len;
+                       sgl++;
+               }
+       } else {
+               sgl += 1;
+               /* clear the last flag in the fcp_rsp map entry */
+               sgl->word2 = le32_to_cpu(sgl->word2);
+               bf_set(lpfc_sli4_sge_last, sgl, 1);
+               sgl->word2 = cpu_to_le32(sgl->word2);
+       }
+
+       /*
+        * Finish initializing those IOCB fields that are dependent on the
+        * scsi_cmnd request_buffer.  Note that for SLI-2 the bdeSize is
+        * explicitly reinitialized.
+        * all iocb memory resources are reused.
+        */
+       fcp_cmnd->fcpDl = cpu_to_be32(scsi_bufflen(scsi_cmnd));
+
+       /*
+        * Due to difference in data length between DIF/non-DIF paths,
+        * we need to set word 4 of IOCB here
+        */
+       iocb_cmd->un.fcpi.fcpi_parm = scsi_bufflen(scsi_cmnd);
+       return 0;
+}
+
+/**
+ * lpfc_scsi_prep_dma_buf - Wrapper function for DMA mapping of scsi buffer
+ * @phba: The Hba for which this call is being executed.
+ * @lpfc_cmd: The scsi buffer which is going to be mapped.
+ *
+ * This routine wraps the actual DMA mapping function pointer from the
+ * lpfc_hba struct.
+ *
+ * Return codes:
+ *     1 - Error
+ *     0 - Success
+ **/
+static inline int
+lpfc_scsi_prep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *lpfc_cmd)
+{
+       return phba->lpfc_scsi_prep_dma_buf(phba, lpfc_cmd);
+}
+
+/**
+ * lpfc_send_scsi_error_event - Posts an event when there is SCSI error
+ * @phba: Pointer to hba context object.
+ * @vport: Pointer to vport object.
+ * @lpfc_cmd: Pointer to lpfc scsi command which reported the error.
+ * @rsp_iocb: Pointer to response iocb object which reported error.
+ *
+ * This function posts an event when there is a SCSI command reporting
+ * error from the scsi device.
+ **/
+static void
+lpfc_send_scsi_error_event(struct lpfc_hba *phba, struct lpfc_vport *vport,
+               struct lpfc_scsi_buf *lpfc_cmd, struct lpfc_iocbq *rsp_iocb) {
+       struct scsi_cmnd *cmnd = lpfc_cmd->pCmd;
+       struct fcp_rsp *fcprsp = lpfc_cmd->fcp_rsp;
+       uint32_t resp_info = fcprsp->rspStatus2;
+       uint32_t scsi_status = fcprsp->rspStatus3;
+       uint32_t fcpi_parm = rsp_iocb->iocb.un.fcpi.fcpi_parm;
+       struct lpfc_fast_path_event *fast_path_evt = NULL;
+       struct lpfc_nodelist *pnode = lpfc_cmd->rdata->pnode;
+       unsigned long flags;
+
+       /* If there is queuefull or busy condition send a scsi event */
+       if ((cmnd->result == SAM_STAT_TASK_SET_FULL) ||
+               (cmnd->result == SAM_STAT_BUSY)) {
+               fast_path_evt = lpfc_alloc_fast_evt(phba);
+               if (!fast_path_evt)
+                       return;
                fast_path_evt->un.scsi_evt.event_type =
                        FC_REG_SCSI_EVENT;
                fast_path_evt->un.scsi_evt.subcategory =
@@ -757,15 +2017,15 @@ lpfc_send_scsi_error_event(struct lpfc_hba *phba, struct lpfc_vport *vport,
 }
 
 /**
- * lpfc_scsi_unprep_dma_buf: Routine to un-map DMA mapping of scatter gather.
- * @phba: The Hba for which this call is being executed.
+ * lpfc_scsi_unprep_dma_buf - Un-map DMA mapping of SG-list for dev
+ * @phba: The HBA for which this call is being executed.
  * @psb: The scsi buffer which is going to be un-mapped.
  *
  * This routine does DMA un-mapping of scatter gather list of scsi command
- * field of @lpfc_cmd.
+ * field of @lpfc_cmd for device with SLI-3 interface spec.
  **/
 static void
-lpfc_scsi_unprep_dma_buf(struct lpfc_hba * phba, struct lpfc_scsi_buf * psb)
+lpfc_scsi_unprep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
 {
        /*
         * There are only two special cases to consider.  (1) the scsi command
@@ -775,10 +2035,14 @@ lpfc_scsi_unprep_dma_buf(struct lpfc_hba * phba, struct lpfc_scsi_buf * psb)
         */
        if (psb->seg_cnt > 0)
                scsi_dma_unmap(psb->pCmd);
+       if (psb->prot_seg_cnt > 0)
+               dma_unmap_sg(&phba->pcidev->dev, scsi_prot_sglist(psb->pCmd),
+                               scsi_prot_sg_count(psb->pCmd),
+                               psb->pCmd->sc_data_direction);
 }
 
 /**
- * lpfc_handler_fcp_err: FCP response handler.
+ * lpfc_handler_fcp_err - FCP response handler
  * @vport: The virtual port for which this call is being executed.
  * @lpfc_cmd: Pointer to lpfc_scsi_buf data structure.
  * @rsp_iocb: The response IOCB which contains FCP error.
@@ -813,6 +2077,21 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
                goto out;
        }
 
+       if (resp_info & RSP_LEN_VALID) {
+               rsplen = be32_to_cpu(fcprsp->rspRspLen);
+               if ((rsplen != 0 && rsplen != 4 && rsplen != 8) ||
+                   (fcprsp->rspInfo3 != RSP_NO_FAILURE)) {
+                       lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
+                                "2719 Invalid response length: "
+                                "tgt x%x lun x%x cmnd x%x rsplen x%x\n",
+                                cmnd->device->id,
+                                cmnd->device->lun, cmnd->cmnd[0],
+                                rsplen);
+                       host_status = DID_ERROR;
+                       goto out;
+               }
+       }
+
        if ((resp_info & SNS_LEN_VALID) && fcprsp->rspSnsLen) {
                uint32_t snslen = be32_to_cpu(fcprsp->rspSnsLen);
                if (snslen > SCSI_SENSE_BUFFERSIZE)
@@ -828,7 +2107,7 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
                logit = LOG_FCP;
 
        lpfc_printf_vlog(vport, KERN_WARNING, logit,
-                        "0730 FCP command x%x failed: x%x SNS x%x x%x "
+                        "9024 FCP command x%x failed: x%x SNS x%x x%x "
                         "Data: x%x x%x x%x x%x x%x\n",
                         cmnd->cmnd[0], scsi_status,
                         be32_to_cpu(*lp), be32_to_cpu(*(lp + 3)), resp_info,
@@ -837,21 +2116,12 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
                         be32_to_cpu(fcprsp->rspRspLen),
                         fcprsp->rspInfo3);
 
-       if (resp_info & RSP_LEN_VALID) {
-               rsplen = be32_to_cpu(fcprsp->rspRspLen);
-               if ((rsplen != 0 && rsplen != 4 && rsplen != 8) ||
-                   (fcprsp->rspInfo3 != RSP_NO_FAILURE)) {
-                       host_status = DID_ERROR;
-                       goto out;
-               }
-       }
-
        scsi_set_resid(cmnd, 0);
        if (resp_info & RESID_UNDER) {
                scsi_set_resid(cmnd, be32_to_cpu(fcprsp->rspResId));
 
                lpfc_printf_vlog(vport, KERN_INFO, LOG_FCP,
-                                "0716 FCP Read Underrun, expected %d, "
+                                "9025 FCP Read Underrun, expected %d, "
                                 "residual %d Data: x%x x%x x%x\n",
                                 be32_to_cpu(fcpcmd->fcpDl),
                                 scsi_get_resid(cmnd), fcpi_parm, cmnd->cmnd[0],
@@ -867,7 +2137,7 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
                        (scsi_get_resid(cmnd) != fcpi_parm)) {
                        lpfc_printf_vlog(vport, KERN_WARNING,
                                         LOG_FCP | LOG_FCP_ERROR,
-                                        "0735 FCP Read Check Error "
+                                        "9026 FCP Read Check Error "
                                         "and Underrun Data: x%x x%x x%x x%x\n",
                                         be32_to_cpu(fcpcmd->fcpDl),
                                         scsi_get_resid(cmnd), fcpi_parm,
@@ -886,7 +2156,7 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
                    (scsi_bufflen(cmnd) - scsi_get_resid(cmnd)
                     < cmnd->underflow)) {
                        lpfc_printf_vlog(vport, KERN_INFO, LOG_FCP,
-                                        "0717 FCP command x%x residual "
+                                        "9027 FCP command x%x residual "
                                         "underrun converted to error "
                                         "Data: x%x x%x x%x\n",
                                         cmnd->cmnd[0], scsi_bufflen(cmnd),
@@ -895,8 +2165,8 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
                }
        } else if (resp_info & RESID_OVER) {
                lpfc_printf_vlog(vport, KERN_WARNING, LOG_FCP,
-                                "0720 FCP command x%x residual overrun error. "
-                                "Data: x%x x%x \n", cmnd->cmnd[0],
+                                "9028 FCP command x%x residual overrun error. "
+                                "Data: x%x x%x\n", cmnd->cmnd[0],
                                 scsi_bufflen(cmnd), scsi_get_resid(cmnd));
                host_status = DID_ERROR;
 
