Re: [PATCH 1/6] RFC: beiscsi : handles core routines

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



I might have given you some of the comments below offlist. Sorry about the duplicates.


On 08/13/2009 02:08 AM, Jayamohan Kallickal wrote:
This file handles initiallization/teardown, allocation/free as well
as IO/Management flows.

Signed-off-by: Jayamohan Kallickal<jayamohank@xxxxxxxxxxxxxxxxx>
---
  drivers/scsi/beiscsi/be_main.c | 3733 ++++++++++++++++++++++++++++++++++++++++
  1 files changed, 3733 insertions(+), 0 deletions(-)
  create mode 100644 drivers/scsi/beiscsi/be_main.c

diff --git a/drivers/scsi/beiscsi/be_main.c b/drivers/scsi/beiscsi/be_main.c
new file mode 100644
index 0000000..2cb0b1c

+
+static struct scsi_host_template beiscsi_sht = {
+	.module = THIS_MODULE,
+	.name = "ServerEngines 10Gbe open-iscsi Initiator Driver",
+	.proc_name = DRV_NAME,
+	.queuecommand = iscsi_queuecommand,
+	.eh_abort_handler = iscsi_eh_abort,
+	.eh_device_reset_handler = iscsi_eh_device_reset,
+	.eh_target_reset_handler = iscsi_eh_target_reset,
+	.sg_tablesize = BEISCSI_SGLIST_ELEMENTS,
+	.can_queue = BEISCSI_MAX_CMNDS,
+	.this_id = -1,
+	.max_sectors = 127,


What is up with this limit? Is it a hardware or driver limit?


+	.cmd_per_lun = BEISCSI_CMD_PER_LUN,
+	.use_clustering = ENABLE_CLUSTERING,

You could set the .change_queue_depth to iscsi_change_queue_depth.


+};
+static struct scsi_transport_template *beiscsi_scsi_transport;
+
+/*------------------- PCI Driver operations and data ----------------- */
+static DEFINE_PCI_DEVICE_TABLE(beiscsi_pci_id_table) = {
+	{ PCI_DEVICE(BE_VENDOR_ID, BE_DEVICE_ID1) },
+	{ PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID1) },
+	{ PCI_DEVICE(BE_VENDOR_ID, OC_DEVICE_ID2) },
+	{ 0 }
+};
+MODULE_DEVICE_TABLE(pci, beiscsi_pci_id_table);
+
+static struct pci_driver beiscsi_pci_driver = {
+	.name = DRV_NAME,
+	.probe = beiscsi_dev_probe,
+	.remove = beiscsi_remove,
+	.id_table = beiscsi_pci_id_table
+};
+
+struct beiscsi_hba *beiscsi_hba_alloc(struct pci_dev *pcidev)
+{


Should be static.


+	struct beiscsi_hba *phba;
+	struct Scsi_Host *shost;
+
+	shost = iscsi_host_alloc(&beiscsi_sht, sizeof(*phba), 0);
+	if (!shost) {
+		dev_err(&pcidev->dev, "beiscsi_hba_alloc -"
+				"iscsi_host_alloc failed \n");
+		return NULL;
+	}
+	shost->dma_boundary = pcidev->dma_mask;
+	shost->max_id = BE2_MAX_SESSIONS;

Can you really only support 64 sessions?

Targets like EQL do target per lun, so 64 runs out quickly.


+	shost->max_channel = 0;
+	shost->max_cmd_len = BEISCSI_MAX_CMD_LEN;
+	shost->max_lun = BEISCSI_NUM_MAX_LUN;
+	shost->transportt = beiscsi_scsi_transport;
+	shost->can_queue = BEISCSI_MAX_CMNDS;
+
+	phba = iscsi_host_priv(shost);
+	memset(phba, 0x0, sizeof(*phba));
+	phba->shost = shost;
+	pci_dev_get(pcidev);
+
+	phba->shost = shost;
+	phba->pcidev = pcidev;
+
+	if (iscsi_host_add(shost,&phba->pcidev->dev))

Add space after ",".


+		goto free_devices;
+	return phba;
+
+free_devices:
+	pci_dev_put(phba->pcidev);
+	return NULL;
+}
+
+static int be_map_pci_bars(struct beiscsi_hba *phba, struct pci_dev *pdev)
+{
+	u8 __iomem *addr;
+
+	phba->csr_va = NULL;
+	phba->db_va = NULL;
+	phba->pci_va = NULL;
+
+	addr = ioremap_nocache(pci_resource_start(pdev, 2),
+			       pci_resource_len(pdev, 2));
+	if (addr == NULL)
+		return -ENOMEM;
+	phba->ctrl.csr = addr;
+	phba->csr_va = addr;
+
+	addr = ioremap_nocache(pci_resource_start(pdev, 4), 128 * 1024);
+	if (addr == NULL)
+		goto pci_map_err;
+	phba->ctrl.db = addr;
+	phba->db_va = addr;
+
+	addr = ioremap_nocache(pci_resource_start(pdev, 1),
+			       pci_resource_len(pdev, 1));
+	if (addr == NULL)
+		goto pci_map_err;
+	phba->ctrl.pcicfg = addr;
+	phba->pci_va = addr;
+
+	return 0;
+pci_map_err:
+	beiscsi_unmap_pci_function(phba);
+	return -ENOMEM;
+}
+
+static int be_iopoll(struct blk_iopoll *iop, int budget)
+{
+	unsigned int ret = 0;
+	struct beiscsi_hba *phba;
+
+	phba = container_of(iop, struct beiscsi_hba, iopoll);
+
+	if (!enable_msix)
+		ret = beiscsi_process_mcc_cq(phba);
+
+	ret = beiscsi_process_cq(phba);
+	if (ret<  budget) {
+		struct hwi_controller *phwi_ctrlr;
+		struct hwi_context_memory *phwi_context;
+
+		phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);


What is up with this macro? Just wondering because it seems more work to type all that out vs just doing phba->phwi_ctrlr.



+		phwi_context = phwi_ctrlr->phwi_ctxt;
+		blk_iopoll_complete(iop);
+		hwi_ring_eq_db(phba, phwi_context->be_eq.q.id, 0,
+							0, 1, 1);
+	}
+	return ret;
+}
+
+static int be_ctrl_init(struct beiscsi_hba *phba, struct pci_dev *pdev)
+{
+	struct be_ctrl_info *ctrl =&phba->ctrl;
+	struct be_dma_mem *mbox_mem_alloc =&ctrl->mbox_mem_alloced;
+	struct be_dma_mem *mbox_mem_align =&ctrl->mbox_mem;
+	int status = 0;
+
+	ctrl->pdev = pdev;
+	status = be_map_pci_bars(phba, pdev);
+	if (status)
+		return status;
+
+	mbox_mem_alloc->size = sizeof(struct be_mcc_mailbox) + 16;
+	mbox_mem_alloc->va = pci_alloc_consistent(pdev,
+						  mbox_mem_alloc->size,
+						&mbox_mem_alloc->dma);
+	if (!mbox_mem_alloc->va) {
+		beiscsi_unmap_pci_function(phba);
+		status = -ENOMEM;
+		return status;
+	}
+
+	mbox_mem_align->size = sizeof(struct be_mcc_mailbox);
+	mbox_mem_align->va = PTR_ALIGN(mbox_mem_alloc->va, 16);
+	mbox_mem_align->dma = PTR_ALIGN(mbox_mem_alloc->dma, 16);
+	memset(mbox_mem_align->va, 0, sizeof(struct be_mcc_mailbox));
+	spin_lock_init(&ctrl->mbox_lock);
+	return status;
+}
+
+int __devinit beiscsi_dev_probe(struct pci_dev *pcidev,


should be static


+				const struct pci_device_id *id)
+{
+	struct beiscsi_hba *phba = NULL;
+	int ret = -1;
+
+	ret = beiscsi_enable_pci(pcidev);
+	if (ret<  0) {
+		dev_err(&pcidev->dev, "beiscsi_dev_probe-"
+				"Failed to enable pci device \n");
+		return ret;
+	}
+
+	phba = beiscsi_hba_alloc(pcidev);
+	if (!phba) {
+		dev_err(&pcidev->dev, "beiscsi_dev_probe-"
+				" Failed in beiscsi_hba_alloc \n");
+		goto disable_pci;
+	}
+
+	pci_set_drvdata(pcidev, phba);
+
+	ret = beiscsi_init_pci_function(phba, pcidev);
+	if (ret<  0) {


space after ret.

+		dev_err(&pcidev->dev, "beiscsi_dev_probe-"
+				"Failed in beiscsi_init_pci_function \n");
+		goto hba_free;
+	}
+
+	ret = be_ctrl_init(phba, pcidev);
+	if (ret) {
+		dev_err(&pcidev->dev, "beiscsi_dev_probe-"
+				"Failed in be_ctrl_init\n");
+		goto hba_free;
+	}
+
+	spin_lock_init(&phba->sgl_lock);
+	spin_lock_init(&phba->isr_lock);
+	beiscsi_get_params(phba);
+
+	ret = beiscsi_init_port(phba);
+	if (ret<  0) {
+		dev_err(&pcidev->dev, "beiscsi_dev_probe-"
+				"Failed in beiscsi_init_port\n");
+		goto free_controller;
+	}
+
+	snprintf(phba->wq_name, sizeof(phba->wq_name), "beiscsi_queue_irqs");
+

I think you want to add a id in there in case there are multiple hosts.

snprintf(phba->wq_name, sizeof(phba->wq_name), "beiscsi_queue_irqs%u", phba->shost->host_no);


+	phba->wq = create_singlethread_workqueue(phba->wq_name);
+	if (!phba->wq) {
+		dev_err(&pcidev->dev, "beiscsi_dev_probe-"
+				"Failed to allocate work queue\n");
+		goto free_controller;
+	}
+
+	if (!blk_iopoll_enabled)
+		INIT_WORK(&phba->work_cqs, beiscsi_process_all_cqs);
+	else {
+		blk_iopoll_init(&phba->iopoll, be_iopoll_budget, be_iopoll);
+		blk_iopoll_enable(&phba->iopoll);
+	}


Can you enable/disable blk napi dynamically? If so then you might always want to do INIT_WORK.


+	ret = beiscsi_init_irqs(phba);
+	if (ret<  0)
+		goto free_controller;
+	ret = hwi_enable_intr(phba);
+	if (ret<  0) {
+		dev_err(&pcidev->dev, "beiscsi_dev_probe-"
+				"Failed to hwi_enable_intr\n");
+		goto free_controller;
+	}
+
+	SE_DEBUG(DBG_LVL_8, "\n\n\n SUCCESS - DRIVER LOADED \n\n\n");
+	return 0;
+
+free_controller:



If we fail but the workqueue was created then you need to desotry the workqueue by calling destroy_workqueue. Also in the normal remaval path you need to do this.


I think you also leave resournces allocated by beiscsi_init_port and some memory. I think you need a beiscsi_clean_port and beiscsi_free_mem call.



+	if (blk_iopoll_enabled)
+		blk_iopoll_disable(&phba->iopoll);
+	iscsi_host_remove(phba->shost);
+	iscsi_host_free(phba->shost);
+	beiscsi_unmap_pci_function(phba);
+hba_free:
+	pci_dev_put(phba->pcidev);
+disable_pci:
+	pci_disable_device(pcidev);
+	return ret;


I think in some functions you switch between different error codes. I think you want to try and return something consistent. ret is set to -1, but other places you return a -EXYZ value.


+}
+
+int beiscsi_init_irqs(struct beiscsi_hba *phba)
+{
+	struct pci_dev *pcidev = phba->pcidev;
+	int ret;
+
+	ret = request_irq(pcidev->irq, be_isr, IRQF_SHARED, "beiscsi", phba);
+	if (ret)	{
+		dev_err(&pcidev->dev, "beiscsi_init_irqs-"
+					"Failed to register irq\\n");
+		return ret;
+	}
+	return 0;
+}
+
+void beiscsi_get_params(struct beiscsi_hba *phba)
+{
+	phba->params.ios_per_ctrl =
+	    MAX_ICDS / 2 - (LOGOUTS + TMFS + NOPOUT_REQ);

The names seem a little generic. Maybe add a BE2 in front of them.

What is ios_per_ctrl? Do you want shost/scsi_host_template->can_queue to be this value?

What are the defined values? For TMFS, are you planning on only being able to handle 16 tmfs at one time? Same for NOPOUT_REQs? For NOPOUT_REQ I think you would want that to be a couple per session.



+	phba->params.cxns_per_ctrl = BE2_MAX_SESSIONS;
+	phba->params.asyncpdus_per_ctrl = ASYNCPDUS;
+	phba->params.icds_per_ctrl = MAX_ICDS / 2;
+	phba->params.num_sge_per_io = SGE;
+	phba->params.defpdu_hdr_sz = DEFPDU_HDR_SZ;
+	phba->params.defpdu_data_sz = DEFPDU_DATA_SZ;


What are the defpdu_hdr_sz and defpdu_data_sz values?

For the defpdu_data_sz does it mean you want max xmit/recv data segment length values to be that value?


+	phba->params.eq_timer = 64;
+	phba->params.num_eq_entries =
+	    (((CMDS_PER_CXN * 2 + LOGOUTS + TMFS + ASYNCPDUS) / 512) + 1) * 512;


What is num_eq_entries? Does it relate to the number of commands you can send the card?


+	phba->params.num_eq_entries = (phba->params.num_eq_entries<  1024)

space after entries and before "<"


+				? 1024 : phba->params.num_eq_entries;
+	SE_DEBUG(DBG_LVL_8, "phba->params.num_eq_entries=%d \n",
+				phba->params.num_eq_entries);
+	phba->params.num_cq_entries =
+	    (((CMDS_PER_CXN * 2 + LOGOUTS + TMFS + ASYNCPDUS) / 512) + 1) * 512;


What is num_cq_entries?


+	SE_DEBUG(DBG_LVL_8,
+		 "phba->params.num_cq_entries=%d CMDS_PER_CXN=%d LOGOUTS=%d"
+		 "TMFS=%d ASYNCPDUS=%d \n", phba->params.num_cq_entries,
+		 CMDS_PER_CXN, LOGOUTS, TMFS, ASYNCPDUS);
+	phba->params.wrbs_per_cxn = 256;	/* Later to increase to 512 */
+}
+
+/*
+ * be_isr : The isr routine of the driver.
+ * @irq:  Not used
+ * @dev_id: Pointer to host adapter structure
+ */
+irqreturn_t be_isr(int irq, void *dev_id)
+{

static


+	struct beiscsi_hba *phba;
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_context_memory *phwi_context;
+	struct be_eq_entry *eqe = NULL;
+	struct be_queue_info *eq;
+	struct be_queue_info *cq;
+	unsigned long flags, index;
+	unsigned int num_eq_processed;
+	struct be_ctrl_info *ctrl;
+	int isr;
+
+	phba = dev_id;
+	if (!enable_msix) {
+		ctrl =&phba->ctrl;;
+		isr = ioread32(ctrl->csr + CEV_ISR0_OFFSET +
+			(PCI_FUNC(ctrl->pdev->devfn) * CEV_ISR_SIZE));
+		if (!isr)
+			return IRQ_NONE;
+	}
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	phwi_context = phwi_ctrlr->phwi_ctxt;
+	eq =&phwi_context->be_eq.q;
+	cq =&phwi_context->be_cq;
+	index = 0;
+	eqe = queue_tail_node(eq);
+	if (!eqe)
+		SE_DEBUG(DBG_LVL_1, "eqe is NULL\n");
+
+	num_eq_processed = 0;
+	if (blk_iopoll_enabled) {
+		while (eqe->dw[offsetof(struct amap_eq_entry, valid) / 32]
+					&  EQE_VALID_MASK) {
+			if (blk_iopoll_sched_prep(&phba->iopoll))
+					blk_iopoll_sched(&phba->iopoll);
+
+			AMAP_SET_BITS(struct amap_eq_entry, valid, eqe, 0);
+			queue_tail_inc(eq);
+			eqe = queue_tail_node(eq);
+			num_eq_processed++;
+			SE_DEBUG(DBG_LVL_8, "Vaid EQE\n");
+		}
+		if (num_eq_processed) {
+			hwi_ring_eq_db(phba, eq->id, 0,	num_eq_processed, 0, 1);
+			return IRQ_HANDLED;
+		} else {
+			return IRQ_NONE;
+		}
+	} else {
+		while (eqe->dw[offsetof(struct amap_eq_entry, valid) / 32]
+						&  EQE_VALID_MASK) {
+
+			if (((eqe->dw[offsetof(struct amap_eq_entry,
+			resource_id) / 32]&  EQE_RESID_MASK)>>16) != cq->id) {
+				spin_lock_irqsave(&phba->isr_lock, flags);
+				phba->todo_mcc_cq = 1;
+				spin_unlock_irqrestore(&phba->isr_lock, flags);
+			} else {
+				spin_lock_irqsave(&phba->isr_lock, flags);
+				phba->todo_cq = 1;
+				spin_unlock_irqrestore(&phba->isr_lock, flags);
+			}
+			AMAP_SET_BITS(struct amap_eq_entry, valid, eqe, 0);
+			queue_tail_inc(eq);
+			eqe = queue_tail_node(eq);
+			num_eq_processed++;
+		}
+		if (phba->todo_cq || phba->todo_mcc_cq)
+			queue_work(phba->wq,&phba->work_cqs);
+
+		if (num_eq_processed) {
+			hwi_ring_eq_db(phba, eq->id, 0,
+						num_eq_processed, 1, 1);
+			return IRQ_HANDLED;
+		} else {
+			return IRQ_NONE;
+		}
+	}
+}
+
+void beiscsi_process_all_cqs(struct work_struct *work)


static. How about from here on, just check all your functions in all your files.


