[SCSI] Merge branch 'linus'
[pandora-kernel.git] / drivers / scsi / lpfc / lpfc_scsi.c
index 8032c5a..e9fa676 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              *
 #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"
@@ -57,6 +59,8 @@ static char *dif_op_str[] = {
        "SCSI_PROT_READ_CONVERT",
        "SCSI_PROT_WRITE_CONVERT"
 };
+static void
+lpfc_release_scsi_buf_s4(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb);
 
 static void
 lpfc_debug_save_data(struct scsi_cmnd *cmnd)
@@ -325,7 +329,7 @@ lpfc_ramp_down_queue_handler(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) {
                                new_queue_depth =
@@ -379,7 +383,7 @@ lpfc_ramp_up_queue_handler(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) {
                                if (vports[i]->cfg_lun_queue_depth <=
@@ -427,7 +431,7 @@ 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));
@@ -438,22 +442,23 @@ 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;
@@ -463,107 +468,401 @@ 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_s4(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->status = IOSTAT_SUCCESS;
+                       spin_unlock_irqrestore(
+                               &phba->sli4_hba.abts_scsi_buf_list_lock,
+                               iflag);
+                       lpfc_release_scsi_buf_s4(phba, psb);
+                       return;
+               }
+       }
+       spin_unlock_irqrestore(&phba->sli4_hba.abts_scsi_buf_list_lock,
+                               iflag);
+}
+
+/**
+ * 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;
+
+               if (index == phba->sli4_hba.scsi_xri_cnt - 1)
+                       /* End of XRI array for SCSI buffer, complete */
+                       bcnt = rcnt;
+
+               /* 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->status = IOSTAT_LOCAL_REJECT;
+                               psb->result = IOERR_ABORT_REQUESTED;
+                               rc++;
+                       } else
+                               psb->status = IOSTAT_SUCCESS;
+                       /* Put it back into the SCSI buffer list */
+                       lpfc_release_scsi_buf_s4(phba, psb);
+               }
        }
-       psb->cur_iocbq.iocb_flag |= LPFC_IO_FCP;
+       return rc;
+}
 
-       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_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;
 
-       /* 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);
+               /*
+                * 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;
+               }
 
-       /*
-        * 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);
+               /* Initialize virtual ptrs to dma_buf region. */
+               memset(psb->data, 0, phba->cfg_sg_dma_buf_size);
 
-       /*
-        * 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;
+               /* 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_len, sgl, sizeof(struct fcp_cmnd));
+               bf_set(lpfc_sli4_sge_last, sgl, 0);
+               sgl->word2 = cpu_to_le32(sgl->word2);
+               sgl->word3 = cpu_to_le32(sgl->word3);
+               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_len, sgl, sizeof(struct fcp_rsp));
+               bf_set(lpfc_sli4_sge_last, sgl, 1);
+               sgl->word2 = cpu_to_le32(sgl->word2);
+               sgl->word3 = cpu_to_le32(sgl->word3);
+
+               /*
+                * 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->status = IOSTAT_LOCAL_REJECT;
+                               psb->result = IOERR_ABORT_REQUESTED;
+                               rc++;
+                       } else
+                               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->status = IOSTAT_LOCAL_REJECT;
+                               psb->result = IOERR_ABORT_REQUESTED;
+                               rc++;
+                       } else
+                               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.
@@ -591,7 +890,7 @@ lpfc_get_scsi_buf(struct lpfc_hba * phba)
 }
 
 /**
- * lpfc_release_scsi_buf - Return a scsi buffer back to hba's lpfc_scsi_buf_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.
  *
@@ -599,7 +898,7 @@ 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;
 