@@ -907,7 +2177,7 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
        } else if ((scsi_status == SAM_STAT_GOOD) && fcpi_parm &&
                        (cmnd->sc_data_direction == DMA_FROM_DEVICE)) {
                lpfc_printf_vlog(vport, KERN_WARNING, LOG_FCP | LOG_FCP_ERROR,
-                                "0734 FCP Read Check Error Data: "
+                                "9029 FCP Read Check Error Data: "
                                 "x%x x%x x%x x%x\n",
                                 be32_to_cpu(fcpcmd->fcpDl),
                                 be32_to_cpu(fcprsp->rspResId),
@@ -922,10 +2192,10 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
 }
 
 /**
- * lpfc_scsi_cmd_iocb_cmpl: Scsi cmnd IOCB completion routine.
+ * lpfc_scsi_cmd_iocb_cmpl - Scsi cmnd IOCB completion routine
  * @phba: The Hba for which this call is being executed.
  * @pIocbIn: The command IOCBQ for the scsi cmnd.
- * @pIocbOut: The response IOCBQ for the scsi cmnd .
+ * @pIocbOut: The response IOCBQ for the scsi cmnd.
  *
  * This routine assigns scsi command result by looking into response IOCB
  * status field appropriately. This routine handles QUEUE FULL condition as
@@ -942,13 +2212,18 @@ lpfc_scsi_cmd_iocb_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pIocbIn,
        struct lpfc_nodelist *pnode = rdata->pnode;
        struct scsi_cmnd *cmd = lpfc_cmd->pCmd;
        int result;
-       struct scsi_device *sdev, *tmp_sdev;
-       int depth = 0;
+       struct scsi_device *tmp_sdev;
+       int depth;
        unsigned long flags;
        struct lpfc_fast_path_event *fast_path_evt;
+       struct Scsi_Host *shost = cmd->device->host;
+       uint32_t queue_depth, scsi_id;
 
        lpfc_cmd->result = pIocbOut->iocb.un.ulpWord[4];
        lpfc_cmd->status = pIocbOut->iocb.ulpStatus;
+       /* pick up SLI4 exhange busy status from HBA */
+       lpfc_cmd->exch_busy = pIocbOut->iocb_flag & LPFC_EXCHANGE_BUSY;
+
        if (pnode && NLP_CHK_NODE_ACT(pnode))
                atomic_dec(&pnode->cmd_pending);
 
@@ -960,7 +2235,7 @@ lpfc_scsi_cmd_iocb_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pIocbIn,
                        lpfc_cmd->status = IOSTAT_DEFAULT;
 
                lpfc_printf_vlog(vport, KERN_WARNING, LOG_FCP,
-                                "0729 FCP cmd x%x failed <%d/%d> "
+                                "9030 FCP cmd x%x failed <%d/%d> "
                                 "status: x%x result: x%x Data: x%x x%x\n",
                                 cmd->cmnd[0],
                                 cmd->device ? cmd->device->id : 0xffff,
@@ -1008,7 +2283,28 @@ lpfc_scsi_cmd_iocb_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pIocbIn,
                            lpfc_cmd->result == IOERR_ABORT_REQUESTED) {
                                cmd->result = ScsiResult(DID_REQUEUE, 0);
                                break;
-                       } /* else: fall through */
+                       }
+
+                       if ((lpfc_cmd->result == IOERR_RX_DMA_FAILED ||
+                            lpfc_cmd->result == IOERR_TX_DMA_FAILED) &&
+                            pIocbOut->iocb.unsli3.sli3_bg.bgstat) {
+                               if (scsi_get_prot_op(cmd) != SCSI_PROT_NORMAL) {
+                                       /*
+                                        * This is a response for a BG enabled
+                                        * cmd. Parse BG error
+                                        */
+                                       lpfc_parse_bg_err(phba, lpfc_cmd,
+                                                       pIocbOut);
+                                       break;
+                               } else {
+                                       lpfc_printf_vlog(vport, KERN_WARNING,
+                                                       LOG_BG,
+                                                       "9031 non-zero BGSTAT "
+                                                       "on unprotected cmd\n");
+                               }
+                       }
+
+               /* else: fall through */
                default:
                        cmd->result = ScsiResult(DID_ERROR, 0);
                        break;
@@ -1035,11 +2331,10 @@ lpfc_scsi_cmd_iocb_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pIocbIn,
 
        lpfc_update_stats(phba, lpfc_cmd);
        result = cmd->result;
-       sdev = cmd->device;
        if (vport->cfg_max_scsicmpl_time &&
           time_after(jiffies, lpfc_cmd->start_time +
                msecs_to_jiffies(vport->cfg_max_scsicmpl_time))) {
-               spin_lock_irqsave(sdev->host->host_lock, flags);
+               spin_lock_irqsave(shost->host_lock, flags);
                if (pnode && NLP_CHK_NODE_ACT(pnode)) {
                        if (pnode->cmd_qdepth >
                                atomic_read(&pnode->cmd_pending) &&
@@ -1052,22 +2347,26 @@ lpfc_scsi_cmd_iocb_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pIocbIn,
 
                        pnode->last_change_time = jiffies;
                }
-               spin_unlock_irqrestore(sdev->host->host_lock, flags);
+               spin_unlock_irqrestore(shost->host_lock, flags);
        } else if (pnode && NLP_CHK_NODE_ACT(pnode)) {
                if ((pnode->cmd_qdepth < LPFC_MAX_TGT_QDEPTH) &&
                   time_after(jiffies, pnode->last_change_time +
                              msecs_to_jiffies(LPFC_TGTQ_INTERVAL))) {
-                       spin_lock_irqsave(sdev->host->host_lock, flags);
+                       spin_lock_irqsave(shost->host_lock, flags);
                        pnode->cmd_qdepth += pnode->cmd_qdepth *
                                LPFC_TGTQ_RAMPUP_PCENT / 100;
                        if (pnode->cmd_qdepth > LPFC_MAX_TGT_QDEPTH)
                                pnode->cmd_qdepth = LPFC_MAX_TGT_QDEPTH;
                        pnode->last_change_time = jiffies;
-                       spin_unlock_irqrestore(sdev->host->host_lock, flags);
+                       spin_unlock_irqrestore(shost->host_lock, flags);
                }
        }
 
        lpfc_scsi_unprep_dma_buf(phba, lpfc_cmd);
+
+       /* The sdev is not guaranteed to be valid post scsi_done upcall. */
+       queue_depth = cmd->device->queue_depth;
+       scsi_id = cmd->device->id;
        cmd->scsi_done(cmd);
 
        if (phba->cfg_poll & ENABLE_FCP_RING_POLLING) {
@@ -1075,45 +2374,17 @@ lpfc_scsi_cmd_iocb_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pIocbIn,
                 * If there is a thread waiting for command completion
                 * wake up the thread.
                 */
-               spin_lock_irqsave(sdev->host->host_lock, flags);
+               spin_lock_irqsave(shost->host_lock, flags);
                lpfc_cmd->pCmd = NULL;
                if (lpfc_cmd->waitq)
                        wake_up(lpfc_cmd->waitq);
-               spin_unlock_irqrestore(sdev->host->host_lock, flags);
+               spin_unlock_irqrestore(shost->host_lock, flags);
                lpfc_release_scsi_buf(phba, lpfc_cmd);
                return;
        }
 
-
        if (!result)
-               lpfc_rampup_queue_depth(vport, sdev);
-
-       if (!result && pnode && NLP_CHK_NODE_ACT(pnode) &&
-          ((jiffies - pnode->last_ramp_up_time) >
-               LPFC_Q_RAMP_UP_INTERVAL * HZ) &&
-          ((jiffies - pnode->last_q_full_time) >
-               LPFC_Q_RAMP_UP_INTERVAL * HZ) &&
-          (vport->cfg_lun_queue_depth > sdev->queue_depth)) {
-               shost_for_each_device(tmp_sdev, sdev->host) {
-                       if (vport->cfg_lun_queue_depth > tmp_sdev->queue_depth){
-                               if (tmp_sdev->id != sdev->id)
-                                       continue;
-                               if (tmp_sdev->ordered_tags)
-                                       scsi_adjust_queue_depth(tmp_sdev,
-                                               MSG_ORDERED_TAG,
-                                               tmp_sdev->queue_depth+1);
-                               else
-                                       scsi_adjust_queue_depth(tmp_sdev,
-                                               MSG_SIMPLE_TAG,
-                                               tmp_sdev->queue_depth+1);
-
-                               pnode->last_ramp_up_time = jiffies;
-                       }
-               }
-               lpfc_send_sdev_queuedepth_change_event(phba, vport, pnode,
-                       0xFFFFFFFF,
-                       sdev->queue_depth - 1, sdev->queue_depth);
-       }
+               lpfc_rampup_queue_depth(vport, queue_depth);
 