+{
+	unsigned long flags;
+	unsigned int  ret;
+	struct beiscsi_hba *phba =
+	    container_of(work, struct beiscsi_hba, work_cqs);
+
+
+	if (phba->todo_mcc_cq) {
+		spin_lock_irqsave(&phba->isr_lock, flags);
+		phba->todo_mcc_cq = 0;
+		spin_unlock_irqrestore(&phba->isr_lock, flags);
+		ret = beiscsi_process_mcc_cq(phba);


ret is not checked here and below.

+	}
+
+	if (phba->todo_cq) {
+		spin_lock_irqsave(&phba->isr_lock, flags);
+		phba->todo_cq = 0;
+		spin_unlock_irqrestore(&phba->isr_lock, flags);
+		ret = beiscsi_process_cq(phba);
+	}
+}
+
+int beiscsi_init_pci_function(struct beiscsi_hba *phba, struct pci_dev *pcidev)
+{
+
+	/* CSR */
+	phba->csr_pa.u.a64.address = pci_resource_start(pcidev, 2);
+
+	/* Door Bell */
+	phba->db_pa.u.a64.address =  pci_resource_start(pcidev, 4);
+
+	/* PCI */
+	phba->pci_pa.u.a64.address = pci_resource_start(pcidev, 1);
+
+	return 0;
+}
+
+void beiscsi_unmap_pci_function(struct beiscsi_hba *phba)
+{
+	if (phba->csr_va) {
+		iounmap(phba->csr_va);
+		phba->csr_va = NULL;
+	}
+	if (phba->db_va) {
+		iounmap(phba->db_va);
+		phba->db_va = NULL;
+	}
+	if (phba->pci_va) {
+		iounmap(phba->pci_va);
+		phba->pci_va = NULL;
+	}
+}
+
+int beiscsi_enable_pci(struct pci_dev *pcidev)
+{
+	if (pci_enable_device(pcidev)) {
+		dev_err(&pcidev->dev, "beiscsi_enable_pci - enable device "
+			"failed. Returning -ENODEV\n");
+		return -ENODEV;
+	}
+
+	if (pci_set_consistent_dma_mask(pcidev, DMA_BIT_MASK(64))) {
+		if (pci_set_consistent_dma_mask(pcidev, DMA_BIT_MASK(32))) {
+			dev_err(&pcidev->dev, "Could not set PCI DMA Mask\n");
+			return -ENODEV;
+		}
+	}
+	return 0;
+}
+
+/* Under host lock */
+static struct sgl_handle *alloc_io_sgl_handle(struct beiscsi_hba *phba)
+{
+	struct sgl_handle *psgl_handle;
+
+	if (phba->io_sgl_hndl_avbl) {
+		SE_DEBUG(DBG_LVL_8,
+			 "In alloc_io_sgl_handle,io_sgl_alloc_index=%d \n",
+						phba->io_sgl_alloc_index);
+		psgl_handle = phba->io_sgl_hndl_base[phba->
+						io_sgl_alloc_index];
+		phba->io_sgl_hndl_base[phba->io_sgl_alloc_index] = NULL;
+		phba->io_sgl_hndl_avbl--;
+		if (phba->io_sgl_alloc_index == (phba->params.ios_per_ctrl - 1))
+			phba->io_sgl_alloc_index = 0;
+		else
+			phba->io_sgl_alloc_index++;
+	} else {
+		psgl_handle = NULL;
+	}
+	return psgl_handle;
+}
+
+/*  Under host lock */
+static void
+free_io_sgl_handle(struct beiscsi_hba *phba, struct sgl_handle *psgl_handle)
+{
+	SE_DEBUG(DBG_LVL_8, "In free_io_sgl_handle ,io_sgl_free_index=%d \n",
+						 phba->io_sgl_free_index);
+	if (phba->io_sgl_hndl_base[phba->io_sgl_free_index]) {
+		SE_DEBUG(DBG_LVL_8,
+			 "Double Free in IO SGL io_sgl_free_index=%d,"
+			 "value there=%p \n", phba->io_sgl_free_index,
+			 phba->io_sgl_hndl_base[phba->io_sgl_free_index]);
+		return;
+	}
+	phba->io_sgl_hndl_base[phba->io_sgl_free_index] = psgl_handle;
+	phba->io_sgl_hndl_avbl++;
+	if (phba->io_sgl_free_index == (phba->params.ios_per_ctrl - 1))
+		phba->io_sgl_free_index = 0;
+	else
+		phba->io_sgl_free_index++;
+}
+
+/* This happens under session_lock until
+ * submission to chip
+ */
+struct wrb_handle *alloc_wrb_handle(struct beiscsi_hba *phba, unsigned int cid)
+{
+	struct hwi_wrb_context *pwrb_context;
+	struct hwi_controller *phwi_ctrlr;
+	struct wrb_handle *pwrb_handle;
+	struct wrb_handle *pwrb_handle_tmp;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	pwrb_context =&phwi_ctrlr->wrb_context[cid];
+
+	if (pwrb_context->wrb_handles_available) {
+		pwrb_handle = pwrb_context->
+				pwrb_handle_base[pwrb_context->alloc_index];
+		pwrb_context->wrb_handles_available--;
+		if (pwrb_context->alloc_index ==
+		    (phba->params.wrbs_per_cxn - 1)) {
+			pwrb_context->alloc_index = 0;
+		} else
+			pwrb_context->alloc_index++;
+
+		pwrb_handle_tmp = pwrb_context->
+				    pwrb_handle_base[pwrb_context->alloc_index];
+		pwrb_handle_tmp->nxt_wrb_index = pwrb_handle_tmp->wrb_index;
+	} else
+		pwrb_handle = NULL;
+	return pwrb_handle;
+}
+
+/* This is invoked
+ * under session_lock
+ */
+static void
+free_wrb_handle(struct beiscsi_hba *phba, struct hwi_wrb_context *pwrb_context,
+		struct wrb_handle *pwrb_handle)
+{
+	pwrb_context->pwrb_handle_base[pwrb_context->free_index] = pwrb_handle;
+	pwrb_context->wrb_handles_available++;
+	if (pwrb_context->free_index == (phba->params.wrbs_per_cxn - 1))
+		pwrb_context->free_index = 0;
+	else
+		pwrb_context->free_index++;
+	SE_DEBUG(DBG_LVL_8,
+		 "FREE WRB: pwrb_handle=%p free_index=%d=0x%x"
+		 "wrb_handles_available=%d \n",
+		 pwrb_handle, pwrb_context->free_index,
+		 pwrb_context->free_index, pwrb_context->wrb_handles_available);
+}
+
+/* Under host lock */
+static struct sgl_handle *alloc_eh_sgl_handle(struct beiscsi_hba *phba)
+{
+	struct sgl_handle *psgl_handle;
+
+	if (phba->eh_sgl_hndl_avbl) {
+		psgl_handle = phba->eh_sgl_hndl_base[phba->
+							 eh_sgl_alloc_index];
+		phba->eh_sgl_hndl_base[phba->eh_sgl_alloc_index] = NULL;
+		SE_DEBUG(DBG_LVL_8, "eh_sgl_alloc_index=%d=0x%x \n",
+			 phba->eh_sgl_alloc_index, phba->eh_sgl_alloc_index);
+		phba->eh_sgl_hndl_avbl--;
+		if (phba->eh_sgl_alloc_index ==
+		    (phba->params.icds_per_ctrl - phba->params.ios_per_ctrl -
+		     1))
+			phba->eh_sgl_alloc_index = 0;
+		else
+			phba->eh_sgl_alloc_index++;
+	} else
+		psgl_handle = NULL;
+	return psgl_handle;
+}
+
+/*  Under host lock */
+void
+free_eh_sgl_handle(struct beiscsi_hba *phba, struct sgl_handle *psgl_handle)
+{
+
+	if (phba->eh_sgl_hndl_base[phba->eh_sgl_free_index]) {
+		dev_warn(&phba->pcidev->dev,
+			 "Double Free in eh SGL ,eh_sgl_free_index=%d \n",
+			 phba->eh_sgl_free_index);
+		return;
+	}
+	phba->eh_sgl_hndl_base[phba->eh_sgl_free_index] =
+	    (struct sgl_handle *)psgl_handle;
+	phba->eh_sgl_hndl_avbl++;
+	if (phba->eh_sgl_free_index ==
+	    (phba->params.icds_per_ctrl - phba->params.ios_per_ctrl - 1))
+		phba->eh_sgl_free_index = 0;
+	else
+		phba->eh_sgl_free_index++;
+}
+
+static struct list_head *hwi_get_async_busy_list(struct hwi_async_pdu_context
+					  *pasync_ctx, unsigned int is_header,
+					  unsigned int host_write_ptr)
+{
+	if (is_header)
+		return&pasync_ctx->async_entry[host_write_ptr].
+		    header_busy_list;
+	else
+		return&pasync_ctx->async_entry[host_write_ptr].data_busy_list;
+}
+
+static struct async_pdu_handle *hwi_get_async_handle(struct beiscsi_hba *phba,
+					struct beiscsi_conn *beiscsi_conn,
+				struct hwi_async_pdu_context *pasync_ctx,
+				      struct i_t_dpdu_cqe *pdpdu_cqe,
+						unsigned int *pcq_index)
+{
+	struct be_bus_address phys_addr;
+	struct list_head *pbusy_list, *plink;
+	struct async_pdu_handle *pasync_handle = NULL;
+	int buffer_len = 0;
+	unsigned char buffer_index = -1;
+	unsigned char is_header = 0;
+
+	/* This function is invoked to get the right async_handle structure
+	 * from   a given default PDU CQ entry.
+	 * The CQ entry contains the address offset where the DMA last ended.
+	 * So subtract the dataplacementlength to get to the base
+	 * */
+	phys_addr.u.a32.address_lo =
+	    pdpdu_cqe->dw[offsetof(struct amap_i_t_dpdu_cqe, db_addr_lo) / 32] -
+	    ((pdpdu_cqe->dw[offsetof(struct amap_i_t_dpdu_cqe, dpl) / 32]
+						&  PDUCQE_DPL_MASK)>>  16);
+	phys_addr.u.a32.address_hi =
+	    pdpdu_cqe->dw[offsetof(struct amap_i_t_dpdu_cqe, db_addr_hi) / 32];
+
+	phys_addr.u.a64.address =
+	    *((unsigned long long *)(&phys_addr.u.a64.address));
+
+	switch (pdpdu_cqe->dw[offsetof(struct amap_i_t_dpdu_cqe, code) / 32]
+				&  PDUCQE_CODE_MASK) {
+	case UNSOL_HDR_NOTIFY:
+
+		is_header = 1;
+
+		/* the index in the CQ entry gives the vertical index */
+		pbusy_list = hwi_get_async_busy_list(pasync_ctx, 1,
+			    (pdpdu_cqe->dw[offsetof(struct amap_i_t_dpdu_cqe,
+					index) / 32]&  PDUCQE_INDEX_MASK));
+
+		/* and the buf_index from the PA will help walk the list
+		 * horizontally
+		 */
+		buffer_len = (uint)(phys_addr.u.a64.address -
+			pasync_ctx->async_header.pa_base.u.a64.address);
+
+		buffer_index =
+		    buffer_len / pasync_ctx->async_header.buffer_size;
+
+		break;
+
+	case UNSOL_DATA_NOTIFY:
+
+		/* the index in the CQ entry gives the vertical index */
+		pbusy_list =
+		    hwi_get_async_busy_list(pasync_ctx, 0, (pdpdu_cqe->
+					  dw[offsetof(struct amap_i_t_dpdu_cqe,
+					  index) / 32]&  PDUCQE_INDEX_MASK));
+
+		/* and the buf_index from the PA will help walk the list
+		 * horizontally
+		 */
+		buffer_len = (unsigned long)(phys_addr.u.a64.address -
+					      pasync_ctx->async_data.pa_base.u.
+					      a64.address);
+		buffer_index = buffer_len / pasync_ctx->async_data.buffer_size;
+		break;
+	default:
+		pbusy_list = NULL;
+		dev_warn(&phba->pcidev->dev,
+			"Unexpected code=%d \n",
+			 pdpdu_cqe->
+			 dw[offsetof(struct amap_i_t_dpdu_cqe, code) / 32]
+							&  PDUCQE_CODE_MASK);
+		return NULL;
+	}
+
+	WARN_ON(!(buffer_index<= pasync_ctx->async_data.num_entries));
+	WARN_ON(list_empty(pbusy_list));
+	/* walk the list and locate the default_handle with index buf_index */
+	plink = pbusy_list->next;
+
+	while (plink != pbusy_list) {


I think you want list_for_each_entry_safe

+		pasync_handle = list_entry(plink, struct async_pdu_handle,
+					   link);
+
+		WARN_ON(!pasync_handle);
+		WARN_ON(pasync_handle->consumed);
+
+		if (pasync_handle->index == buffer_index) {
+			/* found a match, exit out of the loop */
+			break;
+		}
+
+		plink = plink->next;
+		pasync_handle = NULL;
+	}
+
+	WARN_ON(!pasync_handle);
+
+	pasync_handle->cri = (unsigned short)beiscsi_conn->beiscsi_conn_cid;
+	pasync_handle->is_header = is_header;
+	pasync_handle->buffer_len =
+		((pdpdu_cqe->
+		dw[offsetof(struct amap_i_t_dpdu_cqe, dpl) / 32]
+					&  PDUCQE_DPL_MASK)>>  16);
+
+	*pcq_index = (pdpdu_cqe->dw[offsetof(struct amap_i_t_dpdu_cqe,
+			index) / 32]&  PDUCQE_INDEX_MASK);
+	return pasync_handle;
+}
+
+static unsigned int
+hwi_update_async_writables(struct hwi_async_pdu_context *pasync_ctx,
+			   unsigned int is_header, unsigned int cq_index)
+{
+	struct list_head *pbusy_list;
+	struct async_pdu_handle *pasync_handle;
+	unsigned int num_entries, writables = 0;
+	unsigned int *pep_read_ptr, *pwritables;
+
+	/* Some buffers have been consumed and the last was cq_index.
+	 * So from the last ep_read_ptr upto cq_index,
+	 * update all default_handles to have a consumed state.
+	 */
+
+	if (is_header) {
+		pep_read_ptr =&pasync_ctx->async_header.ep_read_ptr;
+		pwritables =&pasync_ctx->async_header.writables;
+		num_entries = pasync_ctx->async_header.num_entries;
+	} else {
+		pep_read_ptr =&pasync_ctx->async_data.ep_read_ptr;
+		pwritables =&pasync_ctx->async_data.writables;
+		num_entries = pasync_ctx->async_data.num_entries;
+	}
+
+	while ((*pep_read_ptr) != cq_index) {
+		/* Pick only the first entry in the list for modification.
+		 * Though the default_list can grow horizontally, we only
+		 * need to update the element at the head.
+		 */
+		(*pep_read_ptr)++;
+		*pep_read_ptr = (*pep_read_ptr) % num_entries;
+
+		pbusy_list =
+		    hwi_get_async_busy_list(pasync_ctx, is_header,
+					    *pep_read_ptr);
+
+		if (writables == 0)
+			WARN_ON(list_empty(pbusy_list));
+
+		if (!list_empty(pbusy_list)) {
+			/* valid entry found, mark it consumed */
+			pasync_handle = list_entry(pbusy_list->next,
+						struct async_pdu_handle, link);
+			WARN_ON(!pasync_handle);
+			pasync_handle->consumed = 1;
+		}
+
+		/* increment the writables since its completed and we
+		 * could re-post
+		 * */
+		writables++;
+	}
+
+	/* no updates were done?? This is a EP fw error! */
+	if (!writables) {
+		SE_DEBUG(DBG_LVL_1,
+			 "Duplicate notification received - index 0x%x!!\n",
+			 cq_index);
+		WARN_ON(1);
+	}
+
+	/* update the writables in the corresponding structures */
+	*pwritables = *pwritables + writables;
+	return 0;
+}
+
+static unsigned int hwi_free_async_msg(struct beiscsi_hba *phba,
+					unsigned int cri)
+{
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_async_pdu_context *pasync_ctx;
+	struct async_pdu_handle *pasync_handle;
+	struct list_head *plink;
+	unsigned int i = 0;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	pasync_ctx = HWI_GET_ASYNC_PDU_CTX(phwi_ctrlr);
+
+	plink = pasync_ctx->async_entry[cri].wait_queue.list.next;
+
+	while (plink !=&pasync_ctx->async_entry[cri].wait_queue.list) {

can this just be a list_for_each_entry_safe too?