@@ -610,21 +909,69 @@ 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->status == IOSTAT_LOCAL_REJECT
+               && psb->result == IOERR_ABORT_REQUESTED) {
+               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;
@@ -1411,6 +1758,133 @@ 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) {
+                       printk(KERN_ERR "%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);
+                       bf_set(lpfc_sli4_sge_len, sgl, 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->word3 = cpu_to_le32(sgl->word3);
+                       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.
@@ -1504,15 +1978,15 @@ 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_s3 - Un-map DMA mapping of SG-list for SLI3 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_s3(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
 {
        /*
         * There are only two special cases to consider.  (1) the scsi command
@@ -1528,6 +2002,36 @@ lpfc_scsi_unprep_dma_buf(struct lpfc_hba * phba, struct lpfc_scsi_buf * psb)
                                psb->pCmd->sc_data_direction);
 }
 
+/**
+ * lpfc_scsi_unprep_dma_buf_s4 - Un-map DMA mapping of SG-list for SLI4 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 for device with SLI-4 interface spec. If we have to
+ * remove the sgl for this scsi buffer then we will do it here. For now
+ * we should be able to just call the sli3 unprep routine.
+ **/
+static void
+lpfc_scsi_unprep_dma_buf_s4(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
+{
+       lpfc_scsi_unprep_dma_buf_s3(phba, psb);
+}
+
+/**
+ * lpfc_scsi_unprep_dma_buf - Wrapper function for unmap DMA mapping of SG-list
+ * @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 for device with SLI-4 interface spec.
+ **/
+static void
+lpfc_scsi_unprep_dma_buf(struct lpfc_hba *phba, struct lpfc_scsi_buf *psb)
+{
+       phba->lpfc_scsi_unprep_dma_buf(phba, psb);
+}
+
 /**
  * lpfc_handler_fcp_err - FCP response handler
  * @vport: The virtual port for which this call is being executed.
@@ -1676,7 +2180,7 @@ lpfc_handle_fcp_err(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
  * 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
@@ -1957,16 +2461,16 @@ 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_s3 - Convert scsi cmnd to FCP infor unit for SLI3 dev
  * @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,
+lpfc_scsi_prep_cmnd_s3(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
                    struct lpfc_nodelist *pnode)
 {
        struct lpfc_hba *phba = vport->phba;
@@ -2013,8 +2517,11 @@ 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 {
@@ -2051,20 +2558,60 @@ 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_cmnd_s4 - Convert scsi cmnd to FCP infor unit for SLI4 dev
+ * @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 for device with SLI4 interface spec.
+ **/
+static void
+lpfc_scsi_prep_cmnd_s4(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
+                      struct lpfc_nodelist *pnode)
+{
+       /*
+        * The prep cmnd routines do not touch the sgl or its
+        * entries. We may not have to do anything different.
+        * I will leave this function in place until we can
+        * run some IO through the driver and determine if changes
+        * are needed.
+        */
+       return lpfc_scsi_prep_cmnd_s3(vport, lpfc_cmd, pnode);
+}
+
+/**
+ * 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 wraps the actual convert SCSI cmnd function pointer from
+ * the lpfc_hba struct.
+ **/
+static inline void
+lpfc_scsi_prep_cmnd(struct lpfc_vport *vport, struct lpfc_scsi_buf *lpfc_cmd,
+                   struct lpfc_nodelist *pnode)
+{
+       vport->phba->lpfc_scsi_prep_cmnd(vport, lpfc_cmd, pnode);
+}
+
+/**
+ * lpfc_scsi_prep_task_mgmt_cmnd_s3 - 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
  *   1 - Success
  **/
 static int
-lpfc_scsi_prep_task_mgmt_cmd(struct lpfc_vport *vport,
+lpfc_scsi_prep_task_mgmt_cmd_s3(struct lpfc_vport *vport,
                             struct lpfc_scsi_buf *lpfc_cmd,
                             unsigned int lun,
                             uint8_t task_mgmt_cmd)
@@ -2113,6 +2660,107 @@ lpfc_scsi_prep_task_mgmt_cmd(struct lpfc_vport *vport,
        return 1;
 }
 