        /*
         * Check for queue full.  If the lun is reporting queue full, then
@@ -1121,30 +2392,20 @@ lpfc_scsi_cmd_iocb_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pIocbIn,
         */
        if (result == SAM_STAT_TASK_SET_FULL && pnode &&
            NLP_CHK_NODE_ACT(pnode)) {
-               pnode->last_q_full_time = jiffies;
-
-               shost_for_each_device(tmp_sdev, sdev->host) {
-                       if (tmp_sdev->id != sdev->id)
+               shost_for_each_device(tmp_sdev, shost) {
+                       if (tmp_sdev->id != scsi_id)
                                continue;
                        depth = scsi_track_queue_full(tmp_sdev,
-                                       tmp_sdev->queue_depth - 1);
-               }
-               /*
-                * The queue depth cannot be lowered any more.
-                * Modify the returned error code to store
-                * the final depth value set by
-                * scsi_track_queue_full.
-                */
-               if (depth == -1)
-                       depth = sdev->host->cmd_per_lun;
-
-               if (depth) {
+                                                     tmp_sdev->queue_depth-1);
+                       if (depth <= 0)
+                               continue;
                        lpfc_printf_vlog(vport, KERN_WARNING, LOG_FCP,
                                         "0711 detected queue full - lun queue "
                                         "depth adjusted to %d.\n", depth);
                        lpfc_send_sdev_queuedepth_change_event(phba, vport,
-                               pnode, 0xFFFFFFFF,
-                               depth+1, depth);
+                                                              pnode,
+                                                              tmp_sdev->lun,
+                                                              depth+1, depth);
                }
        }
 
@@ -1152,17 +2413,17 @@ lpfc_scsi_cmd_iocb_cmpl(struct lpfc_hba *phba, struct lpfc_iocbq *pIocbIn,
         * If there is a thread waiting for command completion
         * wake up the thread.
         */
-       spin_lock_irqsave(sdev->host->host_lock, flags);
+       spin_lock_irqsave(shost->host_lock, flags);
        lpfc_cmd->pCmd = NULL;
        if (lpfc_cmd->waitq)
                wake_up(lpfc_cmd->waitq);
-       spin_unlock_irqrestore(sdev->host->host_lock, flags);
+       spin_unlock_irqrestore(shost->host_lock, flags);
 
        lpfc_release_scsi_buf(phba, lpfc_cmd);
 }
 
 /**
- * lpfc_fcpcmd_to_iocb - copy the fcp_cmd data into the IOCB.
+ * lpfc_fcpcmd_to_iocb - copy the fcp_cmd data into the IOCB
  * @data: A pointer to the immediate command data portion of the IOCB.
  * @fcp_cmnd: The FCP Command that is provided by the SCSI layer.
  *
@@ -1180,13 +2441,13 @@ lpfc_fcpcmd_to_iocb(uint8_t *data, struct fcp_cmnd *fcp_cmnd)
 }
 
 /**
- * lpfc_scsi_prep_cmnd:  Routine to convert scsi cmnd to FCP information unit.
+ * lpfc_scsi_prep_cmnd - Wrapper func for convert scsi cmnd to FCP info unit
  * @vport: The virtual port for which this call is being executed.
  * @lpfc_cmd: The scsi command which needs to send.
  * @pnode: Pointer to lpfc_nodelist.
  *
  * This routine initializes fcp_cmnd and iocb data structure from scsi command
- * to transfer.
+ * to transfer for device with SLI3 interface spec.
  **/
 static void
 lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
@@ -1236,14 +2497,16 @@ lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
        if (scsi_sg_count(scsi_cmnd)) {
                if (datadir == DMA_TO_DEVICE) {
                        iocb_cmd->ulpCommand = CMD_FCP_IWRITE64_CR;
-                       iocb_cmd->un.fcpi.fcpi_parm = 0;
-                       iocb_cmd->ulpPU = 0;
+                       if (phba->sli_rev < LPFC_SLI_REV4) {
+                               iocb_cmd->un.fcpi.fcpi_parm = 0;
+                               iocb_cmd->ulpPU = 0;
+                       } else
+                               iocb_cmd->ulpPU = PARM_READ_CHECK;
                        fcp_cmnd->fcpCntl3 = WRITE_DATA;
                        phba->fc4OutputRequests++;
                } else {
                        iocb_cmd->ulpCommand = CMD_FCP_IREAD64_CR;
                        iocb_cmd->ulpPU = PARM_READ_CHECK;
-                       iocb_cmd->un.fcpi.fcpi_parm = scsi_bufflen(scsi_cmnd);
                        fcp_cmnd->fcpCntl3 = READ_DATA;
                        phba->fc4InputRequests++;
                }
@@ -1254,7 +2517,8 @@ lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
                fcp_cmnd->fcpCntl3 = 0;
                phba->fc4ControlRequests++;
        }
-       if (phba->sli_rev == 3)
+       if (phba->sli_rev == 3 &&
+           !(phba->sli3_options & LPFC_SLI3_BG_ENABLED))
                lpfc_fcpcmd_to_iocb(iocb_cmd->unsli3.fcp_ext.icd, fcp_cmnd);
        /*
         * Finish initializing those IOCB fields that are independent
@@ -1274,13 +2538,14 @@ lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
 }
 
 /**
- * lpfc_scsi_prep_task_mgmt_cmnd: Convert scsi TM cmnd to FCP information unit.
+ * lpfc_scsi_prep_task_mgmt_cmnd - Convert SLI3 scsi TM cmd to FCP info unit
  * @vport: The virtual port for which this call is being executed.
  * @lpfc_cmd: Pointer to lpfc_scsi_buf data structure.
  * @lun: Logical unit number.
  * @task_mgmt_cmd: SCSI task management command.
  *
- * This routine creates FCP information unit corresponding to @task_mgmt_cmd.
+ * This routine creates FCP information unit corresponding to @task_mgmt_cmd
+ * for device with SLI-3 interface spec.
  *
  * Return codes:
  *   0 - Error
@@ -1312,7 +2577,8 @@ lpfc_scsi_prep_task_mgmt_cmd(struct lpfc_vport *vport,
        memset(fcp_cmnd, 0, sizeof(struct fcp_cmnd));
        int_to_scsilun(lun, &fcp_cmnd->fcp_lun);
        fcp_cmnd->fcpCntl2 = task_mgmt_cmd;
-       if (vport->phba->sli_rev == 3)
+       if (vport->phba->sli_rev == 3 &&
+           !(vport->phba->sli3_options & LPFC_SLI3_BG_ENABLED))
                lpfc_fcpcmd_to_iocb(piocb->unsli3.fcp_ext.icd, fcp_cmnd);
        piocb->ulpCommand = CMD_FCP_ICMND64_CR;
        piocb->ulpContext = ndlp->nlp_rpi;
@@ -1328,15 +2594,58 @@ lpfc_scsi_prep_task_mgmt_cmd(struct lpfc_vport *vport,
                 * The driver will provide the timeout mechanism.
                 */
                piocb->ulpTimeout = 0;
-       } else {
+       } else
                piocb->ulpTimeout = lpfc_cmd->timeout;
-       }
+
+       if (vport->phba->sli_rev == LPFC_SLI_REV4)
+               lpfc_sli4_set_rsp_sgl_last(vport->phba, lpfc_cmd);
 
        return 1;
 }
 
 /**
- * lpc_taskmgmt_def_cmpl: IOCB completion routine for task management command.
+ * lpfc_scsi_api_table_setup - Set up scsi api fucntion jump table
+ * @phba: The hba struct for which this call is being executed.
+ * @dev_grp: The HBA PCI-Device group number.
+ *
+ * This routine sets up the SCSI interface API function jump table in @phba
+ * struct.
+ * Returns: 0 - success, -ENODEV - failure.
+ **/
+int
+lpfc_scsi_api_table_setup(struct lpfc_hba *phba, uint8_t dev_grp)
+{
+
+       phba->lpfc_scsi_unprep_dma_buf = lpfc_scsi_unprep_dma_buf;
+       phba->lpfc_scsi_prep_cmnd = lpfc_scsi_prep_cmnd;
+       phba->lpfc_get_scsi_buf = lpfc_get_scsi_buf;
+
+       switch (dev_grp) {
+       case LPFC_PCI_DEV_LP:
+               phba->lpfc_new_scsi_buf = lpfc_new_scsi_buf_s3;
+               phba->lpfc_scsi_prep_dma_buf = lpfc_scsi_prep_dma_buf_s3;
+               phba->lpfc_release_scsi_buf = lpfc_release_scsi_buf_s3;
+               break;
+       case LPFC_PCI_DEV_OC:
+               phba->lpfc_new_scsi_buf = lpfc_new_scsi_buf_s4;
+               phba->lpfc_scsi_prep_dma_buf = lpfc_scsi_prep_dma_buf_s4;
+               phba->lpfc_release_scsi_buf = lpfc_release_scsi_buf_s4;
+               break;
+       default:
+               lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
+                               "1418 Invalid HBA PCI-device group: 0x%x\n",
+                               dev_grp);
+               return -ENODEV;
+               break;
+       }
+       phba->lpfc_get_scsi_buf = lpfc_get_scsi_buf;
+       phba->lpfc_rampdown_queue_depth = lpfc_rampdown_queue_depth;
+       phba->lpfc_scsi_cmd_iocb_cmpl = lpfc_scsi_cmd_iocb_cmpl;
+       return 0;
+}
+
+/**
+ * lpfc_taskmgmt_def_cmpl - IOCB completion routine for task management command
  * @phba: The Hba for which this call is being executed.
  * @cmdiocbq: Pointer to lpfc_iocbq data structure.
  * @rspiocbq: Pointer to lpfc_iocbq data structure.
@@ -1357,74 +2666,7 @@ lpfc_tskmgmt_def_cmpl(struct lpfc_hba *phba,
 }
 