space between != and &

+		pasync_handle =
+		    list_entry(plink, struct async_pdu_handle, link);
+		WARN_ON(!pasync_handle);
+
+		plink = plink->next;
+		list_del(&pasync_handle->link);
+
+		if (i == 0) {
+			list_add_tail(&pasync_handle->link,
+				&pasync_ctx->async_header.free_list);
+			pasync_ctx->async_header.free_entries++;
+			i++;
+		} else {
+			list_add_tail(&pasync_handle->link,
+				&pasync_ctx->async_data.free_list);
+			pasync_ctx->async_data.free_entries++;
+			i++;
+		}
+	}
+
+	/* reset the list header */
+	INIT_LIST_HEAD(&pasync_ctx->async_entry[cri].wait_queue.list);
+	pasync_ctx->async_entry[cri].wait_queue.hdr_received = 0;
+	pasync_ctx->async_entry[cri].wait_queue.bytes_received = 0;
+
+	return 0;
+}
+
+static struct phys_addr *hwi_get_ring_address(struct hwi_async_pdu_context
+					*pasync_ctx, unsigned int is_header,
+					unsigned int host_write_ptr)
+{
+	struct phys_addr *pasync_sge = NULL;
+
+	if (is_header)
+		pasync_sge = pasync_ctx->async_header.ring_base;
+	else
+		pasync_sge = pasync_ctx->async_data.ring_base;
+
+	return pasync_sge + host_write_ptr;
+}
+
+static void hwi_post_async_buffers(struct beiscsi_hba *phba,
+					unsigned int is_header)
+{
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_async_pdu_context *pasync_ctx;
+	struct async_pdu_handle *pasync_handle;
+	struct list_head *pfree_link, *pbusy_list;
+	struct phys_addr *pasync_sge;
+	unsigned int ring_id, num_entries;
+	unsigned int host_write_num;
+	unsigned int writables;
+	unsigned int i = 0;
+	u32 doorbell = 0;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	pasync_ctx = HWI_GET_ASYNC_PDU_CTX(phwi_ctrlr);
+
+	/* The rule during posting buffers into the default PDU ring
+	 * is that the number of entries posted must be in multiples of 8.
+	 * There are 2 resources that matter, the free_entries and the
+	 * writables. free_entries is the count of number of buffers
+	 * iSCSI has finished
+	 * processing and have posted them back. writables is the count
+	 * of numberof buffers we know the EP firmware has consumed
+	 * */
+
+	if (is_header) {
+		num_entries = pasync_ctx->async_header.num_entries;
+		writables = min(pasync_ctx->async_header.writables,
+				pasync_ctx->async_header.free_entries);
+		pfree_link = pasync_ctx->async_header.free_list.next;
+		host_write_num = pasync_ctx->async_header.host_write_ptr;
+		ring_id = phwi_ctrlr->default_pdu_hdr.id;
+	} else {
+		num_entries = pasync_ctx->async_data.num_entries;
+		writables = min(pasync_ctx->async_data.writables,
+				pasync_ctx->async_data.free_entries);
+		pfree_link = pasync_ctx->async_data.free_list.next;
+		host_write_num = pasync_ctx->async_data.host_write_ptr;
+		ring_id = phwi_ctrlr->default_pdu_data.id;
+	}
+
+	writables = (writables / 8) * 8;
+
+	if (writables) {
+		for (i = 0; i<  writables; i++) {
+
+			pbusy_list =
+			    hwi_get_async_busy_list(pasync_ctx, is_header,
+						    host_write_num);
+			pasync_handle =
+			    list_entry(pfree_link, struct async_pdu_handle,
+								link);
+			WARN_ON(!pasync_handle);
+			pasync_handle->consumed = 0;
+
+			pfree_link = pfree_link->next;
+
+			pasync_sge = hwi_get_ring_address(pasync_ctx,
+						is_header, host_write_num);
+
+			pasync_sge->hi = pasync_handle->pa.u.a32.address_lo;
+			pasync_sge->lo = pasync_handle->pa.u.a32.address_hi;
+
+			/* remove from free list and add to busy list */
+			list_del(&pasync_handle->link);
+			list_add(&pasync_handle->link, pbusy_list);


list_move()/list_move_tail()



+
+			host_write_num++;
+			host_write_num = host_write_num % num_entries;
+		}
+
+		if (is_header) {
+			pasync_ctx->async_header.host_write_ptr =
+							host_write_num;
+			pasync_ctx->async_header.free_entries -= writables;
+			pasync_ctx->async_header.writables -= writables;
+			pasync_ctx->async_header.busy_entries += writables;
+		} else {
+			pasync_ctx->async_data.host_write_ptr = host_write_num;
+			pasync_ctx->async_data.free_entries -= writables;
+			pasync_ctx->async_data.writables -= writables;
+			pasync_ctx->async_data.busy_entries += writables;
+		}
+
+		doorbell |= ring_id&  DB_DEF_PDU_RING_ID_MASK;
+		doorbell |= 1<<  DB_DEF_PDU_REARM_SHIFT;
+		doorbell |= 0<<  DB_DEF_PDU_EVENT_SHIFT;
+		doorbell |= (writables&  DB_DEF_PDU_CQPROC_MASK)
+					<<  DB_DEF_PDU_CQPROC_SHIFT;
+
+		iowrite32(doorbell, phba->db_va + DB_RXULP0_OFFSET);
+	}
+
+}
+
+/* This function is called when there is a default PDU buffer digest
+ * (data digest error). The driver consumes the entry and drops
+ * the contents. The EP fw gives a CQ entry indicating completion
+ * on either the default PDU header ring the default PDU buffer ring.
+ * The CQ entry will contain the index and the physical address of
+ * the entry. Get our reference from the CQ entry
+ */
+static unsigned int
+hwi_flush_default_pdu_buffer(struct beiscsi_hba *phba,
+			     struct beiscsi_conn *beiscsi_conn,
+			     struct i_t_dpdu_cqe *pdpdu_cqe)
+{
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_async_pdu_context *pasync_ctx;
+	struct async_pdu_handle *pasync_handle = NULL;
+	unsigned int cq_index = -1;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	pasync_ctx = HWI_GET_ASYNC_PDU_CTX(phwi_ctrlr);
+
+	pasync_handle = hwi_get_async_handle(phba, beiscsi_conn, pasync_ctx,
+							pdpdu_cqe,&cq_index);
+	BUG_ON(pasync_handle->is_header != 0);
+	if (pasync_handle->consumed == 0)
+		hwi_update_async_writables(pasync_ctx,
+				   pasync_handle->is_header, cq_index);
+
+	hwi_free_async_msg(phba, pasync_handle->cri);
+	hwi_post_async_buffers(phba, pasync_handle->is_header);
+
+	return 0;
+}
+
+static void
+hwi_write_sgl(struct iscsi_wrb *pwrb, struct scatterlist *sg,
+	      unsigned int num_sg, struct beiscsi_io_task *io_task)
+{
+	struct iscsi_sge *psgl;
+	unsigned short sg_len, index;
+	unsigned int sge_len = 0;
+	unsigned long long addr;
+	struct scatterlist *l_sg;
+	unsigned int offset;
+
+	AMAP_SET_BITS(struct amap_iscsi_wrb, iscsi_bhs_addr_lo, pwrb,
+				      io_task->bhs_pa.u.a32.address_lo);
+	AMAP_SET_BITS(struct amap_iscsi_wrb, iscsi_bhs_addr_hi, pwrb,
+				      io_task->bhs_pa.u.a32.address_hi);
+
+	l_sg = sg;
+	for (index = 0; (index<  num_sg)&&  (index<  2); index++, sg++) {


You want to use the scatterlist accessor helpers in case chaining is used. for_each_sg, sg_next, scsi_sglist, etc.

Add some spaces after index too.



+		if (index == 0) {
+			sg_len = sg_dma_len(sg);
+			addr = (u64) sg_dma_address(sg);
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_addr_lo, pwrb,
+							(addr&  0xFFFFFFFF));
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_addr_hi, pwrb,
+							(addr>>  32));
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_len, pwrb,
+							sg_len);
+			sge_len = sg_len;
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_last, pwrb,
+							1);
+		} else {
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_last, pwrb,
+							0);
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge1_r2t_offset,
+							pwrb, sge_len);
+			sg_len = sg_dma_len(sg);
+			addr = (u64) sg_dma_address(sg);
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge1_addr_lo, pwrb,
+							(addr&  0xFFFFFFFF));
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge1_addr_hi, pwrb,
+							(addr>>  32));
+			AMAP_SET_BITS(struct amap_iscsi_wrb, sge1_len, pwrb,
+							sg_len);
+		}
+	}
+	psgl = (struct iscsi_sge *)io_task->psgl_handle->pfrag;
+	memset(psgl, 0x0, sizeof(*psgl) * SGE);
+
+	AMAP_SET_BITS(struct amap_iscsi_sge, len, psgl, io_task->bhs_len - 2);
+
+	AMAP_SET_BITS(struct amap_iscsi_sge, addr_hi, psgl,
+			io_task->bhs_pa.u.a32.address_hi);
+	AMAP_SET_BITS(struct amap_iscsi_sge, addr_lo, psgl,
+			io_task->bhs_pa.u.a32.address_lo);
+
+	if (num_sg == 2)
+		AMAP_SET_BITS(struct amap_iscsi_wrb, sge1_last, pwrb, 1);
+	sg = l_sg;
+	psgl++;
+	psgl++;
+	offset = 0;
+	for (index = 0; index<  num_sg; index++, sg++, psgl++) {


same as above.

+		sg_len = sg_dma_len(sg);
+		addr = (u64) sg_dma_address(sg);
+		AMAP_SET_BITS(struct amap_iscsi_sge, addr_lo, psgl,
+						(addr&  0xFFFFFFFF));
+		AMAP_SET_BITS(struct amap_iscsi_sge, addr_hi, psgl,
+						(addr>>  32));
+		AMAP_SET_BITS(struct amap_iscsi_sge, len, psgl, sg_len);
+		AMAP_SET_BITS(struct amap_iscsi_sge, sge_offset, psgl, offset);
+		AMAP_SET_BITS(struct amap_iscsi_sge, last_sge, psgl, 0);
+		offset += sg_len;
+	}
+	psgl--;
+	AMAP_SET_BITS(struct amap_iscsi_sge, last_sge, psgl, 1);
+}
+
+static void hwi_write_buffer(struct iscsi_wrb *pwrb, struct iscsi_task *task)
+{
+	struct iscsi_sge *psgl;
+	unsigned long long addr;
+	struct beiscsi_io_task *io_task = task->dd_data;
+	struct beiscsi_conn *beiscsi_conn = io_task->conn;
+	struct beiscsi_hba *phba = beiscsi_conn->phba;
+
+	io_task->bhs_len = sizeof(struct be_nonio_bhs) - 2;
+	io_task->bhs_pa.u.a64.address = virt_to_bus(&io_task->cmd_bhs);
+	AMAP_SET_BITS(struct amap_iscsi_wrb, iscsi_bhs_addr_lo, pwrb,
+				io_task->bhs_pa.u.a32.address_lo);
+	AMAP_SET_BITS(struct amap_iscsi_wrb, iscsi_bhs_addr_hi, pwrb,
+				io_task->bhs_pa.u.a32.address_hi);
+
+	if (task->data) {
+		/* Should set this if sge is present.
+		 * Verify in all cases
+		 * */
+
+		if (task->data_count) {
+			AMAP_SET_BITS(struct amap_iscsi_wrb, dsp, pwrb, 1);
+			addr = (u64) pci_map_single(phba->pcidev,
+						    task->data,
+						    task->data_count, 1);
+		} else {
+			AMAP_SET_BITS(struct amap_iscsi_wrb, dsp, pwrb, 0);
+			addr = 0;
+		}
+		AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_addr_lo, pwrb,
+						(addr&  0xFFFFFFFF));
+		AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_addr_hi, pwrb,
+						(addr>>  32));


space after addr

+		AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_len, pwrb,
+						task->data_count);
+
+		AMAP_SET_BITS(struct amap_iscsi_wrb, sge0_last, pwrb, 1);
+	} else {
+		AMAP_SET_BITS(struct amap_iscsi_wrb, dsp, pwrb, 0);
+		addr = 0;
+	}
+
+	psgl = (struct iscsi_sge *)io_task->psgl_handle->pfrag;
+
+	AMAP_SET_BITS(struct amap_iscsi_sge, len, psgl, io_task->bhs_len);
+
+	AMAP_SET_BITS(struct amap_iscsi_sge, addr_hi, psgl,
+		      io_task->bhs_pa.u.a32.address_hi);
+	AMAP_SET_BITS(struct amap_iscsi_sge, addr_lo, psgl,
+		      io_task->bhs_pa.u.a32.address_lo);
+	if (task->data) {
+		psgl++;
+		AMAP_SET_BITS(struct amap_iscsi_sge, addr_hi, psgl, 0);
+		AMAP_SET_BITS(struct amap_iscsi_sge, addr_lo, psgl, 0);
+		AMAP_SET_BITS(struct amap_iscsi_sge, len, psgl, 0);
+		AMAP_SET_BITS(struct amap_iscsi_sge, sge_offset, psgl, 0);
+		AMAP_SET_BITS(struct amap_iscsi_sge, rsvd0, psgl, 0);
+		AMAP_SET_BITS(struct amap_iscsi_sge, last_sge, psgl, 0);
+
+		psgl++;
+		if (task->data) {
+			AMAP_SET_BITS(struct amap_iscsi_sge, addr_lo, psgl,
+						(addr&  0xFFFFFFFF));
+			AMAP_SET_BITS(struct amap_iscsi_sge, addr_hi, psgl,
+						(addr>>  32));
+		}
+		AMAP_SET_BITS(struct amap_iscsi_sge, len, psgl, 0x106);
+	}
+	AMAP_SET_BITS(struct amap_iscsi_sge, last_sge, psgl, 1);
+}
+
+static unsigned int
+hwi_fwd_async_msg(struct beiscsi_conn *beiscsi_conn,
+		  struct beiscsi_hba *phba,
+		  struct hwi_async_pdu_context *pasync_ctx, unsigned short cri)
+{
+	struct list_head *plink;
+	struct async_pdu_handle *pasync_handle;
+	void *phdr = NULL;
+	unsigned int hdr_len = 0, buf_len = 0;
+	unsigned int status, index = 0, offset = 0;
+	void *pfirst_buffer = NULL;
+	unsigned int num_buf = 0;
+
+	plink = pasync_ctx->async_entry[cri].wait_queue.list.next;
+
+	while (plink !=&pasync_ctx->async_entry[cri].wait_queue.list) {



list_for_each_entry_safe?



+		pasync_handle = list_entry(plink, struct async_pdu_handle,
+					   link);
+
+		if (index == 0) {
+			/* first element is the header, get its reference */
+			phdr = pasync_handle->pbuffer;
+			hdr_len = pasync_handle->buffer_len;
+		} else {
+			buf_len = pasync_handle->buffer_len;
+
+			if (!num_buf) {
+				pfirst_buffer = pasync_handle->pbuffer;
+				num_buf++;
+			}
+
+			memcpy(pfirst_buffer + offset,
+			       pasync_handle->pbuffer, buf_len);
+
+			offset = buf_len;
+		}
+
+		index++;
+		plink = plink->next;
+	}
+
+	status = beiscsi_process_async_pdu(beiscsi_conn, phba,
+				   beiscsi_conn->beiscsi_conn_cid,
+				   phdr, hdr_len, pfirst_buffer, buf_len);
+
+	if (status == 0) {
+		/* iscsi has finished processing the buffers, add it
+		 * to the free pool. This will remove the element
+		 * from wait list and add it to free list
+		 * */
+		hwi_free_async_msg(phba, cri);
+	}
+	return 0;
+}
+
+static unsigned int
+hwi_gather_async_pdu(struct beiscsi_conn *beiscsi_conn,
+		     struct beiscsi_hba *phba,
+		     struct async_pdu_handle *pasync_handle)
+{
+
+	struct hwi_async_pdu_context *pasync_ctx;
+	struct hwi_controller *phwi_ctrlr;
+	unsigned int bytes_needed = 0, status = 0;
+	unsigned short cri = pasync_handle->cri;
+	struct pdu_base *ppdu;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	pasync_ctx = HWI_GET_ASYNC_PDU_CTX(phwi_ctrlr);
+
+	/* remove the element from the busylist and insert into the waitlist */
+	list_del(&pasync_handle->link);
+	if (pasync_handle->is_header) {
+		pasync_ctx->async_header.busy_entries--;
+		if (pasync_ctx->async_entry[cri].wait_queue.hdr_received) {
+			hwi_free_async_msg(phba, cri);
+			BUG();
+		}
+
+		pasync_ctx->async_entry[cri].wait_queue.bytes_received = 0;
+		pasync_ctx->async_entry[cri].wait_queue.hdr_received = 1;
+		pasync_ctx->async_entry[cri].wait_queue.hdr_len =
+		    (unsigned char)pasync_handle->buffer_len;
+		list_add_tail(&pasync_handle->link,
+			&pasync_ctx->async_entry[cri].wait_queue.list);
+
+		ppdu = pasync_handle->pbuffer;
+		bytes_needed = ((((ppdu->dw[offsetof(struct amap_pdu_base,
+			data_len_hi) / 32]&  PDUBASE_DATALENHI_MASK)<<  8)&
+			0xFFFF0000) | ((be16_to_cpu((ppdu->
+			dw[offsetof(struct amap_pdu_base, data_len_lo) / 32]
+			&  PDUBASE_DATALENLO_MASK)>>  16))&  0x0000FFFF));
+
+		if (status == 0) {
+			pasync_ctx->async_entry[cri].wait_queue.bytes_needed =
+			    bytes_needed;
+
+			if (bytes_needed == 0) {
+				/* all data has been received,
+				 * forward it to iscsi
+				 * */
+				status = hwi_fwd_async_msg(beiscsi_conn, phba,
+							pasync_ctx, cri);
+			}
+		}
+	} else {
+		pasync_ctx->async_data.busy_entries--;
+		if (pasync_ctx->async_entry[cri].wait_queue.hdr_received) {
+			list_add_tail(&pasync_handle->link,
+				&pasync_ctx->async_entry[cri].wait_queue.
+				      list);
+			pasync_ctx->async_entry[cri].wait_queue.
+			    bytes_received +=
+			    (unsigned short)pasync_handle->buffer_len;
+
+			if (pasync_ctx->async_entry[cri].wait_queue.
+			    bytes_received>=
+			    pasync_ctx->async_entry[cri].wait_queue.
+			    bytes_needed) {
+				/* all data has been received, forward it to
+				 * iscsi
+				 * */
+				status = hwi_fwd_async_msg(beiscsi_conn, phba,
+						      pasync_ctx, cri);
+			}
+		}
+	}
+	return status;
+}
+
+static unsigned int
+hwi_process_default_pdu_ring(struct beiscsi_conn *beiscsi_conn,
+			     struct beiscsi_hba *phba,
+			     struct i_t_dpdu_cqe *pdpdu_cqe)
+{
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_async_pdu_context *pasync_ctx;
+	struct async_pdu_handle *pasync_handle = NULL;
+	unsigned int cq_index = -1;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	pasync_ctx = HWI_GET_ASYNC_PDU_CTX(phwi_ctrlr);
+	pasync_handle =
+	    hwi_get_async_handle(phba, beiscsi_conn, pasync_ctx, pdpdu_cqe,
+				&cq_index);
+
+	if (pasync_handle->consumed == 0) {
+		hwi_update_async_writables(pasync_ctx,
+					   pasync_handle->is_header, cq_index);
+	}
+	hwi_gather_async_pdu(beiscsi_conn, phba, pasync_handle);
+	hwi_post_async_buffers(phba, pasync_handle->is_header);
+
+	return 0;
+}
+
+/*
+ * Place holder for handling MCC Interrupts
+ */
+unsigned int beiscsi_process_mcc_cq(struct beiscsi_hba *phba)
+{
+	return 0;
+}
+
+unsigned int beiscsi_process_cq(struct beiscsi_hba *phba)
+{
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_context_memory *phwi_context;
+	struct be_queue_info *cq;
+	struct sol_cqe *sol;
+	struct dmsg_cqe *dmsg;
+	unsigned int num_processed = 0;
+	unsigned int tot_nump = 0;
+	struct beiscsi_conn *beiscsi_conn;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	phwi_context = phwi_ctrlr->phwi_ctxt;
+	cq =&phwi_context->be_cq;
+	sol = queue_tail_node(cq);
+
+	while (sol->dw[offsetof(struct amap_sol_cqe, valid) / 32]
+				&  CQE_VALID_MASK) {
+	be_dws_le_to_cpu(sol, sizeof(struct sol_cqe));
+
+	beiscsi_conn = phba->conn_table[(u32) (sol->
+			 dw[offsetof(struct amap_sol_cqe, cid) / 32]
+			&  SOL_CID_MASK)>>  6];
+
+	if (!beiscsi_conn || !beiscsi_conn->ep) {
+		dev_warn(&phba->pcidev->dev,
+			"Connection table empty for cid = %d\n",
+			 (u32)(sol->dw[offsetof(struct amap_sol_cqe, cid) / 32]
+			&  SOL_CID_MASK)>>  6);
+		return 0;
+	}



Ehhh. Not sure what happened here. Extra tabbing maybe or was this in another if statement?