+/**
+ * lpfc_scsi_prep_task_mgmt_cmnd_s4 - Convert SLI4 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
+ * for device with SLI-4 interface spec.
+ *
+ * Return codes:
+ *     0 - Error
+ *     1 - Success
+ **/
+static int
+lpfc_scsi_prep_task_mgmt_cmd_s4(struct lpfc_vport *vport,
+                               struct lpfc_scsi_buf *lpfc_cmd,
+                               unsigned int lun,
+                               uint8_t task_mgmt_cmd)
+{
+       /*
+        * The prep cmnd routines do not touch the sgl or its
+        * entries. We may not have to do anything different.
+        * I will leave this function in place until we can
+        * run some IO through the driver and determine if changes
+        * are needed.
+        */
+       return lpfc_scsi_prep_task_mgmt_cmd_s3(vport, lpfc_cmd, lun,
+                                               task_mgmt_cmd);
+}
+
+/**
+ * lpfc_scsi_prep_task_mgmt_cmnd - Wrapper func convert scsi TM cmd to FCP info
+ * @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 wraps the actual convert SCSI TM to FCP information unit
+ * function pointer from the lpfc_hba struct.
+ *
+ * Return codes:
+ *     0 - Error
+ *     1 - Success
+ **/
+static inline int
+lpfc_scsi_prep_task_mgmt_cmd(struct lpfc_vport *vport,
+                            struct lpfc_scsi_buf *lpfc_cmd,
+                            unsigned int lun,
+                            uint8_t task_mgmt_cmd)
+{
+       struct lpfc_hba *phba = vport->phba;
+
+       return phba->lpfc_scsi_prep_task_mgmt_cmd(vport, lpfc_cmd, lun,
+                                                 task_mgmt_cmd);
+}
+
+/**
+ * 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)
+{
+
+       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_scsi_prep_cmnd = lpfc_scsi_prep_cmnd_s3;
+               phba->lpfc_scsi_unprep_dma_buf = lpfc_scsi_unprep_dma_buf_s3;
+               phba->lpfc_scsi_prep_task_mgmt_cmd =
+                                       lpfc_scsi_prep_task_mgmt_cmd_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_scsi_prep_cmnd = lpfc_scsi_prep_cmnd_s4;
+               phba->lpfc_scsi_unprep_dma_buf = lpfc_scsi_unprep_dma_buf_s4;
+               phba->lpfc_scsi_prep_task_mgmt_cmd =
+                                       lpfc_scsi_prep_task_mgmt_cmd_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;
+       return 0;
+}
+
 /**
  * lpfc_taskmgmt_def_cmpl - IOCB completion routine for task management command
  * @phba: The Hba for which this call is being executed.
@@ -2178,9 +2826,8 @@ lpfc_scsi_tgt_reset(struct lpfc_scsi_buf *lpfc_cmd, struct lpfc_vport *vport,
        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);
+       status = lpfc_sli_issue_iocb_wait(phba, LPFC_FCP_RING,
+                                         iocbq, iocbqrsp, lpfc_cmd->timeout);
        if (status != IOCB_SUCCESS) {
                if (status == IOCB_TIMEDOUT) {
                        iocbq->iocb_cmpl = lpfc_tskmgmt_def_cmpl;
@@ -2305,7 +2952,6 @@ 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_scsi_buf *lpfc_cmd;
@@ -2427,7 +3073,7 @@ lpfc_queuecommand(struct scsi_cmnd *cmnd, void (*done) (struct scsi_cmnd *))
        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);
@@ -2490,7 +3136,6 @@ 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;
@@ -2531,7 +3176,10 @@ 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;
@@ -2542,7 +3190,8 @@ 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;
@@ -2668,8 +3317,7 @@ lpfc_device_reset_handler(struct scsi_cmnd *cmnd)
                         "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],
+       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;
@@ -2825,11 +3473,10 @@ 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;
@@ -2863,20 +3510,13 @@ 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);
        }
        return 0;
 }