 /**
- * lpfc_scsi_tgt_reset: Target reset handler.
- * @lpfc_cmd: Pointer to lpfc_scsi_buf data structure
- * @vport: The virtual port for which this call is being executed.
- * @tgt_id: Target ID.
- * @lun: Lun number.
- * @rdata: Pointer to lpfc_rport_data.
- *
- * This routine issues a TARGET RESET iocb to reset a target with @tgt_id ID.
- *
- * Return Code:
- *   0x2003 - Error
- *   0x2002 - Success.
- **/
-static int
-lpfc_scsi_tgt_reset(struct lpfc_scsi_buf *lpfc_cmd, struct lpfc_vport *vport,
-                   unsigned  tgt_id, unsigned int lun,
-                   struct lpfc_rport_data *rdata)
-{
-       struct lpfc_hba   *phba = vport->phba;
-       struct lpfc_iocbq *iocbq;
-       struct lpfc_iocbq *iocbqrsp;
-       int ret;
-       int status;
-
-       if (!rdata->pnode || !NLP_CHK_NODE_ACT(rdata->pnode))
-               return FAILED;
-
-       lpfc_cmd->rdata = rdata;
-       status = lpfc_scsi_prep_task_mgmt_cmd(vport, lpfc_cmd, lun,
-                                          FCP_TARGET_RESET);
-       if (!status)
-               return FAILED;
-
-       iocbq = &lpfc_cmd->cur_iocbq;
-       iocbqrsp = lpfc_sli_get_iocbq(phba);
-
-       if (!iocbqrsp)
-               return FAILED;
-
-       /* Issue Target Reset to TGT <num> */
-       lpfc_printf_vlog(vport, KERN_INFO, LOG_FCP,
-                        "0702 Issue Target Reset to TGT %d Data: x%x x%x\n",
-                        tgt_id, rdata->pnode->nlp_rpi, rdata->pnode->nlp_flag);
-       status = lpfc_sli_issue_iocb_wait(phba,
-                                      &phba->sli.ring[phba->sli.fcp_ring],
-                                      iocbq, iocbqrsp, lpfc_cmd->timeout);
-       if (status != IOCB_SUCCESS) {
-               if (status == IOCB_TIMEDOUT) {
-                       iocbq->iocb_cmpl = lpfc_tskmgmt_def_cmpl;
-                       ret = TIMEOUT_ERROR;
-               } else
-                       ret = FAILED;
-               lpfc_cmd->status = IOSTAT_DRIVER_REJECT;
-       } else {
-               ret = SUCCESS;
-               lpfc_cmd->result = iocbqrsp->iocb.un.ulpWord[4];
-               lpfc_cmd->status = iocbqrsp->iocb.ulpStatus;
-               if (lpfc_cmd->status == IOSTAT_LOCAL_REJECT &&
-                       (lpfc_cmd->result & IOERR_DRVR_MASK))
-                               lpfc_cmd->status = IOSTAT_DRIVER_REJECT;
-       }
-
-       lpfc_sli_release_iocbq(phba, iocbqrsp);
-       return ret;
-}
-
-/**
- * lpfc_info: Info entry point of scsi_host_template data structure.
+ * lpfc_info - Info entry point of scsi_host_template data structure
  * @host: The scsi host for which this call is being executed.
  *
  * This routine provides module information about hba.
@@ -1457,12 +2699,19 @@ lpfc_info(struct Scsi_Host *host)
                                 " port %s",
                                 phba->Port);
                }
+               len = strlen(lpfcinfobuf);
+               if (phba->sli4_hba.link_state.logical_speed) {
+                       snprintf(lpfcinfobuf + len,
+                                384-len,
+                                " Logical Link Speed: %d Mbps",
+                                phba->sli4_hba.link_state.logical_speed * 10);
+               }
        }
        return lpfcinfobuf;
 }
 
 /**
- * lpfc_poll_rearm_time: Routine to modify fcp_poll timer of hba.
+ * lpfc_poll_rearm_time - Routine to modify fcp_poll timer of hba
  * @phba: The Hba for which this call is being executed.
  *
  * This routine modifies fcp_poll_timer  field of @phba by cfg_poll_tmo.
@@ -1479,7 +2728,7 @@ static __inline__ void lpfc_poll_rearm_timer(struct lpfc_hba * phba)
 }
 
 /**
- * lpfc_poll_start_timer: Routine to start fcp_poll_timer of HBA.
+ * lpfc_poll_start_timer - Routine to start fcp_poll_timer of HBA
  * @phba: The Hba for which this call is being executed.
  *
  * This routine starts the fcp_poll_timer of @phba.
@@ -1490,7 +2739,7 @@ void lpfc_poll_start_timer(struct lpfc_hba * phba)
 }
 
 /**
- * lpfc_poll_timeout: Restart polling timer.
+ * lpfc_poll_timeout - Restart polling timer
  * @ptr: Map to lpfc_hba data structure pointer.
  *
  * This routine restarts fcp_poll timer, when FCP ring  polling is enable
@@ -1502,15 +2751,16 @@ void lpfc_poll_timeout(unsigned long ptr)
        struct lpfc_hba *phba = (struct lpfc_hba *) ptr;
 
        if (phba->cfg_poll & ENABLE_FCP_RING_POLLING) {
-               lpfc_sli_poll_fcp_ring (phba);
+               lpfc_sli_handle_fast_ring_event(phba,
+                       &phba->sli.ring[LPFC_FCP_RING], HA_R0RE_REQ);
+
                if (phba->cfg_poll & DISABLE_FCP_RING_INT)
                        lpfc_poll_rearm_timer(phba);
        }
 }
 
 /**
- * lpfc_queuecommand: Queuecommand entry point of Scsi Host Templater data
- * structure.
+ * lpfc_queuecommand - scsi_host_template queuecommand entry point
  * @cmnd: Pointer to scsi_cmnd data structure.
  * @done: Pointer to done routine.
  *
@@ -1528,9 +2778,8 @@ lpfc_queuecommand(struct scsi_cmnd *cmnd, void (*done) (struct scsi_cmnd *))
        struct Scsi_Host  *shost = cmnd->device->host;
        struct lpfc_vport *vport = (struct lpfc_vport *) shost->hostdata;
        struct lpfc_hba   *phba = vport->phba;
-       struct lpfc_sli   *psli = &phba->sli;
        struct lpfc_rport_data *rdata = cmnd->device->hostdata;
-       struct lpfc_nodelist *ndlp = rdata->pnode;
+       struct lpfc_nodelist *ndlp;
        struct lpfc_scsi_buf *lpfc_cmd;
        struct fc_rport *rport = starget_to_rport(scsi_target(cmnd->device));
        int err;
@@ -1540,6 +2789,19 @@ lpfc_queuecommand(struct scsi_cmnd *cmnd, void (*done) (struct scsi_cmnd *))
                cmnd->result = err;
                goto out_fail_command;
        }
+       ndlp = rdata->pnode;
+
+       if (!(phba->sli3_options & LPFC_SLI3_BG_ENABLED) &&
+               scsi_get_prot_op(cmnd) != SCSI_PROT_NORMAL) {
+
+               lpfc_printf_log(phba, KERN_ERR, LOG_BG,
+                               "9058 BLKGRD: ERROR: rcvd protected cmd:%02x"
+                               " op:%02x str=%s without registering for"
+                               " BlockGuard - Rejecting command\n",
+                               cmnd->cmnd[0], scsi_get_prot_op(cmnd),
+                               dif_op_str[scsi_get_prot_op(cmnd)]);
+               goto out_fail_command;
+       }
 
        /*
         * Catch race where our node has transitioned, but the
@@ -1574,21 +2836,88 @@ lpfc_queuecommand(struct scsi_cmnd *cmnd, void (*done) (struct scsi_cmnd *))
        cmnd->host_scribble = (unsigned char *)lpfc_cmd;
        cmnd->scsi_done = done;
 