+		if (num_processed>= 32) {
+			/* Occassionally free space in CQ for the hardware to
+			 * write more entries without enabling the interrupt
+			 * yet.
+			 */
+			hwi_ring_cq_db(phba, phwi_context->be_cq.id,
+					num_processed, 0, 0);
+			tot_nump += num_processed;
+			num_processed = 0;
+		}
+
+		switch ((u32) sol->dw[offsetof(struct amap_sol_cqe, code) / 32]
+				&  CQE_CODE_MASK) {
+		case SOL_CMD_COMPLETE:
+			hwi_complete_cmd(beiscsi_conn, phba, sol);
+			break;
+
+		case DRIVERMSG_NOTIFY:
+			SE_DEBUG(DBG_LVL_8,
+				 "Received DRIVERMSG_NOTIFY \n");
+			dmsg = (struct dmsg_cqe *)sol;
+			hwi_complete_drvr_msgs(beiscsi_conn, phba, sol);
+			break;
+
+		case UNSOL_HDR_NOTIFY:
+		case UNSOL_DATA_NOTIFY:
+			SE_DEBUG(DBG_LVL_8,
+				 "Received UNSOL_HDR/DATA_NOTIFY\n");
+			hwi_process_default_pdu_ring(beiscsi_conn, phba,
+					(struct i_t_dpdu_cqe *)sol);
+			break;
+
+		case CXN_INVALIDATE_INDEX_NOTIFY:
+		case CMD_INVALIDATED_NOTIFY:
+		case CXN_INVALIDATE_NOTIFY:
+			SE_DEBUG(DBG_LVL_1,
+				 "Ignoring CQ Error notification for cmd/cxn"
+				 "invalidate\n");
+			break;
+
+		case SOL_CMD_KILLED_DATA_DIGEST_ERR:
+		case CMD_KILLED_INVALID_STATSN_RCVD:
+		case CMD_KILLED_INVALID_R2T_RCVD:
+		case CMD_CXN_KILLED_LUN_INVALID:
+		case CMD_CXN_KILLED_ICD_INVALID:
+		case CMD_CXN_KILLED_ITT_INVALID:
+		case CMD_CXN_KILLED_SEQ_OUTOFORDER:
+		case CMD_CXN_KILLED_INVALID_DATASN_RCVD:


What are these error codes for? How would youhit a lun invalid or itt invalid error?


+			/* Though some of the above code say CMD_CXN, it only
+			 * affects the command in initiator mode.
+			 * */
+
+			SE_DEBUG(DBG_LVL_1,
+			 "CQ Error notification for cmd.. code %d cid 0x%x\n",
+			 sol->dw[offsetof(struct amap_sol_cqe, code) / 32]
+			&  CQE_CODE_MASK,
+			 (sol->dw[offsetof(struct amap_sol_cqe, cid) / 32]
+			&  SOL_CID_MASK));
+			/* Wait for mid-layer timeout to abort and clean */
+			break;
+
+		case UNSOL_DATA_DIGEST_ERROR_NOTIFY:
+			SE_DEBUG(DBG_LVL_1,
+				 "Digest error on def pdu ring, dropping..\n");
+			/* Increment default PDU producer index and post
+			 * it back
+			 * */
+			hwi_flush_default_pdu_buffer(phba, beiscsi_conn,
+						     (struct i_t_dpdu_cqe *)
+						     sol);
+			break;
+
+		case CXN_KILLED_PDU_SIZE_EXCEEDS_DSL:
+		case CXN_KILLED_BURST_LEN_MISMATCH:
+		case CXN_KILLED_AHS_RCVD:
+		case CXN_KILLED_HDR_DIGEST_ERR:
+		case CXN_KILLED_UNKNOWN_HDR:
+		case CXN_KILLED_STALE_ITT_TTT_RCVD:
+		case CXN_KILLED_INVALID_ITT_TTT_RCVD:
+		case CXN_KILLED_TIMED_OUT:
+		case CXN_KILLED_FIN_RCVD:
+		case CXN_KILLED_BAD_UNSOL_PDU_RCVD:
+		case CXN_KILLED_BAD_WRB_INDEX_ERROR:
+		case CXN_KILLED_OVER_RUN_RESIDUAL:
+		case CXN_KILLED_UNDER_RUN_RESIDUAL:


For the resid errors I do not think you want to fire a conn_failure. We can just fail the command and retry it normally.


+		case CXN_KILLED_CMND_DATA_NOT_ON_SAME_CONN:
+
+			SE_DEBUG(DBG_LVL_1,
+			"CQ Error %d, resetting CID 0x%x...\n",
+			sol->dw[offsetof(struct amap_sol_cqe, code) / 32]
+			&  CQE_CODE_MASK,
+			sol->dw[offsetof(struct amap_sol_cqe, cid) / 32]
+			&  CQE_CID_MASK);
+			iscsi_conn_failure((struct iscsi_conn *)beiscsi_conn->
+					   conn, ISCSI_ERR_CONN_FAILED);
+			break;
+
+		case CXN_KILLED_RST_SENT:
+		case CXN_KILLED_RST_RCVD:
+
+			SE_DEBUG(DBG_LVL_1,
+			"CQ Error %d, reset received/sent on"
+			"CID 0x%x...\n",
+			sol->dw[offsetof(struct amap_sol_cqe, code) / 32]
+			&  CQE_CODE_MASK,
+			sol->dw[offsetof(struct amap_sol_cqe, cid) / 32]
+			&  CQE_CID_MASK);
+			iscsi_conn_failure((struct iscsi_conn *)beiscsi_conn->
+					   conn, ISCSI_ERR_CONN_FAILED);
+			break;
+
+		default:
+			SE_DEBUG(DBG_LVL_1,
+			"CQ Error Invalid code= %d received on"
+			"CID 0x%x...\n",
+			sol->dw[offsetof(struct amap_sol_cqe, code) / 32]
+						&  CQE_CODE_MASK,
+			sol->dw[offsetof(struct amap_sol_cqe, cid) / 32]
+						&  CQE_CID_MASK);
+			break;
+		}
+
+		AMAP_SET_BITS(struct amap_sol_cqe, valid, sol, 0);
+		queue_tail_inc(cq);


I think instead of your queue stuff you might be able to use the kfifi stuff instead. Not sure though. I have not looked at the code closely enough yet. Check it out when you get a sec.



+		sol = queue_tail_node(cq);
+		num_processed++;
+	}
+
+	if (num_processed>  0) {
+		tot_nump += num_processed;
+		hwi_ring_cq_db(phba, phwi_context->be_cq.id, num_processed,
+			       1, 0);
+
+	}
+	return tot_nump;
+}
+
+static void beiscsi_find_mem_req(struct beiscsi_hba *phba)
+{
+
+	unsigned int num_cq_pages, num_eq_pages, num_async_pdu_buf_pages,
+	    num_async_pdu_data_pages;
+	unsigned int num_async_pdu_buf_sgl_pages, num_async_pdu_data_sgl_pages;
+
+	num_cq_pages =
+	    PAGES_REQUIRED(phba->params.num_cq_entries * \
+				    sizeof(struct sol_cqe));
+	num_eq_pages =
+	    PAGES_REQUIRED(phba->params.num_eq_entries * \
+			    sizeof(struct be_eq_entry));
+	num_async_pdu_buf_pages =
+	    PAGES_REQUIRED(phba->params.asyncpdus_per_ctrl * \
+			phba->params.defpdu_hdr_sz);
+	num_async_pdu_buf_sgl_pages =
+	    PAGES_REQUIRED(phba->params.asyncpdus_per_ctrl * \
+		       sizeof(struct phys_addr));
+	num_async_pdu_data_pages =
+	    PAGES_REQUIRED(phba->params.asyncpdus_per_ctrl * \
+			phba->params.defpdu_data_sz);
+	num_async_pdu_data_sgl_pages =
+	    PAGES_REQUIRED(phba->params.asyncpdus_per_ctrl * \
+		       sizeof(struct phys_addr));
+
+	phba->params.hwi_ws_sz = sizeof(struct hwi_controller);
+
+	phba->mem_req[ISCSI_MEM_GLOBAL_HEADER] = 2 *
+		sizeof(BE_ISCSI_PDU_HEADER_SIZE);	/* g_header_sz */
+
+	/* hwi_ctxt_sz      */
+	phba->mem_req[HWI_MEM_ADDN_CONTEXT] =
+	    sizeof(struct hwi_context_memory);
+	phba->mem_req[HWI_MEM_MAILBOX] = sizeof(struct mcc_wrb) +
+		sizeof(struct be_mcc_compl);	/* mailbox_sz */
+
+	phba->mem_req[HWI_MEM_CQ] = num_cq_pages * PAGE_SIZE;
+	/* cq_sz            */
+	phba->mem_req[HWI_MEM_EQ] = num_eq_pages * PAGE_SIZE;
+	/* eq_sz            */
+
+	phba->mem_req[HWI_MEM_WRB] = sizeof(struct iscsi_wrb)
+	    * (phba->params.wrbs_per_cxn)
+	    * phba->params.cxns_per_ctrl;	/* queue depth */
+	phba->mem_req[HWI_MEM_WRBH] =
+	    sizeof(struct wrb_handle) * (phba->params.wrbs_per_cxn) *
+	    phba->params.cxns_per_ctrl;
+
+	phba->mem_req[HWI_MEM_SGLH] = sizeof(struct sgl_handle) *
+		phba->params.icds_per_ctrl;
+	phba->mem_req[HWI_MEM_SGE] = sizeof(struct iscsi_sge) *
+	    phba->params.num_sge_per_io * phba->params.icds_per_ctrl;
+	/* sge_sz   */
+
+	phba->mem_req[HWI_MEM_ASYNC_HEADER_BUF] =
+		num_async_pdu_buf_pages * PAGE_SIZE;	/* async_hdr_sz  */
+	phba->mem_req[HWI_MEM_ASYNC_DATA_BUF] =
+		num_async_pdu_data_pages * PAGE_SIZE;	/* def_hdr_sz  */
+	phba->mem_req[HWI_MEM_ASYNC_HEADER_RING] =
+		num_async_pdu_buf_sgl_pages * PAGE_SIZE;
+	/* async_hdr_ring_sz */
+	phba->mem_req[HWI_MEM_ASYNC_DATA_RING] =
+	    num_async_pdu_data_sgl_pages * PAGE_SIZE;
+	/* def_buf_ring_sz  */
+	phba->mem_req[HWI_MEM_ASYNC_HEADER_HANDLE] =
+	    phba->params.asyncpdus_per_ctrl * sizeof(struct async_pdu_handle);
+	/* async_hdr_h_sz   */
+	phba->mem_req[HWI_MEM_ASYNC_DATA_HANDLE] =
+	    phba->params.asyncpdus_per_ctrl * sizeof(struct async_pdu_handle);
+	/* def_buf_h_sz     */
+	phba->mem_req[HWI_MEM_ASYNC_PDU_CONTEXT] =
+	    sizeof(struct hwi_async_pdu_context) +
+	    (phba->params.cxns_per_ctrl * sizeof(struct hwi_async_entry));
+	/* ctxt_sz    */
+}
+
+static int beiscsi_alloc_mem(struct beiscsi_hba *phba)
+{
+	struct be_mem_descriptor *mem_descr;
+	dma_addr_t bus_add;
+	unsigned int num_size, i, j;
+	phba->phwi_ctrlr = kmalloc(phba->params.hwi_ws_sz, GFP_KERNEL);
+	if (!phba->phwi_ctrlr)
+		return -ENOMEM;
+	phba->init_mem =
+	    kzalloc(sizeof(struct be_mem_descriptor) * SE_MEM_MAX, GFP_KERNEL);
+	mem_descr = phba->init_mem;
+
+	for (i = 0; i<  SE_MEM_MAX; i++) {


Maybe you want some sort of pci/dma mem pool here?

Also intead of the alloc_consistent calls maybe you want to be allocting in pages if they allocations are pretty large like this.



+		j = 0;
+
+		num_size = phba->mem_req[i];
+		while (num_size) {
+			if (j>= BEISCSI_MAX_FRAGS_INIT) {
+				SE_DEBUG(DBG_LVL_1,
+				 "Memory Fragment exceeded %d for"
+				 "index=%d. Failing to Load thedriver\n",
+				 BEISCSI_MAX_FRAGS_INIT, i);
+				goto free_mem;
+			}
+
+			if (num_size>= 131072) {
+				mem_descr->mem_array[j].virtual_address =
+				    pci_alloc_consistent(phba->pcidev,
+						 131072,&bus_add);
+				if (!mem_descr->mem_array[j].virtual_address) {
+					SE_DEBUG(DBG_LVL_1, "Memory too"
+					"fragmented to Load the driver");
+					goto free_mem;
+				} else {
+					mem_descr->mem_array[j].bus_address.u.
+					    a64.address = (__u64) bus_add;
+					mem_descr->mem_array[j].size = 131072;
+					memset(mem_descr->mem_array[j].
+					       virtual_address, 0x0, 131072);
+					j++;
+					num_size -= 131072;
+				}
+			} else if (num_size>= 65536) {
+				mem_descr->mem_array[j].virtual_address =
+				    pci_alloc_consistent(phba->pcidev,
+						 num_size,&bus_add);
+				if (!mem_descr->mem_array[j].virtual_address) {
+					goto free_mem;
+				} else {
+					mem_descr->mem_array[j].bus_address.u.
+					    a64.address = (__u64) bus_add;
+					mem_descr->mem_array[j].size = num_size;
+					memset(mem_descr->mem_array[j].
+					       virtual_address, 0x0, num_size);
+					j++;
+					num_size -= num_size;
+				}
+			} else {
+				mem_descr->mem_array[j].virtual_address =
+				    pci_alloc_consistent(phba->pcidev,
+						 num_size,&bus_add);
+				if (!mem_descr->mem_array[j].virtual_address) {
+					SE_DEBUG(DBG_LVL_1,
+					"Memory too fragmented to Load driver");
+					goto free_mem;
+				} else {
+					mem_descr->mem_array[j].bus_address.u.
+					    a64.address = (__u64) bus_add;
+					mem_descr->mem_array[j].size = num_size;
+					memset(mem_descr->mem_array[j].
+					       virtual_address, 0x0, num_size);
+					j++;
+					num_size -= num_size;
+				}
+			}
+		}
+		mem_descr->num_elements = j;
+		mem_descr->size_in_bytes = phba->mem_req[i];
+		SE_DEBUG(DBG_LVL_8,
+		"In alloc_mem, i=%d , mem_descr=%p mem_descr->num_elements=%d"
+		"mem_descr->size_in_bytes =%d \n", i, mem_descr,
+		mem_descr->num_elements, mem_descr->size_in_bytes);
+		mem_descr++;
+	}
+	return 0;
+free_mem:
+	mem_descr->num_elements = j;
+	while ((i) || (j)) {
+		for (j = mem_descr->num_elements; j>  0; j--) {
+			pci_free_consistent(phba->pcidev,
+					    mem_descr->mem_array[j - 1].size,
+					    mem_descr->mem_array[j - 1].
+					    virtual_address,
+					    mem_descr->mem_array[j - 1].
+					    bus_address.u.a64.address);
+
+		}
+		if (i) {
+			i--;
+			mem_descr--;
+		}
+	}
+	kfree(phba->init_mem);
+	kfree(phba->phwi_ctrlr);
+	return -ENOMEM;
+}
+
+static int beiscsi_get_memory(struct beiscsi_hba *phba)
+{
+	beiscsi_find_mem_req(phba);
+	return beiscsi_alloc_mem(phba);
+}
+
+static void iscsi_init_global_templates(struct beiscsi_hba *phba)
+{
+	struct pdu_data_out *pdata_out;
+	struct pdu_nop_out *pnop_out;
+	struct be_mem_descriptor *mem_descr;
+
+	/*
+	 * Patch the opcode field of global template headers with write code.
+	 * The rest  will be patched up by the EP
+	 */
+	mem_descr = phba->init_mem;
+	mem_descr += ISCSI_MEM_GLOBAL_HEADER;
+	pdata_out =
+	    (struct pdu_data_out *)mem_descr->mem_array[0].virtual_address;
+	memset(pdata_out, 0, BE_ISCSI_PDU_HEADER_SIZE);
+
+	AMAP_SET_BITS(struct amap_pdu_data_out, opcode, pdata_out,
+		      IIOC_SCSI_DATA);
+
+	pnop_out =
+	    (struct pdu_nop_out *)((unsigned char *)mem_descr->mem_array[0].
+				   virtual_address + BE_ISCSI_PDU_HEADER_SIZE);
+
+	memset(pnop_out, 0, BE_ISCSI_PDU_HEADER_SIZE);
+	AMAP_SET_BITS(struct amap_pdu_nop_out, ttt, pnop_out, 0xFFFFFFFF);
+	AMAP_SET_BITS(struct amap_pdu_nop_out, f_bit, pnop_out, 1);
+	AMAP_SET_BITS(struct amap_pdu_nop_out, i_bit, pnop_out, 0);
+}
+
+static void beiscsi_init_wrb_handle(struct beiscsi_hba *phba)
+{
+
+	struct be_mem_descriptor *mem_descr_wrbh, *mem_descr_wrb;
+	struct wrb_handle *pwrb_handle;
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_wrb_context *pwrb_context;
+	struct iscsi_wrb *pwrb;
+	unsigned short arr_index;
+	unsigned int num_cxn_wrbh;
+	unsigned int num_cxn_wrb, j, idx, index;
+
+	/* Initiallize IO Handle */
+	mem_descr_wrbh = phba->init_mem;
+	mem_descr_wrbh += HWI_MEM_WRBH;
+
+	mem_descr_wrb = phba->init_mem;
+	mem_descr_wrb += HWI_MEM_WRB;
+
+	idx = 0;
+	pwrb_handle = mem_descr_wrbh->mem_array[idx].virtual_address;
+	num_cxn_wrbh = ((mem_descr_wrbh->mem_array[idx].size) /
+			((sizeof(struct wrb_handle)) *
+			 phba->params.wrbs_per_cxn));
+	phwi_ctrlr = phba->phwi_ctrlr;
+


Are you preallocating memory for all possible connections when the host/hba is setup? That seems like a lot of memory that could be wasted.

Could we allocate this when iscsi_transport->create_conn is called?

+	for (index = 0; index<  phba->params.cxns_per_ctrl * 2; index += 2) {
+		arr_index = 0;
+		pwrb_context =&phwi_ctrlr->wrb_context[index];
+		SE_DEBUG(DBG_LVL_8, "cid=%d pwrb_context=%p \n", index,
+						pwrb_context);
+		pwrb_context->pwrb_handle_base =
+				kzalloc(sizeof(struct wrb_handle *) *
+				phba->params.wrbs_per_cxn, GFP_KERNEL);
+		pwrb_context->pwrb_handle_basestd =
+		    kzalloc(sizeof(struct wrb_handle *) *
+			    phba->params.wrbs_per_cxn, GFP_KERNEL);
+		if (num_cxn_wrbh) {
+			pwrb_context->alloc_index = 0;
+			pwrb_context->wrb_handles_available = 0;
+			for (j = 0; j<  phba->params.wrbs_per_cxn; j++) {
+				pwrb_handle->wrb_index =
+				    (unsigned short)arr_index++;
+				pwrb_context->pwrb_handle_base[j] = pwrb_handle;
+				pwrb_context->pwrb_handle_basestd[j] =
+				    pwrb_handle;
+				pwrb_context->wrb_handles_available++;
+				pwrb_handle++;
+			}
+			pwrb_context->free_index = 0;
+			num_cxn_wrbh--;
+		} else {
+			idx++;
+			pwrb_handle =
+			    mem_descr_wrbh->mem_array[idx].virtual_address;
+			num_cxn_wrbh =
+			    ((mem_descr_wrbh->mem_array[idx].size) /
+			     ((sizeof(struct wrb_handle)) *
+			      phba->params.wrbs_per_cxn));
+			pwrb_context->alloc_index = 0;
+			for (j = 0; j<  phba->params.wrbs_per_cxn; j++) {
+				pwrb_handle->wrb_index =
+				    (unsigned short)arr_index++;
+				pwrb_context->pwrb_handle_base[j] = pwrb_handle;
+				pwrb_context->pwrb_handle_basestd[j] =
+				    pwrb_handle;
+				pwrb_context->wrb_handles_available++;
+				pwrb_handle++;
+			}
+			pwrb_context->free_index = 0;
+			num_cxn_wrbh--;
+		}
+	}
+
+	/* Initiallize pwrb */
+	idx = 0;
+	pwrb = mem_descr_wrb->mem_array[idx].virtual_address;
+	num_cxn_wrb =
+	    ((mem_descr_wrb->mem_array[idx].size) / (sizeof(struct iscsi_wrb)) *
+	     phba->params.wrbs_per_cxn);
+
+	for (index = 0; index<  phba->params.cxns_per_ctrl; index += 2) {
+		pwrb_context =&phwi_ctrlr->wrb_context[index];
+		if (num_cxn_wrb) {
+			for (j = 0; j<  phba->params.wrbs_per_cxn; j++) {
+				pwrb_handle = pwrb_context->pwrb_handle_base[j];
+				pwrb_handle->pwrb = pwrb;
+				pwrb++;
+			}
+			num_cxn_wrb--;
+		} else {
+			idx++;
+			pwrb = mem_descr_wrb->mem_array[idx].virtual_address;
+			num_cxn_wrb =
+			    ((mem_descr_wrb->mem_array[idx].size) /
+			     (sizeof(struct iscsi_wrb)) *
+			     phba->params.wrbs_per_cxn);
+			for (j = 0; j<  phba->params.wrbs_per_cxn; j++) {
+				pwrb_handle = pwrb_context->pwrb_handle_base[j];
+				pwrb_handle->pwrb = pwrb;
+				pwrb++;
+			}
+			num_cxn_wrb--;
+		}
+	}
+}
+
+/* Initializes the async_pdu_context structure. This will setup the
+ *  context to enable posting of buffers later.
+ */
+static void hwi_init_async_pdu_ctx(struct beiscsi_hba *phba)
+{
+	struct hwi_controller *phwi_ctrlr;
+	struct hba_parameters *p =&phba->params;
+	struct hwi_async_pdu_context *pasync_ctx;
+	struct async_pdu_handle *pasync_header_h, *pasync_data_h;
+	unsigned int index;
+	struct be_mem_descriptor *mem_descr;
+	mem_descr = (struct be_mem_descriptor *)phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_PDU_CONTEXT;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	phwi_ctrlr->phwi_ctxt->pasync_ctx = (struct hwi_async_pdu_context *)
+	    mem_descr->mem_array[0].virtual_address;
+	pasync_ctx = phwi_ctrlr->phwi_ctxt->pasync_ctx;
+	memset(pasync_ctx, 0, sizeof(*pasync_ctx));
+
+	pasync_ctx->async_header.num_entries = p->asyncpdus_per_ctrl;
+	pasync_ctx->async_header.buffer_size = p->defpdu_hdr_sz;
+	pasync_ctx->async_data.buffer_size = p->defpdu_data_sz;
+	pasync_ctx->async_data.num_entries = p->asyncpdus_per_ctrl;
+
+	/* The Header */
+	mem_descr = (struct be_mem_descriptor *)phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_HEADER_BUF;
+	if (mem_descr->mem_array[0].virtual_address) {
+		SE_DEBUG(DBG_LVL_8,
+			 "hwi_init_async_pdu_ctx HWI_MEM_ASYNC_HEADER_BUF"
+			 "va=%p \n", mem_descr->mem_array[0].virtual_address);
+	} else {
+		dev_warn(&phba->pcidev->dev, "No Virtual address \n");
+	}
+	pasync_ctx->async_header.va_base =
+	    mem_descr->mem_array[0].virtual_address;
+
+	pasync_ctx->async_header.pa_base.u.a64.address =
+	    mem_descr->mem_array[0].bus_address.u.a64.address;
+
+	mem_descr = (struct be_mem_descriptor *)phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_HEADER_RING;
+	if (mem_descr->mem_array[0].virtual_address) {
+		SE_DEBUG(DBG_LVL_8,
+			 "hwi_init_async_pdu_ctx HWI_MEM_ASYNC_HEADER_RING"
+			 "va=%p \n", mem_descr->mem_array[0].virtual_address);
+	} else {
+		dev_warn(&phba->pcidev->dev, "No Virtual address \n");
+	}
+	pasync_ctx->async_header.ring_base =
+	    mem_descr->mem_array[0].virtual_address;
+
+	mem_descr = (struct be_mem_descriptor *)phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_HEADER_HANDLE;
+	if (mem_descr->mem_array[0].virtual_address) {
+		SE_DEBUG(DBG_LVL_8,
+			 "hwi_init_async_pdu_ctx HWI_MEM_ASYNC_HEADER_HANDLE"
+			 "va=%p \n", mem_descr->mem_array[0].virtual_address);
+	} else {
+		dev_warn(&phba->pcidev->dev, "No Virtual address \n");
+	}
+	pasync_ctx->async_header.handle_base =
+	    mem_descr->mem_array[0].virtual_address;
+	pasync_ctx->async_header.writables = 0;
+	INIT_LIST_HEAD(&pasync_ctx->async_header.free_list);
+
+	/* Now, for Data */
+
+	mem_descr = (struct be_mem_descriptor *)phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_DATA_BUF;
+	if (mem_descr->mem_array[0].virtual_address) {
+		SE_DEBUG(DBG_LVL_8,
+			 "hwi_init_async_pdu_ctx HWI_MEM_ASYNC_DATA_BUF"
+			"va=%p \n", mem_descr->mem_array[0].virtual_address);
+	} else {
+		dev_warn(&phba->pcidev->dev, "No Virtual address \n");
+	}
+	pasync_ctx->async_data.va_base =
+	    mem_descr->mem_array[0].virtual_address;
+	pasync_ctx->async_data.pa_base.u.a64.address =
+	    mem_descr->mem_array[0].bus_address.u.a64.address;
+
+	mem_descr = (struct be_mem_descriptor *)phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_DATA_RING;
+	if (mem_descr->mem_array[0].virtual_address) {
+		SE_DEBUG(DBG_LVL_8,
+			 "hwi_init_async_pdu_ctx HWI_MEM_ASYNC_DATA_RING"
+			 "va=%p \n", mem_descr->mem_array[0].virtual_address);
+	} else {
+		dev_warn(&phba->pcidev->dev, "No Virtual address \n");
+	}
+
+	pasync_ctx->async_data.ring_base =
+	    mem_descr->mem_array[0].virtual_address;
+
+	mem_descr = (struct be_mem_descriptor *)phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_DATA_HANDLE;
+	if (!mem_descr->mem_array[0].virtual_address)
+		dev_warn(&phba->pcidev->dev, "No Virtual address \n");
+
+	pasync_ctx->async_data.handle_base =
+	    mem_descr->mem_array[0].virtual_address;
+	pasync_ctx->async_data.writables = 0;
+	INIT_LIST_HEAD(&pasync_ctx->async_data.free_list);
+
+	pasync_header_h =
+	    (struct async_pdu_handle *)pasync_ctx->async_header.handle_base;
+	pasync_data_h =
+	    (struct async_pdu_handle *)pasync_ctx->async_data.handle_base;
+
+	for (index = 0; index<  p->asyncpdus_per_ctrl; index++) {
+
+		/* initialize the header part
+		 *  setup the free_list
+		 */
+		pasync_header_h->cri = -1;
+		pasync_header_h->index = (char)index;
+		INIT_LIST_HEAD(&pasync_header_h->link);
+		pasync_header_h->pbuffer =
+		   (void *)((unsigned long)(pasync_ctx->async_header.va_base) +
+			     (p->defpdu_hdr_sz * index));
+
+		pasync_header_h->pa.u.a64.address =
+		    pasync_ctx->async_header.pa_base.u.a64.address +
+		    (p->defpdu_hdr_sz * index);
+
+		list_add_tail(&pasync_header_h->link,
+			&pasync_ctx->async_header.free_list);
+		pasync_header_h++;
+		pasync_ctx->async_header.free_entries++;
+		pasync_ctx->async_header.writables++;
+
+		/* initialize the common part
+		 * initialize wait queue
+		 */
+		INIT_LIST_HEAD(&pasync_ctx->async_entry[index].wait_queue.list);
+		/* initialize the busy list */
+		INIT_LIST_HEAD(&pasync_ctx->async_entry[index].
+			       header_busy_list);
+
+		/* initialize the data part
+		 * setup the free_list
+		 */
+		pasync_data_h->cri = -1;
+		pasync_data_h->index = (char)index;
+		INIT_LIST_HEAD(&pasync_data_h->link);
+		pasync_data_h->pbuffer =
+		    (void *)((unsigned long)(pasync_ctx->async_data.va_base) +
+			     (p->defpdu_data_sz * index));
+
+		pasync_data_h->pa.u.a64.address =
+		    pasync_ctx->async_data.pa_base.u.a64.address +
+		    (p->defpdu_data_sz * index);
+
+		list_add_tail(&pasync_data_h->link,
+			&pasync_ctx->async_data.free_list);
+		pasync_data_h++;
+		pasync_ctx->async_data.free_entries++;
+		pasync_ctx->async_data.writables++;
+
+		/*      initialize the busy list        */
+		INIT_LIST_HEAD(&pasync_ctx->async_entry[index].data_busy_list);
+	}
+
+	pasync_ctx->async_header.host_write_ptr = 0;
+	pasync_ctx->async_header.ep_read_ptr = -1;
+	pasync_ctx->async_data.host_write_ptr = 0;
+	pasync_ctx->async_data.ep_read_ptr = -1;
+}
+
+static int
+be_sgl_create_contiguous(void *virtual_address,
+			 u64 physical_address, u32 length,
+			 struct be_dma_mem *sgl)
+{
+	WARN_ON(!virtual_address);
+	WARN_ON(!physical_address);
+	WARN_ON(!length>  0);
+	WARN_ON(!sgl);
+
+	sgl->va = virtual_address;
+	sgl->dma = physical_address;
+	sgl->size = length;
+
+	return 0;
+}
+
+static void be_sgl_destroy_contiguous(struct be_dma_mem *sgl)
+{
+	WARN_ON(!sgl);
+	memset(sgl, 0x0, sizeof(*sgl));
+}
+
+static void hwi_build_be_sgl_arr(struct beiscsi_hba *phba,
+		struct mem_array *pmem, struct be_dma_mem *sgl)
+{
+	struct hwi_controller *phwi_ctrlr;
+
+	WARN_ON(!sgl);
+	WARN_ON(!phba);
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	WARN_ON(!phwi_ctrlr);
+
+	/*  Destroy any existing SGL */
+	if (sgl->va)
+		be_sgl_destroy_contiguous(sgl);
+
+	be_sgl_create_contiguous(pmem->virtual_address,
+				 pmem->bus_address.u.a64.address,
+				 pmem->size, sgl);
+}
+
+static void hwi_build_be_sgl_by_offset(struct beiscsi_hba *phba,
+		struct mem_array *pmem, struct be_dma_mem *sgl)
+{
+	struct hwi_controller *phwi_ctrlr;
+
+	WARN_ON(!sgl);
+	WARN_ON(!phba);
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	WARN_ON(!phwi_ctrlr);
+
+	/* Destroy any existing SGL */
+	if (sgl->va)
+		be_sgl_destroy_contiguous(sgl);
+
+	be_sgl_create_contiguous((unsigned char *)pmem->virtual_address,
+				 pmem->bus_address.u.a64.address,
+				 pmem->size, sgl);
+}
+
+static int be_fill_queue(struct be_queue_info *q,
+		u16 len, u16 entry_size, void *vaddress)
+{
+	struct be_dma_mem *mem =&q->dma_mem;
+
+	memset(q, 0, sizeof(*q));
+	q->len = len;
+	q->entry_size = entry_size;
+	mem->size = len * entry_size;
+	mem->va = vaddress;
+	if (!mem->va)
+		return -ENOMEM;
+	memset(mem->va, 0, mem->size);
+	return 0;
+}
+
+static int
+beiscsi_create_eq(struct beiscsi_hba *phba,
+		struct hwi_context_memory *phwi_context)
+{
+	uint idx;
+	int ret;
+	struct be_queue_info *eq;
+	struct be_dma_mem *mem;
+	struct be_mem_descriptor *mem_descr;
+	void *eq_vaddress;
+
+	idx = 0;
+	eq =&phwi_context->be_eq.q;
+	mem =&eq->dma_mem;
+	mem_descr = phba->init_mem;
+	mem_descr += HWI_MEM_EQ;
+	eq_vaddress = mem_descr->mem_array[idx].virtual_address;
+
+	ret = be_fill_queue(eq, phba->params.num_eq_entries,
+			sizeof(struct be_eq_entry), eq_vaddress);
+	if (ret) {
+		dev_err(&phba->pcidev->dev, "be_fill_queue Failed for EQ \n");
+		return ret;
+	}
+
+	mem->dma = mem_descr->mem_array[idx].bus_address.u.a64.address;
+
+	ret = be_cmd_eq_create(&phba->ctrl, eq, phwi_context->be_eq.cur_eqd);
+	if (ret) {
+		dev_err(&phba->pcidev->dev, "be_cmd_eq_creat Failed for EQ \n");
+		return ret;
+	}
+
+	SE_DEBUG(DBG_LVL_8, "eq id is %d\n", phwi_context->be_eq.q.id);
+	return 0;
+}
+
+static int
+beiscsi_create_cq(struct beiscsi_hba *phba,
+		struct hwi_context_memory *phwi_context)
+{
+	uint idx;
+	int ret;
+	struct be_queue_info *cq, *eq;
+	struct be_dma_mem *mem;
+	struct be_mem_descriptor *mem_descr;
+	void *cq_vaddress;
+
+	idx = 0;
+	cq =&phwi_context->be_cq;
+	eq =&phwi_context->be_eq.q;
+	mem =&cq->dma_mem;
+	mem_descr = phba->init_mem;
+	mem_descr += HWI_MEM_CQ;
+	cq_vaddress = mem_descr->mem_array[idx].virtual_address;
+	ret = be_fill_queue(cq, phba->params.icds_per_ctrl / 2,
+		sizeof(struct sol_cqe), cq_vaddress);
+	if (ret) {
+		dev_err(&phba->pcidev->dev,
+				"be_fill_queue Failed for ISCSI CQ \n");
+		return ret;
+	}
+
+	mem->dma = mem_descr->mem_array[idx].bus_address.u.a64.address;
+	ret = be_cmd_cq_create(&phba->ctrl, cq, eq, false, false, 0);
+	if (ret) {
+		dev_err(&phba->pcidev->dev,
+				"be_cmd_eq_creat Failed for ISCSI CQ \n");
+		return ret;
+	}
+	SE_DEBUG(DBG_LVL_8, "iscsi cq id is %d\n", phwi_context->be_cq.id);
+	SE_DEBUG(DBG_LVL_8, "ISCSI CQ CREATED\n");
+	return 0;
+}
+
+static int
+beiscsi_create_def_hdr(struct beiscsi_hba *phba,
+		struct hwi_context_memory *phwi_context,
+		struct hwi_controller *phwi_ctrlr,
+		unsigned int def_pdu_ring_sz)
+{
+	uint idx;
+	int ret;
+	struct be_queue_info *dq, *cq;
+	struct be_dma_mem *mem;
+	struct be_mem_descriptor *mem_descr;
+	void *dq_vaddress;
+
+	idx = 0;
+	dq =&phwi_context->be_def_hdrq;
+	cq =&phwi_context->be_cq;
+	mem =&dq->dma_mem;
+	mem_descr = phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_HEADER_RING;
+	dq_vaddress = mem_descr->mem_array[idx].virtual_address;
+	ret = be_fill_queue
+	    (dq, mem_descr->mem_array[0].size / sizeof(struct phys_addr),
+				     sizeof(struct phys_addr), dq_vaddress);
+	if (ret) {
+		dev_err(&phba->pcidev->dev,
+				"be_fill_queue Failed for DEF PDU HDR\n");
+		return ret;
+	}
+	mem->dma = mem_descr->mem_array[idx].bus_address.u.a64.address;
+	ret = be_cmd_create_default_pdu_queue(&phba->ctrl, cq, dq,
+		def_pdu_ring_sz, phba->params.defpdu_hdr_sz);
+	if (ret) {
+		dev_err(&phba->pcidev->dev,
+				"be_cmd_eq_creat Failed for DEF PDU HDR\n");
+		return ret;
+	}
+	phwi_ctrlr->default_pdu_hdr.id = phwi_context->be_def_hdrq.id;
+	SE_DEBUG(DBG_LVL_8, "iscsi def pdu id is %d\n",
+		 phwi_context->be_def_hdrq.id);
+	/*  Now that the default PDU ring header has been
+	 *  created, let the EP know  about it... */
+	hwi_post_async_buffers(phba, 1);	/* header */
+	return 0;
+}
+
+static int
+beiscsi_create_def_data(struct beiscsi_hba *phba,
+		struct hwi_context_memory *phwi_context,
+		struct hwi_controller *phwi_ctrlr,
+		unsigned int def_pdu_ring_sz)
+{
+	uint idx;
+	int ret;
+	struct be_queue_info *dataq, *cq;
+	struct be_dma_mem *mem;
+	struct be_mem_descriptor *mem_descr;
+	void *dq_vaddress;
+
+	/* Now, for default Data Ring     */
+	idx = 0;
+	dataq =&phwi_context->be_def_dataq;
+	cq =&phwi_context->be_cq;
+	mem =&dataq->dma_mem;
+	mem_descr = phba->init_mem;
+	mem_descr += HWI_MEM_ASYNC_DATA_RING;
+	dq_vaddress = mem_descr->mem_array[idx].virtual_address;
+	ret = be_fill_queue(dataq, mem_descr->mem_array[0].size /
+		sizeof(struct phys_addr),
+		sizeof(struct phys_addr), dq_vaddress);
+	if (ret) {
+		dev_err(&phba->pcidev->dev,
+				"be_fill_queue Failed for DEF PDU DATA\n");
+		return ret;
+	}
+	mem->dma = mem_descr->mem_array[idx].bus_address.u.a64.address;
+	ret = be_cmd_create_default_pdu_queue
+	    (&phba->ctrl, cq, dataq, def_pdu_ring_sz,
+	     phba->params.defpdu_data_sz);
+	if (ret) {
+		dev_err(&phba->pcidev->dev,
+			 "be_cmd_eq_creat Failed for DEF PDU DATA\n");
+		return ret;
+	}
+	phwi_ctrlr->default_pdu_data.id = phwi_context->be_def_dataq.id;
+	SE_DEBUG(DBG_LVL_8, "iscsi def data id is %d\n",
+		 phwi_context->be_def_dataq.id);
+	hwi_post_async_buffers(phba, 0);	/* data */
+	SE_DEBUG(DBG_LVL_8, "DEFAULT PDU DATA RING CREATED \n");
+	return 0;
+}
+
+static int
+beiscsi_post_pages(struct beiscsi_hba *phba)
+{
+	struct be_mem_descriptor *mem_descr;
+	struct mem_array *pm_arr;
+	unsigned int page_offset, i;
+	struct be_dma_mem sgl;
+	int status;
+
+	mem_descr = phba->init_mem;
+	mem_descr += HWI_MEM_SGE;
+	pm_arr = mem_descr->mem_array;
+
+	page_offset =
+	    (sizeof(struct iscsi_sge) * phba->params.num_sge_per_io *
+	     phba->fw_config.iscsi_icd_start) / PAGE_SIZE;
+	for (i = 0; i<  mem_descr->num_elements; i++) {
+		hwi_build_be_sgl_arr(phba, pm_arr, (struct be_dma_mem *)&sgl);
+		status =
+		    be_cmd_iscsi_post_sgl_pages(&phba->ctrl,
+			(struct be_dma_mem *)&sgl, page_offset,
+				(pm_arr->size / PAGE_SIZE));
+		page_offset += pm_arr->size / PAGE_SIZE;
+		if (status != 0) {
+			dev_err(&phba->pcidev->dev,
+					"post sgl failed.\n");
+			return status;
+		}
+		pm_arr++;
+	}
+	SE_DEBUG(DBG_LVL_8, "POSTED PAGES \n");
+	return 0;
+}
+
+static int
+beiscsi_create_wrb_rings(struct beiscsi_hba *phba,
+		struct hwi_context_memory *phwi_context,
+		struct hwi_controller *phwi_ctrlr)
+{
+	unsigned int wrb_mem_index, offset, size, num_wrb_rings;
+	u64 pa_addr_lo;
+	uint idx, num, i;
+	struct mem_array *pwrb_arr;
+	void *wrb_vaddr;
+	struct be_dma_mem sgl;
+	struct be_mem_descriptor *mem_descr;
+	int status;
+
+	idx = 0;
+	mem_descr = phba->init_mem;
+	mem_descr += HWI_MEM_WRB;
+	pwrb_arr = kmalloc(sizeof(*pwrb_arr)
+			* phba->params.cxns_per_ctrl, GFP_KERNEL);
+	if (!pwrb_arr) {
+		dev_err(&phba->pcidev->dev,
+			"Memory alloc failed in create wrb ring.\n");
+		return -ENOMEM;
+	}
+	wrb_vaddr = mem_descr->mem_array[idx].virtual_address;
+	pa_addr_lo = mem_descr->mem_array[idx].bus_address.u.a64.address;
+	num_wrb_rings = mem_descr->mem_array[idx].size /
+	    (phba->params.wrbs_per_cxn * sizeof(struct iscsi_wrb));
+
+	for (num = 0; num<  phba->params.cxns_per_ctrl; num++) {
+		if (num_wrb_rings) {
+			pwrb_arr[num].virtual_address = wrb_vaddr;
+			pwrb_arr[num].bus_address.u.a64.address	= pa_addr_lo;
+			pwrb_arr[num].size = phba->params.wrbs_per_cxn *
+					    sizeof(struct iscsi_wrb);
+			wrb_vaddr += pwrb_arr[num].size;
+			pa_addr_lo += pwrb_arr[num].size;
+			num_wrb_rings--;
+		} else {
+			idx++;
+			wrb_vaddr = mem_descr->mem_array[idx].virtual_address;
+			pa_addr_lo = mem_descr->mem_array[idx].\
+				     bus_address.u.a64.address;
+			num_wrb_rings = mem_descr->mem_array[idx].size /
+			    (phba->params.wrbs_per_cxn *
+					sizeof(struct iscsi_wrb));
+			pwrb_arr[num].virtual_address = wrb_vaddr;
+			pwrb_arr[num].bus_address.u.a64.address\
+						= pa_addr_lo;
+			pwrb_arr[num].size = phba->params.wrbs_per_cxn *
+						 sizeof(struct iscsi_wrb);
+			wrb_vaddr += pwrb_arr[num].size;
+			pa_addr_lo   += pwrb_arr[num].size;
+			num_wrb_rings--;
+		}
+	}
+	for (i = 0; i<  phba->params.cxns_per_ctrl; i++) {
+