-       err = lpfc_scsi_prep_dma_buf(phba, lpfc_cmd);
+       if (scsi_get_prot_op(cmnd) != SCSI_PROT_NORMAL) {
+               if (vport->phba->cfg_enable_bg) {
+                       lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                               "9033 BLKGRD: rcvd protected cmd:%02x op:%02x "
+                               "str=%s\n",
+                               cmnd->cmnd[0], scsi_get_prot_op(cmnd),
+                               dif_op_str[scsi_get_prot_op(cmnd)]);
+                       lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                               "9034 BLKGRD: CDB: %02x %02x %02x %02x %02x "
+                               "%02x %02x %02x %02x %02x\n",
+                               cmnd->cmnd[0], cmnd->cmnd[1], cmnd->cmnd[2],
+                               cmnd->cmnd[3], cmnd->cmnd[4], cmnd->cmnd[5],
+                               cmnd->cmnd[6], cmnd->cmnd[7], cmnd->cmnd[8],
+                               cmnd->cmnd[9]);
+                       if (cmnd->cmnd[0] == READ_10)
+                               lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                                       "9035 BLKGRD: READ @ sector %llu, "
+                                       "count %u\n",
+                                       (unsigned long long)scsi_get_lba(cmnd),
+                                       blk_rq_sectors(cmnd->request));
+                       else if (cmnd->cmnd[0] == WRITE_10)
+                               lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                                       "9036 BLKGRD: WRITE @ sector %llu, "
+                                       "count %u cmd=%p\n",
+                                       (unsigned long long)scsi_get_lba(cmnd),
+                                       blk_rq_sectors(cmnd->request),
+                                       cmnd);
+               }
+
+               err = lpfc_bg_scsi_prep_dma_buf(phba, lpfc_cmd);
+       } else {
+               if (vport->phba->cfg_enable_bg) {
+                       lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                                       "9038 BLKGRD: rcvd unprotected cmd:"
+                                       "%02x op:%02x str=%s\n",
+                                       cmnd->cmnd[0], scsi_get_prot_op(cmnd),
+                                       dif_op_str[scsi_get_prot_op(cmnd)]);
+                               lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                                       "9039 BLKGRD: CDB: %02x %02x %02x "
+                                       "%02x %02x %02x %02x %02x %02x %02x\n",
+                                       cmnd->cmnd[0], cmnd->cmnd[1],
+                                       cmnd->cmnd[2], cmnd->cmnd[3],
+                                       cmnd->cmnd[4], cmnd->cmnd[5],
+                                       cmnd->cmnd[6], cmnd->cmnd[7],
+                                       cmnd->cmnd[8], cmnd->cmnd[9]);
+                       if (cmnd->cmnd[0] == READ_10)
+                               lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                                       "9040 dbg: READ @ sector %llu, "
+                                       "count %u\n",
+                                       (unsigned long long)scsi_get_lba(cmnd),
+                                        blk_rq_sectors(cmnd->request));
+                       else if (cmnd->cmnd[0] == WRITE_10)
+                               lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                                        "9041 dbg: WRITE @ sector %llu, "
+                                        "count %u cmd=%p\n",
+                                        (unsigned long long)scsi_get_lba(cmnd),
+                                        blk_rq_sectors(cmnd->request), cmnd);
+                       else
+                               lpfc_printf_vlog(vport, KERN_WARNING, LOG_BG,
+                                        "9042 dbg: parser not implemented\n");
+               }
+               err = lpfc_scsi_prep_dma_buf(phba, lpfc_cmd);
+       }
+
        if (err)
                goto out_host_busy_free_buf;
 
        lpfc_scsi_prep_cmnd(vport, lpfc_cmd, ndlp);
 
        atomic_inc(&ndlp->cmd_pending);
-       err = lpfc_sli_issue_iocb(phba, &phba->sli.ring[psli->fcp_ring],
+       err = lpfc_sli_issue_iocb(phba, LPFC_FCP_RING,
                                  &lpfc_cmd->cur_iocbq, SLI_IOCB_RET_IOCB);
        if (err) {
                atomic_dec(&ndlp->cmd_pending);
                goto out_host_busy_free_buf;
        }
        if (phba->cfg_poll & ENABLE_FCP_RING_POLLING) {
-               lpfc_sli_poll_fcp_ring(phba);
+               spin_unlock(shost->host_lock);
+               lpfc_sli_handle_fast_ring_event(phba,
+                       &phba->sli.ring[LPFC_FCP_RING], HA_R0RE_REQ);
+
+               spin_lock(shost->host_lock);
                if (phba->cfg_poll & DISABLE_FCP_RING_INT)
                        lpfc_poll_rearm_timer(phba);
        }
@@ -1607,30 +2936,7 @@ lpfc_queuecommand(struct scsi_cmnd *cmnd, void (*done) (struct scsi_cmnd *))
 }
 
 /**
- * lpfc_block_error_handler: Routine to block error  handler.
- * @cmnd: Pointer to scsi_cmnd data structure.
- *
- *  This routine blocks execution till fc_rport state is not FC_PORSTAT_BLCOEKD.
- **/
-static void
-lpfc_block_error_handler(struct scsi_cmnd *cmnd)
-{
-       struct Scsi_Host *shost = cmnd->device->host;
-       struct fc_rport *rport = starget_to_rport(scsi_target(cmnd->device));
-
-       spin_lock_irq(shost->host_lock);
-       while (rport->port_state == FC_PORTSTATE_BLOCKED) {
-               spin_unlock_irq(shost->host_lock);
-               msleep(1000);
-               spin_lock_irq(shost->host_lock);
-       }
-       spin_unlock_irq(shost->host_lock);
-       return;
-}
-
-/**
- * lpfc_abort_handler: Eh_abort_handler entry point of Scsi Host Template data
- *structure.
+ * lpfc_abort_handler - scsi_host_template eh_abort_handler entry point
  * @cmnd: Pointer to scsi_cmnd data structure.
  *
  * This routine aborts @cmnd pending in base driver.
@@ -1645,7 +2951,6 @@ lpfc_abort_handler(struct scsi_cmnd *cmnd)
        struct Scsi_Host  *shost = cmnd->device->host;
        struct lpfc_vport *vport = (struct lpfc_vport *) shost->hostdata;
        struct lpfc_hba   *phba = vport->phba;
-       struct lpfc_sli_ring *pring = &phba->sli.ring[phba->sli.fcp_ring];
        struct lpfc_iocbq *iocb;
        struct lpfc_iocbq *abtsiocb;
        struct lpfc_scsi_buf *lpfc_cmd;
@@ -1653,7 +2958,7 @@ lpfc_abort_handler(struct scsi_cmnd *cmnd)
        int ret = SUCCESS;
        DECLARE_WAIT_QUEUE_HEAD_ONSTACK(waitq);
 
-       lpfc_block_error_handler(cmnd);
+       fc_block_scsi_eh(cmnd);
        lpfc_cmd = (struct lpfc_scsi_buf *)cmnd->host_scribble;
        BUG_ON(!lpfc_cmd);
 
@@ -1686,10 +2991,18 @@ lpfc_abort_handler(struct scsi_cmnd *cmnd)
        icmd = &abtsiocb->iocb;
        icmd->un.acxri.abortType = ABORT_TYPE_ABTS;
        icmd->un.acxri.abortContextTag = cmd->ulpContext;
-       icmd->un.acxri.abortIoTag = cmd->ulpIoTag;
+       if (phba->sli_rev == LPFC_SLI_REV4)
+               icmd->un.acxri.abortIoTag = iocb->sli4_xritag;
+       else
+               icmd->un.acxri.abortIoTag = cmd->ulpIoTag;
 
        icmd->ulpLe = 1;
        icmd->ulpClass = cmd->ulpClass;
+
+       /* ABTS WQE must go to the same WQ as the WQE to be aborted */
+       abtsiocb->fcp_wqidx = iocb->fcp_wqidx;
+       abtsiocb->iocb_flag |= LPFC_USE_FCPWQIDX;
+
        if (lpfc_is_link_up(phba))
                icmd->ulpCommand = CMD_ABORT_XRI_CN;
        else
@@ -1697,14 +3010,16 @@ lpfc_abort_handler(struct scsi_cmnd *cmnd)
 
        abtsiocb->iocb_cmpl = lpfc_sli_abort_fcp_cmpl;
        abtsiocb->vport = vport;
-       if (lpfc_sli_issue_iocb(phba, pring, abtsiocb, 0) == IOCB_ERROR) {
+       if (lpfc_sli_issue_iocb(phba, LPFC_FCP_RING, abtsiocb, 0) ==
+           IOCB_ERROR) {
                lpfc_sli_release_iocbq(phba, abtsiocb);
                ret = FAILED;
                goto out;
        }
 
        if (phba->cfg_poll & DISABLE_FCP_RING_INT)
-               lpfc_sli_poll_fcp_ring (phba);
+               lpfc_sli_handle_fast_ring_event(phba,
+                       &phba->sli.ring[LPFC_FCP_RING], HA_R0RE_REQ);
 
        lpfc_cmd->waitq = &waitq;
        /* Wait for abort to complete */
@@ -1734,159 +3049,352 @@ lpfc_abort_handler(struct scsi_cmnd *cmnd)
        return ret;
 }
 