Extra newline

+		wrb_mem_index = 0;
+		offset = 0;
+		size = 0;
+
+		hwi_build_be_sgl_by_offset(phba,&pwrb_arr[i],
+					   (struct be_dma_mem *)&sgl);
+		status = be_cmd_wrbq_create(&phba->ctrl,
+			 (struct be_dma_mem *)&sgl,&phwi_context->be_wrbq[i]);
+		if (status != 0) {
+			dev_err(&phba->pcidev->dev, "wrbq create failed.");
+			return status;
+		}
+		phwi_ctrlr->wrb_context[i].cid = phwi_context->be_wrbq[i].id;
+	}
+	kfree(pwrb_arr);
+	return 0;
+}
+
+static int hwi_init_port(struct beiscsi_hba *phba)
+{
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_context_memory *phwi_context;
+	unsigned int def_pdu_ring_sz;
+	struct be_ctrl_info *ctrl =&phba->ctrl;
+	int status;
+	unsigned int num_sge = 0;
+	unsigned int sebits = 0;
+
+	def_pdu_ring_sz =
+	    phba->params.asyncpdus_per_ctrl * sizeof(struct phys_addr);
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	WARN_ON(!phwi_ctrlr);
+
+	phwi_context = phwi_ctrlr->phwi_ctxt;
+	WARN_ON(!phwi_context);
+	phwi_context->be_eq.max_eqd = 0;
+	phwi_context->be_eq.min_eqd = 0;
+	phwi_context->be_eq.cur_eqd = 64;
+	phwi_context->be_eq.enable_aic = false;
+
+	be_cmd_fw_initiallize(&phba->ctrl);
+	status = beiscsi_create_eq(phba, phwi_context);
+	if (status != 0) {
+		dev_err(&phba->pcidev->dev, "EQ not created \n");
+		goto error;
+	}
+
+	status = mgmt_check_supported_fw(ctrl);
+	if (status != 0) {
+		dev_err(&phba->pcidev->dev, "Unsupported fw version \n");
+		goto error;
+	}
+
+	status = mgmt_get_fw_config(ctrl, phba);
+	if (status != 0) {
+		dev_err(&phba->pcidev->dev, "Error getting fw config\n");
+		goto error;
+	}
+
+	status = beiscsi_create_cq(phba, phwi_context);
+	if (status != 0) {
+		dev_err(&phba->pcidev->dev, "CQ not created \n");
+		goto error;
+	}
+
+	status = beiscsi_create_def_hdr(phba,
+				phwi_context, phwi_ctrlr, def_pdu_ring_sz);
+	if (status != 0) {
+		dev_err(&phba->pcidev->dev, "Default Header not created \n");
+		goto error;
+	}
+
+	status = beiscsi_create_def_data(phba, phwi_context,
+						phwi_ctrlr, def_pdu_ring_sz);
+	if (status != 0) {
+		dev_err(&phba->pcidev->dev, "Default Data not created \n");
+		goto error;
+	}
+
+	status = beiscsi_post_pages(phba);
+	if (status != 0) {
+		dev_err(&phba->pcidev->dev, "Post SGL Pages Failed \n");
+		goto error;
+	}
+
+	status = beiscsi_create_wrb_rings(phba,	phwi_context, phwi_ctrlr);
+	if (status != 0) {
+		dev_err(&phba->pcidev->dev, "WRB Rings not created\n");
+		goto error;
+	}
+	WARN_ON(!phba->params.num_sge_per_io);
+	num_sge = phba->params.num_sge_per_io;
+
+	/* Set the frag_num sebits. Used when driver does
+	 * not use 32 ISCSI_SGEs.
+	 */
+	if (num_sge != 32) {
+		while (num_sge != 0x1) {
+			num_sge = (num_sge>>  1);
+			sebits++;
+		}
+
+		SE_DEBUG(DBG_LVL_8, "Using %d sebits for frag_num "
+				"in ISCSI_SGL_CRA\n", sebits);
+		status = mgmt_set_fragnum_bits(phba, sebits);
+		if (status != 0) {
+			dev_err(&phba->pcidev->dev, "Cannot set fragnum"
+					"sebits\n");
+			goto error;
+		}
+	}
+
+	SE_DEBUG(DBG_LVL_8, "hwi_init_port success\n");
+	return 0;
+
+error:
+	dev_err(&phba->pcidev->dev, "hwi_init_port failed");
+	hwi_cleanup(phba);
+	return 1;
+}
+
+static int hwi_init_controller(struct beiscsi_hba *phba)
+{
+	struct hwi_controller *phwi_ctrlr;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	if (1 == phba->init_mem[HWI_MEM_ADDN_CONTEXT].num_elements) {
+		phwi_ctrlr->phwi_ctxt = (struct hwi_context_memory *)phba->
+		    init_mem[HWI_MEM_ADDN_CONTEXT].mem_array[0].virtual_address;
+		SE_DEBUG(DBG_LVL_8, " phwi_ctrlr->phwi_ctxt=%p \n",
+			 phwi_ctrlr->phwi_ctxt);
+	} else {
+		dev_err(&phba->pcidev->dev,
+			 "HWI_MEM_ADDN_CONTEXT is more than one element."
+			 "Failing to load\n");
+		return -ENOMEM;
+	}
+
+	iscsi_init_global_templates(phba);
+	beiscsi_init_wrb_handle(phba);
+	hwi_init_async_pdu_ctx(phba);
+	if (hwi_init_port(phba) != 0) {
+		dev_err(&phba->pcidev->dev,
+			"hwi_init_controller failed\n");
+		return 1;

Should use a error value like you did above.

If we fail here what cleans up the resources allocated by beiscsi_init_wrb_handle?



+	}
+
+	return 0;
+}
+
+void beiscsi_free_mem(struct beiscsi_hba *phba)
+{
+	struct be_mem_descriptor *mem_descr;
+	int i, j;
+
+	mem_descr = phba->init_mem;
+	i = 0;
+	j = 0;
+	for (i = 0; i<  SE_MEM_MAX; i++) {
+		for (j = mem_descr->num_elements; j>  0; j--) {
+			pci_free_consistent(phba->pcidev,
+			  mem_descr->mem_array[j - 1].size,
+			  mem_descr->mem_array[j - 1].virtual_address,
+			  mem_descr->mem_array[j - 1].bus_address.
+				u.a64.address);
+		}
+		mem_descr++;
+	}
+	kfree(phba->init_mem);
+	kfree(phba->phwi_ctrlr);
+}
+
+static int beiscsi_init_controller(struct beiscsi_hba *phba)
+{
+	int ret = -1;
+
+	ret = beiscsi_get_memory(phba);
+	if (ret<  0) {
+		dev_err(&phba->pcidev->dev, "beiscsi_dev_probe - Failed in"
+			 "beiscsi_alloc_memory \n");
+		return ret;
+	}
+
+	ret = hwi_init_controller(phba);
+	if (ret)
+		goto free_init;
+	SE_DEBUG(DBG_LVL_8, "Return success from beiscsi_init_controller");
+	return 0;
+
+free_init:
+	beiscsi_free_mem(phba);
+	return -ENOMEM;
+}
+
+static int beiscsi_init_sgl_handle(struct beiscsi_hba *phba)
+{
+	struct be_mem_descriptor *mem_descr_sglh, *mem_descr_sg;
+	struct sgl_handle *psgl_handle;
+	void *puse_sgl_handle;
+	struct iscsi_sge *pfrag;
+	unsigned int arr_index, i, idx;
+
+	/* Initiallize IO Handle */
+
+	phba->io_sgl_hndl_avbl = 0;
+	phba->eh_sgl_hndl_avbl = 0;
+	mem_descr_sglh = phba->init_mem;
+	mem_descr_sglh += HWI_MEM_SGLH;
+	if (1 == mem_descr_sglh->num_elements) {
+		phba->io_sgl_hndl_base =
+		    kmalloc(sizeof(struct sgl_handle *) *
+			    phba->params.ios_per_ctrl, GFP_KERNEL);


Check for NULL.

+		phba->eh_sgl_hndl_base =
+		    kmalloc(sizeof(struct sgl_handle *) *
+			    (phba->params.icds_per_ctrl -
+			     phba->params.ios_per_ctrl), GFP_KERNEL);


check for NULL.