+static char *
+lpfc_taskmgmt_name(uint8_t task_mgmt_cmd)
+{
+       switch (task_mgmt_cmd) {
+       case FCP_ABORT_TASK_SET:
+               return "ABORT_TASK_SET";
+       case FCP_CLEAR_TASK_SET:
+               return "FCP_CLEAR_TASK_SET";
+       case FCP_BUS_RESET:
+               return "FCP_BUS_RESET";
+       case FCP_LUN_RESET:
+               return "FCP_LUN_RESET";
+       case FCP_TARGET_RESET:
+               return "FCP_TARGET_RESET";
+       case FCP_CLEAR_ACA:
+               return "FCP_CLEAR_ACA";
+       case FCP_TERMINATE_TASK:
+               return "FCP_TERMINATE_TASK";
+       default:
+               return "unknown";
+       }
+}
+
 /**
- * lpfc_device_reset_handler: eh_device_reset entry point of Scsi Host Template
- *data structure.
- * @cmnd: Pointer to scsi_cmnd data structure.
+ * lpfc_send_taskmgmt - Generic SCSI Task Mgmt Handler
+ * @vport: The virtual port for which this call is being executed.
+ * @rdata: Pointer to remote port local data
+ * @tgt_id: Target ID of remote device.
+ * @lun_id: Lun number for the TMF
+ * @task_mgmt_cmd: type of TMF to send
  *
- * This routine does a device reset by sending a TARGET_RESET task management
- * command.
+ * This routine builds and sends a TMF (SCSI Task Mgmt Function) to
+ * a remote port.
  *
- * Return code :
- *  0x2003 - Error
- *  0ex2002 - Success
+ * Return Code:
+ *   0x2003 - Error
+ *   0x2002 - Success.
  **/
 static int
-lpfc_device_reset_handler(struct scsi_cmnd *cmnd)
+lpfc_send_taskmgmt(struct lpfc_vport *vport, struct lpfc_rport_data *rdata,
+                   unsigned  tgt_id, unsigned int lun_id,
+                   uint8_t task_mgmt_cmd)
 {
-       struct Scsi_Host  *shost = cmnd->device->host;
-       struct lpfc_vport *vport = (struct lpfc_vport *) shost->hostdata;
        struct lpfc_hba   *phba = vport->phba;
        struct lpfc_scsi_buf *lpfc_cmd;
-       struct lpfc_iocbq *iocbq, *iocbqrsp;
-       struct lpfc_rport_data *rdata = cmnd->device->hostdata;
-       struct lpfc_nodelist *pnode = rdata->pnode;
-       unsigned long later;
-       int ret = SUCCESS;
+       struct lpfc_iocbq *iocbq;
+       struct lpfc_iocbq *iocbqrsp;
+       int ret;
        int status;
-       int cnt;
-       struct lpfc_scsi_event_header scsi_event;
-
-       lpfc_block_error_handler(cmnd);
-       /*
-        * If target is not in a MAPPED state, delay the reset until
-        * target is rediscovered or devloss timeout expires.
-        */
-       later = msecs_to_jiffies(2 * vport->cfg_devloss_tmo * 1000) + jiffies;
-       while (time_after(later, jiffies)) {
-               if (!pnode || !NLP_CHK_NODE_ACT(pnode))
-                       return FAILED;
-               if (pnode->nlp_state == NLP_STE_MAPPED_NODE)
-                       break;
-               schedule_timeout_uninterruptible(msecs_to_jiffies(500));
-               rdata = cmnd->device->hostdata;
-               if (!rdata)
-                       break;
-               pnode = rdata->pnode;
-       }
-
-       scsi_event.event_type = FC_REG_SCSI_EVENT;
-       scsi_event.subcategory = LPFC_EVENT_TGTRESET;
-       scsi_event.lun = 0;
-       memcpy(scsi_event.wwpn, &pnode->nlp_portname, sizeof(struct lpfc_name));
-       memcpy(scsi_event.wwnn, &pnode->nlp_nodename, sizeof(struct lpfc_name));
-
-       fc_host_post_vendor_event(shost,
-               fc_get_event_number(),
-               sizeof(scsi_event),
-               (char *)&scsi_event,
-               LPFC_NL_VENDOR_ID);
 
-       if (!rdata || pnode->nlp_state != NLP_STE_MAPPED_NODE) {
-               lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
-                                "0721 LUN Reset rport "
-                                "failure: msec x%x rdata x%p\n",
-                                jiffies_to_msecs(jiffies - later), rdata);
+       if (!rdata->pnode || !NLP_CHK_NODE_ACT(rdata->pnode))
                return FAILED;
-       }
+
        lpfc_cmd = lpfc_get_scsi_buf(phba);
        if (lpfc_cmd == NULL)
                return FAILED;
        lpfc_cmd->timeout = 60;
        lpfc_cmd->rdata = rdata;
 
-       status = lpfc_scsi_prep_task_mgmt_cmd(vport, lpfc_cmd,
-                                             cmnd->device->lun,
-                                             FCP_TARGET_RESET);
+       status = lpfc_scsi_prep_task_mgmt_cmd(vport, lpfc_cmd, lun_id,
+                                          task_mgmt_cmd);
        if (!status) {
                lpfc_release_scsi_buf(phba, lpfc_cmd);
                return FAILED;
        }
-       iocbq = &lpfc_cmd->cur_iocbq;
 
-       /* get a buffer for this IOCB command response */
+       iocbq = &lpfc_cmd->cur_iocbq;
        iocbqrsp = lpfc_sli_get_iocbq(phba);
        if (iocbqrsp == NULL) {
                lpfc_release_scsi_buf(phba, lpfc_cmd);
                return FAILED;
        }
+
        lpfc_printf_vlog(vport, KERN_INFO, LOG_FCP,
-                        "0703 Issue target reset to TGT %d LUN %d "
-                        "rpi x%x nlp_flag x%x\n", cmnd->device->id,
-                        cmnd->device->lun, pnode->nlp_rpi, pnode->nlp_flag);
-       status = lpfc_sli_issue_iocb_wait(phba,
-                                         &phba->sli.ring[phba->sli.fcp_ring],
+                        "0702 Issue %s to TGT %d LUN %d "
+                        "rpi x%x nlp_flag x%x\n",
+                        lpfc_taskmgmt_name(task_mgmt_cmd), tgt_id, lun_id,
+                        rdata->pnode->nlp_rpi, rdata->pnode->nlp_flag);
+
+       status = lpfc_sli_issue_iocb_wait(phba, LPFC_FCP_RING,
                                          iocbq, iocbqrsp, lpfc_cmd->timeout);
-       if (status == IOCB_TIMEDOUT) {
-               iocbq->iocb_cmpl = lpfc_tskmgmt_def_cmpl;
-               ret = TIMEOUT_ERROR;
-       } else {
-               if (status != IOCB_SUCCESS)
+       if (status != IOCB_SUCCESS) {
+               if (status == IOCB_TIMEDOUT) {
+                       iocbq->iocb_cmpl = lpfc_tskmgmt_def_cmpl;
+                       ret = TIMEOUT_ERROR;
+               } else
                        ret = FAILED;
-               lpfc_release_scsi_buf(phba, lpfc_cmd);
-       }
-       lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
-                        "0713 SCSI layer issued device reset (%d, %d) "
-                        "return x%x status x%x result x%x\n",
-                        cmnd->device->id, cmnd->device->lun, ret,
-                        iocbqrsp->iocb.ulpStatus,
+               lpfc_cmd->status = IOSTAT_DRIVER_REJECT;
+               lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
+                        "0727 TMF %s to TGT %d LUN %d failed (%d, %d)\n",
+                        lpfc_taskmgmt_name(task_mgmt_cmd),
+                        tgt_id, lun_id, iocbqrsp->iocb.ulpStatus,
                         iocbqrsp->iocb.un.ulpWord[4]);
+       } else
+               ret = SUCCESS;
+
        lpfc_sli_release_iocbq(phba, iocbqrsp);
-       cnt = lpfc_sli_sum_iocb(vport, cmnd->device->id, cmnd->device->lun,
-                               LPFC_CTX_TGT);
+
+       if (ret != TIMEOUT_ERROR)
+               lpfc_release_scsi_buf(phba, lpfc_cmd);
+
+       return ret;
+}
+
+/**
+ * lpfc_chk_tgt_mapped -
+ * @vport: The virtual port to check on
+ * @cmnd: Pointer to scsi_cmnd data structure.
+ *
+ * This routine delays until the scsi target (aka rport) for the
+ * command exists (is present and logged in) or we declare it non-existent.
+ *
+ * Return code :
+ *  0x2003 - Error
+ *  0x2002 - Success
+ **/
+static int
+lpfc_chk_tgt_mapped(struct lpfc_vport *vport, struct scsi_cmnd *cmnd)
+{
+       struct lpfc_rport_data *rdata = cmnd->device->hostdata;
+       struct lpfc_nodelist *pnode;
+       unsigned long later;
+
+       if (!rdata) {
+               lpfc_printf_vlog(vport, KERN_INFO, LOG_FCP,
+                       "0797 Tgt Map rport failure: rdata x%p\n", rdata);
+               return FAILED;
+       }
+       pnode = rdata->pnode;
+       /*
+        * If target is not in a MAPPED state, delay until
+        * target is rediscovered or devloss timeout expires.
+        */
+       later = msecs_to_jiffies(2 * vport->cfg_devloss_tmo * 1000) + jiffies;
+       while (time_after(later, jiffies)) {
+               if (!pnode || !NLP_CHK_NODE_ACT(pnode))
+                       return FAILED;
+               if (pnode->nlp_state == NLP_STE_MAPPED_NODE)
+                       return SUCCESS;
+               schedule_timeout_uninterruptible(msecs_to_jiffies(500));
+               rdata = cmnd->device->hostdata;
+               if (!rdata)
+                       return FAILED;
+               pnode = rdata->pnode;
+       }
+       if (!pnode || !NLP_CHK_NODE_ACT(pnode) ||
+           (pnode->nlp_state != NLP_STE_MAPPED_NODE))
+               return FAILED;
+       return SUCCESS;
+}
+
+/**
+ * lpfc_reset_flush_io_context -
+ * @vport: The virtual port (scsi_host) for the flush context
+ * @tgt_id: If aborting by Target contect - specifies the target id
+ * @lun_id: If aborting by Lun context - specifies the lun id
+ * @context: specifies the context level to flush at.
+ *
+ * After a reset condition via TMF, we need to flush orphaned i/o
+ * contexts from the adapter. This routine aborts any contexts
+ * outstanding, then waits for their completions. The wait is
+ * bounded by devloss_tmo though.
+ *
+ * Return code :
+ *  0x2003 - Error
+ *  0x2002 - Success
+ **/
+static int
+lpfc_reset_flush_io_context(struct lpfc_vport *vport, uint16_t tgt_id,
+                       uint64_t lun_id, lpfc_ctx_cmd context)
+{
+       struct lpfc_hba   *phba = vport->phba;
+       unsigned long later;
+       int cnt;
+
+       cnt = lpfc_sli_sum_iocb(vport, tgt_id, lun_id, context);
        if (cnt)
                lpfc_sli_abort_iocb(vport, &phba->sli.ring[phba->sli.fcp_ring],
-                                   cmnd->device->id, cmnd->device->lun,
-                                   LPFC_CTX_TGT);
+                                   tgt_id, lun_id, context);
        later = msecs_to_jiffies(2 * vport->cfg_devloss_tmo * 1000) + jiffies;
        while (time_after(later, jiffies) && cnt) {
                schedule_timeout_uninterruptible(msecs_to_jiffies(20));
-               cnt = lpfc_sli_sum_iocb(vport, cmnd->device->id,
-                                       cmnd->device->lun, LPFC_CTX_TGT);
+               cnt = lpfc_sli_sum_iocb(vport, tgt_id, lun_id, context);
        }
        if (cnt) {
                lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
-                                "0719 device reset I/O flush failure: "
-                                "cnt x%x\n", cnt);
-               ret = FAILED;
+                       "0724 I/O flush failure for context %s : cnt x%x\n",
+                       ((context == LPFC_CTX_LUN) ? "LUN" :
+                        ((context == LPFC_CTX_TGT) ? "TGT" :
+                         ((context == LPFC_CTX_HOST) ? "HOST" : "Unknown"))),
+                       cnt);
+               return FAILED;
        }
-       return ret;
+       return SUCCESS;
 }
 