+	} else {
+		dev_err(&phba->pcidev->dev,
+			 "HWI_MEM_SGLH is more than one element."
+			 "Failing to load\n");
+		return -ENOMEM;
+	}
+	puse_sgl_handle = phba->io_sgl_hndl_base;
+	memset(phba->io_sgl_hndl_base, 0x0,
+	       (sizeof(struct sgl_handle *) * phba->params.ios_per_ctrl));
+
+	arr_index = 0;
+	idx = 0;
+	while (idx<  mem_descr_sglh->num_elements) {
+		psgl_handle = mem_descr_sglh->mem_array[idx].virtual_address;
+
+		for (i = 0; i<  (mem_descr_sglh->mem_array[idx].size /
+		      sizeof(struct sgl_handle)); i++) {
+			if (arr_index<  phba->params.ios_per_ctrl) {
+				phba->io_sgl_hndl_base[arr_index] =
+				    psgl_handle;
+				phba->io_sgl_hndl_avbl++;
+				arr_index++;
+			} else {
+
+				phba->eh_sgl_hndl_base[arr_index -
+				 phba->params.ios_per_ctrl] = psgl_handle;
+				arr_index++;
+				phba->eh_sgl_hndl_avbl++;
+			}
+			psgl_handle++;
+		}
+		idx++;
+	}
+	SE_DEBUG(DBG_LVL_8,
+		 "phba->io_sgl_hndl_avbl=%d"
+		 "phba->eh_sgl_hndl_avbl=%d \n",
+		 phba->io_sgl_hndl_avbl,
+		 phba->eh_sgl_hndl_avbl);
+	/* Initiallize sge_frags and indices */
+	mem_descr_sg = phba->init_mem;
+	mem_descr_sg += HWI_MEM_SGE;
+	SE_DEBUG(DBG_LVL_8, "\n mem_descr_sg->num_elements=%d \n",
+		 mem_descr_sg->num_elements);
+	arr_index = 0;
+	idx = 0;
+	while (idx<  mem_descr_sg->num_elements) {
+		pfrag = mem_descr_sg->mem_array[idx].virtual_address;
+
+		for (i = 0;
+		     i<  (mem_descr_sg->mem_array[idx].size) /
+		     (sizeof(struct iscsi_sge) * phba->params.num_sge_per_io);
+		     i++) {
+			if (arr_index<  phba->params.ios_per_ctrl) {
+				psgl_handle =
+				    (struct sgl_handle *)phba->
+				    io_sgl_hndl_base[arr_index];
+			} else {
+				psgl_handle =
+				    (struct sgl_handle *)phba->
+				    eh_sgl_hndl_base[arr_index -
+						       phba->params.
+						       ios_per_ctrl];
+			}
+			psgl_handle->pfrag = pfrag;
+			AMAP_SET_BITS(struct amap_iscsi_sge, addr_hi, pfrag, 0);
+			AMAP_SET_BITS(struct amap_iscsi_sge, addr_lo, pfrag, 0);
+			pfrag += phba->params.num_sge_per_io;
+			psgl_handle->sgl_index =
+			    phba->fw_config.iscsi_cid_start + arr_index++;
+		}
+		idx++;
+	}
+	phba->io_sgl_free_index = 0;
+	phba->io_sgl_alloc_index = 0;
+
+	phba->eh_sgl_free_index = 0;
+	phba->eh_sgl_alloc_index = 0;
+	return 0;
+}
+
+static int hba_setup_cid_tbls(struct beiscsi_hba *phba)
+{
+	int i, new_cid;
+
+	phba->cid_array =
+	    kmalloc(sizeof(void *) * phba->params.cxns_per_ctrl * 2,
+			    GFP_KERNEL);

What is the * 2 for?


+	if (!phba->cid_array) {
+		dev_err(&phba->pcidev->dev,
+			 "Failed to allocate memory in hba_setup_cid_tbls \n");
+		return -ENOMEM;
+	}
+	phba->ep_array =
+	    kmalloc(sizeof(unsigned long *) *
+			    phba->params.cxns_per_ctrl * 2, GFP_KERNEL);
+	if (!phba->ep_array) {
+		dev_err(&phba->pcidev->dev,
+			 "Failed to allocate memory in hba_setup_cid_tbls \n");
+		return -ENOMEM;
+	}
+	new_cid = phba->fw_config.iscsi_icd_start;
+	for (i = 0; i<  phba->params.cxns_per_ctrl; i++) {
+		phba->cid_array[i] = new_cid;
+		/*for even cid, cid_array contains 0,2,4, etc at locatio


Why is the cid values like that?


+		 * n 0,1,2,...
+		 */
+		new_cid += 2;
+	}
+	phba->avlbl_cids = phba->params.cxns_per_ctrl;
+	return 0;
+}
+
+void hwi_cleanup(struct beiscsi_hba *phba)
+{
+	struct be_queue_info *q;
+	struct be_ctrl_info *ctrl =&phba->ctrl;
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_context_memory *phwi_context;
+	int i;
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	phwi_context = phwi_ctrlr->phwi_ctxt;
+
+	/*Destroy all wrbq's */
+	for (i = 0; i<  phba->params.cxns_per_ctrl; i++) {
+		q =&phwi_context->be_wrbq[i];
+		if (q->created)
+			be_cmd_q_destroy(ctrl, q, QTYPE_WRBQ);
+	}
+
+	/*Destroy the default PDU queues */
+	q =&phwi_context->be_def_hdrq;
+	if (q->created)
+		be_cmd_q_destroy(ctrl, q, QTYPE_DPDUQ);
+
+	/*Destroy the default PDU data queues */
+	q =&phwi_context->be_def_dataq;
+	if (q->created)
+		be_cmd_q_destroy(ctrl, q, QTYPE_DPDUQ);
+
+	/*Destroy the SGL pages */
+	be_cmd_q_destroy(ctrl, NULL, QTYPE_SGL);
+
+	/*Destroy iscsi CQ */
+	q =&phwi_context->be_cq;
+	if (q->created)
+		be_cmd_q_destroy(ctrl, q, QTYPE_CQ);
+
+	/*Destroy EQ */
+	q =&phwi_context->be_eq.q;
+	if (q->created)
+		be_cmd_q_destroy(ctrl, q, QTYPE_EQ);
+
+}
+
+ void hwi_ring_cq_db(struct beiscsi_hba *phba,
+		unsigned int id, unsigned int num_processed,
+		unsigned char rearm, unsigned char event)
+{
+	u32 val = 0;
+	val |= id&  DB_CQ_RING_ID_MASK;
+	if (rearm)
+		val |= 1<<  DB_CQ_REARM_SHIFT;
+	val |= num_processed<<  DB_CQ_NUM_POPPED_SHIFT;
+	iowrite32(val, phba->db_va + DB_CQ_OFFSET);
+}
+
+void hwi_ring_eq_db(struct beiscsi_hba *phba,
+		unsigned int id, unsigned int clr_interrupt,
+		unsigned int num_processed,
+		unsigned char rearm, unsigned char event)
+{
+	u32 val = 0;
+	val |= id&  DB_EQ_RING_ID_MASK;
+	if (rearm)
+		val |= 1<<  DB_EQ_REARM_SHIFT;
+	if (clr_interrupt)
+		val |= 1<<  DB_EQ_CLR_SHIFT;
+	if (event)
+		val |= 1<<  DB_EQ_EVNT_SHIFT;
+	val |= num_processed<<  DB_EQ_NUM_POPPED_SHIFT;
+	iowrite32(val, phba->db_va + DB_EQ_OFFSET);
+}
+
+unsigned char hwi_enable_intr(struct beiscsi_hba *phba)
+{
+	struct be_ctrl_info *ctrl =&phba->ctrl;
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_context_memory *phwi_context;
+	struct be_queue_info *eq;
+	u8 __iomem *addr;
+	u32 reg;
+	u32 enabled;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	phwi_context = phwi_ctrlr->phwi_ctxt;
+
+	eq =&phwi_context->be_eq.q;
+	addr = (u8 __iomem *) ((u8 __iomem *) ctrl->pcicfg +
+			PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET);
+	reg = ioread32(addr);
+	SE_DEBUG(DBG_LVL_8, "reg =x%08x \n", reg);
+
+	enabled = reg&  MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
+	if (!enabled) {
+		reg |= MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
+		SE_DEBUG(DBG_LVL_8, "reg =x%08x addr=%p \n", reg, addr);
+		iowrite32(reg, addr);
+		SE_DEBUG(DBG_LVL_8, "eq->id=%d \n", eq->id);
+
+		hwi_ring_eq_db(phba, eq->id, 0,	/* clear interrupt */
+			       0, 1, 1);
+
+	} else
+		dev_warn(&phba->pcidev->dev,
+				"In hwi_enable_intr, Not Enabled \n");
+	return true;
+}
+
+void hwi_disable_intr(struct beiscsi_hba *phba)
+{
+	struct be_ctrl_info *ctrl =&phba->ctrl;
+
+	u8 __iomem *addr = ctrl->pcicfg + PCICFG_MEMBAR_CTRL_INT_CTRL_OFFSET;
+	u32 reg = ioread32(addr);
+
+	u32 enabled = reg&  MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
+	if (enabled) {
+		reg&= ~MEMBAR_CTRL_INT_CTRL_HOSTINTR_MASK;
+		iowrite32(reg, addr);
+	} else
+		dev_warn(&phba->pcidev->dev,
+				"In hwi_disable_intr, Already Disabled \n");
+
+}
+
+int beiscsi_init_port(struct beiscsi_hba *phba)
+{
+	int ret = -1;
+
+	ret = beiscsi_init_controller(phba);
+	if (ret<  0) {
+		dev_err(&phba->pcidev->dev,
+			 "beiscsi_dev_probe - Failed in"
+			 "beiscsi_init_controller \n");
+		return ret;
+	}
+	ret = beiscsi_init_sgl_handle(phba);
+	if (ret<  0) {
+		dev_err(&phba->pcidev->dev,
+			 "beiscsi_dev_probe - Failed in"
+			 "beiscsi_init_sgl_handle \n");
+		goto do_cleanup_ctrlr;
+	}
+
+	if (hba_setup_cid_tbls(phba)) {


If this fails where do we free the mem allocated by bsiscsi_init_sgl_handle? Where do we free that in general in the normal host removale path?


Where do we free the mem allocated by hba_setup_cid_tbls in the normal host removal path too?


+		dev_err(&phba->pcidev->dev,
+			"Failed in hba_setup_cid_tbls\n");
+		goto do_cleanup_ctrlr;
+	}
+
+	return ret;
+
+do_cleanup_ctrlr:
+	hwi_cleanup(phba);
+
+	return ret;
+}
+
+static void hwi_purge_eq(struct beiscsi_hba *phba)
+{
+	struct hwi_controller *phwi_ctrlr;
+	struct hwi_context_memory *phwi_context;
+	struct be_queue_info *eq;
+	struct be_eq_entry *eqe = NULL;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	phwi_context = phwi_ctrlr->phwi_ctxt;
+	eq =&phwi_context->be_eq.q;
+	eqe = queue_tail_node(eq);
+
+	while (eqe->dw[offsetof(struct amap_eq_entry, valid) / 32]
+						&  EQE_VALID_MASK) {
+		AMAP_SET_BITS(struct amap_eq_entry, valid, eqe, 0);
+		queue_tail_inc(eq);
+		eqe = queue_tail_node(eq);
+	}
+}
+
+void beiscsi_clean_port(struct beiscsi_hba *phba)
+{
+	struct hwi_controller *phwi_ctrlr;
+	unsigned char mgmt_status;
+
+	/* initiate cleanup of EP firmware */
+	mgmt_status = mgmt_epfw_cleanup(phba, CMD_CONNECTION_CHUTE_0);
+
+	/* destroy all rings */
+	hwi_cleanup(phba);
+
+	/* purge all EQ entries that may have been left out
+	 * This is to workaround a problem we've seen occasionally where
+	 * we exit from be_stop_controller and we get an interrupt
+	 * with EQ entry bit set
+	 */
+	hwi_purge_eq(phba);
+
+	/* destroy SA device object  */
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+}
+
+void
+beiscsi_offload_connection(struct beiscsi_conn *beiscsi_conn,
+			   struct beiscsi_offload_params *params)
+{
+	struct wrb_handle *pwrb_handle;
+	struct iscsi_target_context_update_wrb *pwrb = NULL;
+	struct be_mem_descriptor *mem_descr;
+	struct beiscsi_hba *phba = beiscsi_conn->phba;
+	u32 doorbell = 0;
+
+	do {
+		pwrb_handle = alloc_wrb_handle(phba,
+				beiscsi_conn->beiscsi_conn_cid);
+	} while (!pwrb_handle);


Do we hit this loop if we try to create more sessions than phba->params.cxns_per_ctrl?

I think we would just want to fail here. If we loop like this then we can block up other sessions.


+
+	pwrb = (struct iscsi_target_context_update_wrb *)pwrb_handle->pwrb;
+	memset(pwrb, 0x0, sizeof(*pwrb));
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb,
+		      max_burst_length, pwrb, params->dw[offsetof
+			 (struct amap_beiscsi_offload_params,
+			  max_burst_length) / 32]);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb,
+	      max_send_data_segment_length, pwrb,
+	      params->dw[offsetof(struct amap_beiscsi_offload_params,
+			  max_send_data_segment_length) / 32]);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb,
+		first_burst_length,
+		pwrb, params->dw[offsetof(struct amap_beiscsi_offload_params,
+			  first_burst_length) / 32]);
+
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, erl, pwrb,
+		(params->dw[offsetof(struct amap_beiscsi_offload_params, erl)
+						/ 32]&  OFFLD_PARAMS_ERL));
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, dde, pwrb,
+		(params->
+		dw[offsetof(struct amap_beiscsi_offload_params, dde) / 32]
+		&  OFFLD_PARAMS_DDE)>>  2);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, hde, pwrb,
+		(params->
+		dw[offsetof(struct amap_beiscsi_offload_params, hde) / 32]
+		&  OFFLD_PARAMS_HDE)>>  3);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, ir2t, pwrb,
+		(params->
+		dw[offsetof(struct amap_beiscsi_offload_params, ir2t) / 32]
+		&  OFFLD_PARAMS_IR2T)>>  4);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, imd, pwrb,
+		(params->
+		dw[offsetof(struct amap_beiscsi_offload_params, imd) / 32]
+		&  OFFLD_PARAMS_IMD)>>  5);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, stat_sn,
+		pwrb, (params->dw[offsetof(struct amap_beiscsi_offload_params,
+		exp_statsn) / 32] + 1));
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, type, pwrb,
+		      0x7);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, wrb_idx,
+		      pwrb, pwrb_handle->wrb_index);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, ptr2nextwrb,
+		      pwrb, pwrb_handle->nxt_wrb_index);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb,
+			session_state, pwrb, 0);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, compltonack,
+		      pwrb, 1);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, notpredblq,
+		      pwrb, 0);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb, mode, pwrb,
+		      0);
+
+	mem_descr = phba->init_mem;
+	mem_descr += ISCSI_MEM_GLOBAL_HEADER;
+
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb,
+			pad_buffer_addr_hi, pwrb,
+		      mem_descr->mem_array[0].bus_address.u.a32.address_hi);
+	AMAP_SET_BITS(struct amap_iscsi_target_context_update_wrb,
+			pad_buffer_addr_lo, pwrb,
+		      mem_descr->mem_array[0].bus_address.u.a32.address_lo);
+
+	be_dws_le_to_cpu(pwrb, sizeof(struct iscsi_target_context_update_wrb));
+
+	doorbell |= beiscsi_conn->beiscsi_conn_cid&  DB_WRB_POST_CID_MASK;
+	doorbell |= (pwrb_handle->wrb_index&  DB_DEF_PDU_WRB_INDEX_MASK)<<
+					DB_DEF_PDU_WRB_INDEX_SHIFT;
+	doorbell |= 1<<  DB_DEF_PDU_NUM_POSTED_SHIFT;
+
+	iowrite32(doorbell, phba->db_va + DB_TXULP0_OFFSET);
+	return;
+}
+
+/*
+ * beiscsi_task_xmit
+ * task: iscsi_command_task

Kind of a lame comment. Could drop it or could use docbook here and add some info if needed.



+ */
+int beiscsi_task_xmit(struct iscsi_task *task)
+{
+	struct iscsi_conn *conn = task->conn;
+	struct beiscsi_io_task *io_task = task->dd_data;
+	struct iscsi_task *old_task = NULL;
+	struct scsi_cmnd *sc = task->sc;
+	struct beiscsi_conn *beiscsi_conn = conn->dd_data;
+	struct scatterlist *sg;
+	unsigned int num_sg, writedir = 0, xferlen = 0;
+
+	SE_DEBUG(DBG_LVL_4,
+	 "\n cid=%d In beiscsi_task_xmit task=%p conn=%p \t"
+		 "beiscsi_conn=%p \n", beiscsi_conn->beiscsi_conn_cid,
+				task, conn, beiscsi_conn);
+	io_task->conn = beiscsi_conn;
+	if (!sc) {
+		if (ISCSI_OP_SCSI_TMFUNC ==
+		    (task->hdr->opcode&  ISCSI_OPCODE_MASK)) {
+			old_task = (struct iscsi_task *)iscsi_itt_to_task(
+						conn, ((struct iscsi_tm *)
+						task->hdr)->rtt);
+			io_task->old_task = old_task->dd_data;
+		}
+
+		return beiscsi_mtask(task);
+	}
+
+	io_task->scsi_cmnd = sc;
+	num_sg = scsi_dma_map(sc);

check for failure


+	SE_DEBUG(DBG_LVL_4, "xferlen=0x%08x scmd=%p num_sg=%d sernum=%lu\n",
+		  (scsi_bufflen(sc)), sc, num_sg, sc->serial_number);
+	xferlen = scsi_bufflen(sc);
+	sg = (struct scatterlist *)sc->sdb.table.sgl;



Use scsi_sglist


+	if (sc->sc_data_direction == DMA_TO_DEVICE) {
+		writedir = 1;
+		SE_DEBUG(DBG_LVL_4, "task->imm_count=0x%08x \n",
+			 task->imm_count);
+	} else {
+		writedir = 0;
+	}
+
+	return beiscsi_iotask(task, sg, num_sg, xferlen, writedir);
+}
+
+int beiscsi_iotask(struct iscsi_task *task, struct scatterlist *sg,
+		   unsigned int num_sg, unsigned int xferlen,
+		   unsigned int writedir)
+{
+
+	struct beiscsi_io_task *io_task = task->dd_data;
+	struct iscsi_conn *conn = task->conn;
+	struct beiscsi_conn *beiscsi_conn = conn->dd_data;
+	struct beiscsi_hba *phba = beiscsi_conn->phba;
+	struct iscsi_cmd *hdr;
+	struct iscsi_wrb *pwrb = NULL;
+	unsigned int doorbell = 0;
+	io_task->pwrb_handle =
+	    alloc_wrb_handle(phba, beiscsi_conn->beiscsi_conn_cid);

add a newline here

+	spin_lock(&phba->sgl_lock);
+	io_task->psgl_handle = alloc_io_sgl_handle(phba);
+	spin_unlock(&phba->sgl_lock);
+
+	if (!io_task->pwrb_handle || !io_task->psgl_handle) {
+		/* We are in session-lock, so can't loop here untill
+		 * free wrb is available
+		 * have to free sgl-icd index as well
+		 */
+		SE_DEBUG(DBG_LVL_1,
+			 "Alloc of WRB/SGL_ICD Failed \n");
+		return -EAGAIN;
+	}
+
+	io_task->pwrb_handle->pio_handle = task;
+	pwrb = io_task->pwrb_handle->pwrb;
+
+	hdr = (struct iscsi_cmd *)task->hdr;
+	/* Oh, we use a different itt for fw */


I think in the iscsi_transport->alloc_pdu callout you want to set the hdr->itt to whatever you want.