 /**
- * lpfc_bus_reset_handler: eh_bus_reset_handler entry point of Scsi Host
- * Template data structure.
+ * lpfc_device_reset_handler - scsi_host_template eh_device_reset entry point
  * @cmnd: Pointer to scsi_cmnd data structure.
  *
- * This routine does target reset to all target on @cmnd->device->host.
+ * This routine does a device reset by sending a LUN_RESET task management
+ * command.
  *
- * Return Code:
- *   0x2003 - Error
- *   0x2002 - Success
+ * Return code :
+ *  0x2003 - Error
+ *  0x2002 - Success
+ **/
+static int
+lpfc_device_reset_handler(struct scsi_cmnd *cmnd)
+{
+       struct Scsi_Host  *shost = cmnd->device->host;
+       struct lpfc_vport *vport = (struct lpfc_vport *) shost->hostdata;
+       struct lpfc_rport_data *rdata = cmnd->device->hostdata;
+       struct lpfc_nodelist *pnode;
+       unsigned tgt_id = cmnd->device->id;
+       unsigned int lun_id = cmnd->device->lun;
+       struct lpfc_scsi_event_header scsi_event;
+       int status;
+
+       if (!rdata) {
+               lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
+                       "0798 Device Reset rport failure: rdata x%p\n", rdata);
+               return FAILED;
+       }
+       pnode = rdata->pnode;
+       fc_block_scsi_eh(cmnd);
+
+       status = lpfc_chk_tgt_mapped(vport, cmnd);
+       if (status == FAILED) {
+               lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
+                       "0721 Device Reset rport failure: rdata x%p\n", rdata);
+               return FAILED;
+       }
+
+       scsi_event.event_type = FC_REG_SCSI_EVENT;
+       scsi_event.subcategory = LPFC_EVENT_LUNRESET;
+       scsi_event.lun = lun_id;
+       memcpy(scsi_event.wwpn, &pnode->nlp_portname, sizeof(struct lpfc_name));
+       memcpy(scsi_event.wwnn, &pnode->nlp_nodename, sizeof(struct lpfc_name));
+
+       fc_host_post_vendor_event(shost, fc_get_event_number(),
+               sizeof(scsi_event), (char *)&scsi_event, LPFC_NL_VENDOR_ID);
+
+       status = lpfc_send_taskmgmt(vport, rdata, tgt_id, lun_id,
+                                               FCP_LUN_RESET);
+
+       lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
+                        "0713 SCSI layer issued Device Reset (%d, %d) "
+                        "return x%x\n", tgt_id, lun_id, status);
+
+       /*
+        * We have to clean up i/o as : they may be orphaned by the TMF;
+        * or if the TMF failed, they may be in an indeterminate state.
+        * So, continue on.
+        * We will report success if all the i/o aborts successfully.
+        */
+       status = lpfc_reset_flush_io_context(vport, tgt_id, lun_id,
+                                               LPFC_CTX_LUN);
+       return status;
+}
+
+/**
+ * lpfc_target_reset_handler - scsi_host_template eh_target_reset entry point
+ * @cmnd: Pointer to scsi_cmnd data structure.
+ *
+ * This routine does a target reset by sending a TARGET_RESET task management
+ * command.
+ *
+ * Return code :
+ *  0x2003 - Error
+ *  0x2002 - Success
+ **/
+static int
+lpfc_target_reset_handler(struct scsi_cmnd *cmnd)
+{
+       struct Scsi_Host  *shost = cmnd->device->host;
+       struct lpfc_vport *vport = (struct lpfc_vport *) shost->hostdata;
+       struct lpfc_rport_data *rdata = cmnd->device->hostdata;
+       struct lpfc_nodelist *pnode;
+       unsigned tgt_id = cmnd->device->id;
+       unsigned int lun_id = cmnd->device->lun;
+       struct lpfc_scsi_event_header scsi_event;
+       int status;
+
+       if (!rdata) {
+               lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
+                       "0799 Target Reset rport failure: rdata x%p\n", rdata);
+               return FAILED;
+       }
+       pnode = rdata->pnode;
+       fc_block_scsi_eh(cmnd);
+
+       status = lpfc_chk_tgt_mapped(vport, cmnd);
+       if (status == FAILED) {
+               lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
+                       "0722 Target Reset rport failure: rdata x%p\n", rdata);
+               return FAILED;
+       }
+
+       scsi_event.event_type = FC_REG_SCSI_EVENT;
+       scsi_event.subcategory = LPFC_EVENT_TGTRESET;
+       scsi_event.lun = 0;
+       memcpy(scsi_event.wwpn, &pnode->nlp_portname, sizeof(struct lpfc_name));
+       memcpy(scsi_event.wwnn, &pnode->nlp_nodename, sizeof(struct lpfc_name));
+
+       fc_host_post_vendor_event(shost, fc_get_event_number(),
+               sizeof(scsi_event), (char *)&scsi_event, LPFC_NL_VENDOR_ID);
+
+       status = lpfc_send_taskmgmt(vport, rdata, tgt_id, lun_id,
+                                       FCP_TARGET_RESET);
+
+       lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
+                        "0723 SCSI layer issued Target Reset (%d, %d) "
+                        "return x%x\n", tgt_id, lun_id, status);
+
+       /*
+        * We have to clean up i/o as : they may be orphaned by the TMF;
+        * or if the TMF failed, they may be in an indeterminate state.
+        * So, continue on.
+        * We will report success if all the i/o aborts successfully.
+        */
+       status = lpfc_reset_flush_io_context(vport, tgt_id, lun_id,
+                                       LPFC_CTX_TGT);
+       return status;
+}
+
+/**
+ * lpfc_bus_reset_handler - scsi_host_template eh_bus_reset_handler entry point
+ * @cmnd: Pointer to scsi_cmnd data structure.
+ *
+ * This routine does target reset to all targets on @cmnd->device->host.
+ * This emulates Parallel SCSI Bus Reset Semantics.
+ *
+ * Return code :
+ *  0x2003 - Error
+ *  0x2002 - Success
  **/
 static int
 lpfc_bus_reset_handler(struct scsi_cmnd *cmnd)
 {
        struct Scsi_Host  *shost = cmnd->device->host;
        struct lpfc_vport *vport = (struct lpfc_vport *) shost->hostdata;
-       struct lpfc_hba   *phba = vport->phba;
        struct lpfc_nodelist *ndlp = NULL;
-       int match;
-       int ret = SUCCESS, status = SUCCESS, i;
-       int cnt;
-       struct lpfc_scsi_buf * lpfc_cmd;
-       unsigned long later;
        struct lpfc_scsi_event_header scsi_event;
+       int match;
+       int ret = SUCCESS, status, i;
 
        scsi_event.event_type = FC_REG_SCSI_EVENT;
        scsi_event.subcategory = LPFC_EVENT_BUSRESET;
@@ -1894,13 +3402,11 @@ lpfc_bus_reset_handler(struct scsi_cmnd *cmnd)
        memcpy(scsi_event.wwpn, &vport->fc_portname, sizeof(struct lpfc_name));
        memcpy(scsi_event.wwnn, &vport->fc_nodename, sizeof(struct lpfc_name));
 
-       fc_host_post_vendor_event(shost,
-               fc_get_event_number(),
-               sizeof(scsi_event),
-               (char *)&scsi_event,
-               LPFC_NL_VENDOR_ID);
+       fc_host_post_vendor_event(shost, fc_get_event_number(),
+               sizeof(scsi_event), (char *)&scsi_event, LPFC_NL_VENDOR_ID);
+
+       fc_block_scsi_eh(cmnd);
 
-       lpfc_block_error_handler(cmnd);
        /*
         * Since the driver manages a single bus device, reset all
         * targets known to the driver.  Should any target reset
@@ -1923,16 +3429,11 @@ lpfc_bus_reset_handler(struct scsi_cmnd *cmnd)
                spin_unlock_irq(shost->host_lock);
                if (!match)
                        continue;
-               lpfc_cmd = lpfc_get_scsi_buf(phba);
-               if (lpfc_cmd) {
-                       lpfc_cmd->timeout = 60;
-                       status = lpfc_scsi_tgt_reset(lpfc_cmd, vport, i,
-                                                    cmnd->device->lun,
-                                                    ndlp->rport->dd_data);
-                       if (status != TIMEOUT_ERROR)
-                               lpfc_release_scsi_buf(phba, lpfc_cmd);
-               }
-               if (!lpfc_cmd || status != SUCCESS) {
+
+               status = lpfc_send_taskmgmt(vport, ndlp->rport->dd_data,
+                                       i, 0, FCP_TARGET_RESET);
+
+               if (status != SUCCESS) {
                        lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
                                         "0700 Bus Reset on target %d failed\n",
                                         i);
@@ -1940,33 +3441,23 @@ lpfc_bus_reset_handler(struct scsi_cmnd *cmnd)
                }
        }
        /*
-        * All outstanding txcmplq I/Os should have been aborted by
-        * the targets.  Unfortunately, some targets do not abide by
-        * this forcing the driver to double check.
+        * We have to clean up i/o as : they may be orphaned by the TMFs
+        * above; or if any of the TMFs failed, they may be in an
+        * indeterminate state.
+        * We will report success if all the i/o aborts successfully.
         */
-       cnt = lpfc_sli_sum_iocb(vport, 0, 0, LPFC_CTX_HOST);
-       if (cnt)
-               lpfc_sli_abort_iocb(vport, &phba->sli.ring[phba->sli.fcp_ring],
-                                   0, 0, LPFC_CTX_HOST);
-       later = msecs_to_jiffies(2 * vport->cfg_devloss_tmo * 1000) + jiffies;
-       while (time_after(later, jiffies) && cnt) {
-               schedule_timeout_uninterruptible(msecs_to_jiffies(20));
-               cnt = lpfc_sli_sum_iocb(vport, 0, 0, LPFC_CTX_HOST);
-       }
-       if (cnt) {
-               lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
-                                "0715 Bus Reset I/O flush failure: "
-                                "cnt x%x left x%x\n", cnt, i);
+
+       status = lpfc_reset_flush_io_context(vport, 0, 0, LPFC_CTX_HOST);
+       if (status != SUCCESS)
                ret = FAILED;
-       }
+
        lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
                         "0714 SCSI layer issued Bus Reset Data: x%x\n", ret);
        return ret;
 }
 
 /**
- * lpfc_slave_alloc: slave_alloc entry point of Scsi Host Template data
- * structure.
+ * lpfc_slave_alloc - scsi_host_template slave_alloc entry point
  * @sdev: Pointer to scsi_device.
  *
  * This routine populates the cmds_per_lun count + 2 scsi_bufs into  this host's
@@ -1983,11 +3474,10 @@ lpfc_slave_alloc(struct scsi_device *sdev)
 {
        struct lpfc_vport *vport = (struct lpfc_vport *) sdev->host->hostdata;
        struct lpfc_hba   *phba = vport->phba;
-       struct lpfc_scsi_buf *scsi_buf = NULL;
        struct fc_rport *rport = starget_to_rport(scsi_target(sdev));
-       uint32_t total = 0, i;
+       uint32_t total = 0;
        uint32_t num_to_alloc = 0;
-       unsigned long flags;
+       int num_allocated = 0;
 
        if (!rport || fc_remote_port_chkready(rport))
                return -ENXIO;
@@ -2021,27 +3511,21 @@ lpfc_slave_alloc(struct scsi_device *sdev)
                                 (phba->cfg_hba_queue_depth - total));
                num_to_alloc = phba->cfg_hba_queue_depth - total;
        }
-
-       for (i = 0; i < num_to_alloc; i++) {
-               scsi_buf = lpfc_new_scsi_buf(vport);
-               if (!scsi_buf) {
-                       lpfc_printf_vlog(vport, KERN_ERR, LOG_FCP,
-                                        "0706 Failed to allocate "
-                                        "command buffer\n");
-                       break;
-               }
-
-               spin_lock_irqsave(&phba->scsi_buf_list_lock, flags);
-               phba->total_scsi_bufs++;
-               list_add_tail(&scsi_buf->list, &phba->lpfc_scsi_buf_list);
-               spin_unlock_irqrestore(&phba->scsi_buf_list_lock, flags);
+       num_allocated = lpfc_new_scsi_buf(vport, num_to_alloc);
+       if (num_to_alloc != num_allocated) {
+                       lpfc_printf_vlog(vport, KERN_WARNING, LOG_FCP,
+                                "0708 Allocation request of %d "
+                                "command buffers did not succeed.  "
+                                "Allocated %d buffers.\n",
+                                num_to_alloc, num_allocated);
        }
+       if (num_allocated > 0)
+               phba->total_scsi_bufs += num_allocated;
        return 0;
 }
 
 /**
- * lpfc_slave_configure: slave_configure entry point of Scsi Host Templater data
- *  structure.
+ * lpfc_slave_configure - scsi_host_template slave_configure entry point
  * @sdev: Pointer to scsi_device.
  *
  * This routine configures following items
@@ -2073,7 +3557,8 @@ lpfc_slave_configure(struct scsi_device *sdev)
        rport->dev_loss_tmo = vport->cfg_devloss_tmo;
 
        if (phba->cfg_poll & ENABLE_FCP_RING_POLLING) {
-               lpfc_sli_poll_fcp_ring(phba);
+               lpfc_sli_handle_fast_ring_event(phba,
+                       &phba->sli.ring[LPFC_FCP_RING], HA_R0RE_REQ);
                if (phba->cfg_poll & DISABLE_FCP_RING_INT)
                        lpfc_poll_rearm_timer(phba);
        }
@@ -2082,7 +3567,7 @@ lpfc_slave_configure(struct scsi_device *sdev)
 }
 
 /**
- * lpfc_slave_destroy: slave_destroy entry point of SHT data structure.
+ * lpfc_slave_destroy - slave_destroy entry point of SHT data structure
  * @sdev: Pointer to scsi_device.
  *
  * This routine sets @sdev hostatdata filed to null.
@@ -2101,7 +3586,8 @@ struct scsi_host_template lpfc_template = {
        .info                   = lpfc_info,
        .queuecommand           = lpfc_queuecommand,
        .eh_abort_handler       = lpfc_abort_handler,
-       .eh_device_reset_handler= lpfc_device_reset_handler,
+       .eh_device_reset_handler = lpfc_device_reset_handler,
+       .eh_target_reset_handler = lpfc_target_reset_handler,
        .eh_bus_reset_handler   = lpfc_bus_reset_handler,
        .slave_alloc            = lpfc_slave_alloc,
        .slave_configure        = lpfc_slave_configure,
@@ -2113,6 +3599,8 @@ struct scsi_host_template lpfc_template = {
        .use_clustering         = ENABLE_CLUSTERING,
        .shost_attrs            = lpfc_hba_attrs,
        .max_sectors            = 0xFFFF,
+       .vendor_id              = LPFC_NL_VENDOR_ID,
+       .change_queue_depth     = lpfc_change_queue_depth,
 };
 
 struct scsi_host_template lpfc_vport_template = {
@@ -2121,7 +3609,8 @@ struct scsi_host_template lpfc_vport_template = {
        .info                   = lpfc_info,
        .queuecommand           = lpfc_queuecommand,
        .eh_abort_handler       = lpfc_abort_handler,
-       .eh_device_reset_handler= lpfc_device_reset_handler,
+       .eh_device_reset_handler = lpfc_device_reset_handler,
+       .eh_target_reset_handler = lpfc_target_reset_handler,
        .eh_bus_reset_handler   = lpfc_bus_reset_handler,
        .slave_alloc            = lpfc_slave_alloc,
        .slave_configure        = lpfc_slave_configure,
@@ -2133,4 +3622,5 @@ struct scsi_host_template lpfc_vport_template = {
        .use_clustering         = ENABLE_CLUSTERING,
        .shost_attrs            = lpfc_vport_attrs,
        .max_sectors            = 0xFFFF,
+       .change_queue_depth     = lpfc_change_queue_depth,
 };