Then implement a parse_pdu_itt callback so that libiscsi can go from hdr->itt to the iscsi_task->itt.

Check out cxgb3i for an example.


+	io_task->fw_itt = (itt_t)
+	    cpu_to_be32((uint)((io_task->pwrb_handle->wrb_index)<<  16) |
+			(uint)(io_task->psgl_handle->sgl_index));
+	io_task->itt = io_task->cmd_bhs.iscsi_hdr.itt;
+
+	io_task->cmd_bhs.iscsi_hdr.itt = io_task->fw_itt;
+	io_task->cmd_bhs.iscsi_hdr.exp_statsn = 0;
+
+	io_task->bhs_len = sizeof(struct be_cmd_bhs);
+	io_task->bhs_pa.u.a64.address = virt_to_bus(&io_task->cmd_bhs);
+
+	if (writedir) {
+		/* Setting up data_pdu */
+		SE_DEBUG(DBG_LVL_4, " WRITE Command \t");
+		memset(&io_task->cmd_bhs.iscsi_data_pdu, 0, 48);
+		AMAP_SET_BITS(struct amap_pdu_data_out, itt,
+			&io_task->cmd_bhs.iscsi_data_pdu,
+			      (unsigned int)io_task->fw_itt);
+		AMAP_SET_BITS(struct amap_pdu_data_out, opcode,
+			&io_task->cmd_bhs.iscsi_data_pdu,
+			      ISCSI_OPCODE_SCSI_DATA_OUT);
+		AMAP_SET_BITS(struct amap_pdu_data_out, final_bit,
+			&io_task->cmd_bhs.iscsi_data_pdu, 1);
+		AMAP_SET_BITS(struct amap_iscsi_wrb, type, pwrb, INI_WR_CMD);
+		AMAP_SET_BITS(struct amap_iscsi_wrb, dsp, pwrb, 1);
+
+		/* for writes, use r2t_buf_offset field to communicate number
+		 * of sg  elements in order to turn on the EP firmware's
+		 * fast path
+		 * */
+	} else {
+		SE_DEBUG(DBG_LVL_4, "READ Command \t");
+		AMAP_SET_BITS(struct amap_iscsi_wrb, type, pwrb, INI_RD_CMD);
+		AMAP_SET_BITS(struct amap_iscsi_wrb, dsp, pwrb, 0);
+	}
+	memcpy(&io_task->cmd_bhs.iscsi_data_pdu.
+	       dw[offsetof(struct amap_pdu_data_out, lun) / 32],
+	       io_task->cmd_bhs.iscsi_hdr.lun, sizeof(struct scsi_lun));
+
+	if ((hdr->opcode&  ISCSI_OPCODE_MASK) == 0x1) {

Using ISCSI_OP_SCSI_CMD instead of 0x1 is more clear.


+		io_task->resp_len = hdr->data_length;
+		if ((hdr->cdb[0] == 0x12) || (hdr->cdb[0] == 0xa0)


For the scsi cmds using the scsi opcode in scsi.h is also nicer I think.


+		    || (hdr->cdb[0] == 0x25) || (hdr->cdb[0] == 0x1a)
+		    || (hdr->cdb[0] == 0x28)) {
+			io_task->resp_opcode = 0x25;
+		} else if ((hdr->cdb[0] == 0x00) || (hdr->cdb[0] == 0x2a)
+			|| (hdr->cdb[0] == 0x35)) {
+			io_task->resp_opcode = 0x21;
+		} else{
+			SE_DEBUG(DBG_LVL_1, "cdb[0] == 0x%x NOT HANDLED\n",
+				 hdr->cdb[0]);
+			io_task->resp_opcode = 0x21;
+		}


Faking this does not look so nice. Maybe we want to modify iscsi_complete_pdu so you can just complete the scsi_cmnd/iscsi_task? I am not sure. I will look into it.

For login and logout you get the pdu, right?


+	} else {
+		if ((hdr->opcode&  ISCSI_OPCODE_MASK) == 0x05) {


You should never get a data_out, because iscsi_queuecommand just send you the iscsi scsi cmd and the card handles data_outs/data_ins/r2ts right?


+			io_task->resp_opcode = 0x21;
+		} else {
+			io_task->resp_opcode = 0x21;
+			SE_DEBUG(DBG_LVL_1,
+				 "Not yet supported hdr->opcode=0x%x\n",
+				 (hdr->opcode&  ISCSI_OPCODE_MASK));
+		}
+	}
+
+	/* 14 bits of LUN info, from the first level in bigendian format */
+	AMAP_SET_BITS(struct amap_iscsi_wrb, lun, pwrb,
+		      cpu_to_be16((unsigned short)io_task->cmd_bhs.iscsi_hdr.
+				  lun[0]));
+	AMAP_SET_BITS(struct amap_iscsi_wrb, r2t_exp_dtl, pwrb, xferlen);
+	AMAP_SET_BITS(struct amap_iscsi_wrb, wrb_idx, pwrb,
+		      io_task->pwrb_handle->wrb_index);
+	AMAP_SET_BITS(struct amap_iscsi_wrb, cmdsn_itt, pwrb,
+		      be32_to_cpu(task->cmdsn));
+	AMAP_SET_BITS(struct amap_iscsi_wrb, sgl_icd_idx, pwrb,
+		      io_task->psgl_handle->sgl_index);
+
+	hwi_write_sgl(pwrb, sg, num_sg, io_task);
+
+	AMAP_SET_BITS(struct amap_iscsi_wrb, ptr2nextwrb, pwrb,
+		      io_task->pwrb_handle->nxt_wrb_index);
+	be_dws_le_to_cpu(pwrb, sizeof(struct iscsi_wrb));
+
+	doorbell |= beiscsi_conn->beiscsi_conn_cid&  DB_WRB_POST_CID_MASK;
+	doorbell |= (io_task->pwrb_handle->wrb_index&
+		     DB_DEF_PDU_WRB_INDEX_MASK)<<  DB_DEF_PDU_WRB_INDEX_SHIFT;
+	doorbell |= 1<<  DB_DEF_PDU_NUM_POSTED_SHIFT;
+
+	iowrite32(doorbell, phba->db_va + DB_TXULP0_OFFSET);
+	return 0;
+}
+
+int beiscsi_mtask(struct iscsi_task *task)
+{
+	struct beiscsi_io_task *io_task = task->dd_data;
+	struct iscsi_conn *conn = task->conn;
+	struct beiscsi_conn *beiscsi_conn = conn->dd_data;
+	struct beiscsi_hba *phba = beiscsi_conn->phba;
+	struct iscsi_wrb *pwrb = NULL;
+	unsigned int doorbell = 0;
+
+	io_task->scsi_cmnd = NULL;
+	io_task->conn = beiscsi_conn;
+	io_task->pwrb_handle =
+	    alloc_wrb_handle(phba, beiscsi_conn->beiscsi_conn_cid);
+	if (!io_task->pwrb_handle) {
+		/* We are in session-lock, so can't loop here untill
+		 * free wrb is available
+		 */
+		SE_DEBUG(DBG_LVL_1, "No WRB Available \n");
+		return -EAGAIN;
+	}
+	io_task->pwrb_handle->pio_handle = task;
+	pwrb = io_task->pwrb_handle->pwrb;
+
+	if ((task->hdr->opcode&  ISCSI_OPCODE_MASK) == ISCSI_OP_LOGIN) {
+		beiscsi_conn->task = task;
+		if (!beiscsi_conn->login_in_progress) {
+
+			do {
+				io_task->psgl_handle = (struct sgl_handle *)
+						alloc_eh_sgl_handle(phba);
+			} while (!io_task->psgl_handle);
+
+			beiscsi_conn->login_in_progress = 1;
+			beiscsi_conn->plogin_sgl_handle = io_task->psgl_handle;
+		} else {
+			io_task->psgl_handle = beiscsi_conn->plogin_sgl_handle;
+		}
+	} else {
+		beiscsi_conn->task = task;
+
+		do {
+			io_task->psgl_handle =
+				(struct sgl_handle *)alloc_eh_sgl_handle(phba);
+		} while (!io_task->psgl_handle);


How often and how likely is this going to fail? It looks like it could become some crazy loop that takes over the processor and hangs the box.

You want to set things up so we can always do least one TMF per hba.

And you want to be able to handle a nop from the initiator and at least one nop from the target at the same time.



+	}
+
+	/* Oh, we use a different itt for fw */
+	io_task->fw_itt = (itt_t) cpu_to_be32(((uint)(io_task->pwrb_handle->
+			 wrb_index)<<  16) | (uint)(io_task->psgl_handle->
+				sgl_index));

What I suggest for the scsi io path should work here too.


+	io_task->itt = io_task->cmd_bhs.iscsi_hdr.itt;
+	io_task->cmd_bhs.iscsi_hdr.itt = io_task->fw_itt;
+	AMAP_SET_BITS(struct amap_iscsi_wrb, cmdsn_itt, pwrb,
+		      be32_to_cpu(task->cmdsn));
+	AMAP_SET_BITS(struct amap_iscsi_wrb, wrb_idx, pwrb,
+		      io_task->pwrb_handle->wrb_index);
+	AMAP_SET_BITS(struct amap_iscsi_wrb, sgl_icd_idx, pwrb,
+		      io_task->psgl_handle->sgl_index);
+
+	switch (task->hdr->opcode&  ISCSI_OPCODE_MASK) {
+	case ISCSI_OP_LOGIN:
+		AMAP_SET_BITS(struct amap_iscsi_wrb, type, pwrb, TGT_DM_CMD);
+		AMAP_SET_BITS(struct amap_iscsi_wrb, dmsg, pwrb, 0);
+		AMAP_SET_BITS(struct amap_iscsi_wrb, cmdsn_itt, pwrb, 1);
+		hwi_write_buffer(pwrb, task);
+		break;
+	case ISCSI_OP_NOOP_OUT:
+		io_task->resp_opcode = 0x20;
+		AMAP_SET_BITS(struct amap_iscsi_wrb, type, pwrb, INI_RD_CMD);
+		hwi_write_buffer(pwrb, task);
+		break;
+	case ISCSI_OP_SCSI_TMFUNC:
+		{

you do not need these brackets do you?

+			if (!io_task->old_task || !io_task->old_task->scsi_cmnd)
+				return 0;
+			mgmt_invalidate_icds(phba,
+					     io_task->old_task->psgl_handle->
+					     sgl_index,
+					     beiscsi_conn->beiscsi_conn_cid);
+			((struct iscsi_tm *)task->hdr)->rtt =
+				io_task->old_task->fw_itt;	/* Not storing
+								 * original rtt
+								 * here
+								 */
+		}
+		AMAP_SET_BITS(struct amap_iscsi_wrb, type, pwrb, INI_TMF_CMD);
+		AMAP_SET_BITS(struct amap_iscsi_wrb, dmsg, pwrb, 0);
+		hwi_write_buffer(pwrb, task);
+		break;
+	case ISCSI_OP_LOGOUT:
+		AMAP_SET_BITS(struct amap_iscsi_wrb, dmsg, pwrb, 0);
+		AMAP_SET_BITS(struct amap_iscsi_wrb, type, pwrb,
+				HWH_TYPE_LOGOUT);
+		hwi_write_buffer(pwrb, task);
+		break;
+
+	default:
+		SE_DEBUG(DBG_LVL_1, "opcode =%d Not supported \n",
+			 task->hdr->opcode&  ISCSI_OPCODE_MASK);
+		return -EINVAL;
+	}
+
+	AMAP_SET_BITS(struct amap_iscsi_wrb, r2t_exp_dtl, pwrb,
+		      be32_to_cpu(task->data_count));
+	AMAP_SET_BITS(struct amap_iscsi_wrb, ptr2nextwrb, pwrb,
+		      io_task->pwrb_handle->nxt_wrb_index);
+	be_dws_le_to_cpu(pwrb, sizeof(struct iscsi_wrb));
+
+	doorbell |= beiscsi_conn->beiscsi_conn_cid&  DB_WRB_POST_CID_MASK;
+	doorbell |= (io_task->pwrb_handle->wrb_index&
+		     DB_DEF_PDU_WRB_INDEX_MASK)<<  DB_DEF_PDU_WRB_INDEX_SHIFT;
+	doorbell |= 1<<  DB_DEF_PDU_NUM_POSTED_SHIFT;
+	iowrite32(doorbell, phba->db_va + DB_TXULP0_OFFSET);
+
+	return 0;
+}
+
+/* pdu processing */
+
+static void
+be_complete_login(struct beiscsi_conn *beiscsi_conn,
+		  struct iscsi_task *task, struct iscsi_login_rsp *ppdu,
+		  char *pbuffer, unsigned int buf_len)
+{
+	struct beiscsi_io_task *io_task = task->dd_data;
+
+	ppdu->itt = io_task->itt;
+}
+
+/* async pdus include
+ * a. unsolicited NOP-In (target initiated NOP-In)
+ * b. Async Messages
+ * c. Reject PDU
+ * d. Login response
+ * These headers arrive unprocessed by the EP firmware and iSCSI layer
+ * process them
+ */
+unsigned int
+beiscsi_process_async_pdu(struct beiscsi_conn *beiscsi_conn,
+			  struct beiscsi_hba *phba,
+			  unsigned short cid,
+			  struct pdu_base *ppdu,
+			  unsigned long pdu_len,
+			  void *pbuffer, unsigned long buf_len)
+{
+	struct iscsi_task *task;
+	struct beiscsi_io_task *io_task;
+	struct iscsi_conn *conn = beiscsi_conn->conn;
+	struct iscsi_session *session = conn->session;
+	unsigned int status = 0;
+
+	switch (ppdu->
+		dw[offsetof(struct amap_pdu_base, opcode) / 32]
+					&  PDUBASE_OPCODE_MASK) {
+	case ISCSI_OP_NOOP_IN:
+		pbuffer = NULL;
+		buf_len = 0;
+		break;
+
+	case ISCSI_OP_ASYNC_EVENT:
+		break;
+
+	case ISCSI_OP_REJECT:
+		WARN_ON(!pbuffer);
+		WARN_ON(!(buf_len == 48));
+		iscsi_suspend_tx(conn);


Why are you doing this? For the ISCSI_OP_ASYNC_EVENT are you just passing the pdu and data upwards to libiscsi? If so you can just do that here too.

The function can sleep so you cannot call it from the blk napi callout.


+		SE_DEBUG(DBG_LVL_1, "In ISCSI_OP_REJECT\n");
+		break;
+
+	case ISCSI_OP_LOGIN_RSP:
+		/* saved login request task is used */
+		task = beiscsi_conn->task;
+		if (!task) {
+			SE_DEBUG(DBG_LVL_1, "\n No Task\n");
+			status = 1;
+			return status;
+		}
+
+		io_task = task->dd_data;
+		/* make sure that we got the login response for
+		 * the login request received on this connection
+		 */
+		if (io_task->fw_itt != ((struct iscsi_login_rsp *)ppdu)->itt) {
+			SE_DEBUG(DBG_LVL_8,
+					"Error: itt in login request 0x%x ,"
+					"itt in login response 0x%x\n",
+					io_task->fw_itt,
+					((struct iscsi_login_rsp *)ppdu)->itt);
+		}
+		be_complete_login(beiscsi_conn, task,
+				  (struct iscsi_login_rsp *)ppdu,
+				  (char *)pbuffer, buf_len);
+		break;
+
+	default:
+		dev_warn(&phba->pcidev->dev,
+			"Unrecognized opcode 0x%x in async msg \n",
+			 (ppdu->
+			  dw[offsetof(struct amap_pdu_base, opcode) / 32]
+						&  PDUBASE_OPCODE_MASK));
+		status = 1;
+		return status;
+	}
+
+	/* as per open iscsi API, __iscsi_complete_pdu()
+	 * should be invoked under iscsi_session->lock
+	 */
+	SE_DEBUG(DBG_LVL_8, "In %s , grabbing session->lock\n", __func__);
+	spin_lock_bh(&session->lock);
+	__iscsi_complete_pdu(conn, (struct iscsi_hdr *)ppdu, pbuffer, buf_len);
+	spin_unlock_bh(&session->lock);
+
+	return 0;
+}
+
+static void
+be_complete_io(struct beiscsi_conn *beiscsi_conn,
+	       struct iscsi_task *task, struct sol_cqe *psol)
+{
+	struct iscsi_cmd_rsp *hdr;
+	struct beiscsi_io_task *io_task = task->dd_data;
+	struct be_status_bhs *sts_bhs =
+	    (struct be_status_bhs *)&io_task->cmd_bhs;
+	struct iscsi_conn *conn = beiscsi_conn->conn;
+	struct beiscsi_hba *phba = beiscsi_conn->phba;
+	unsigned int sense_len = 0;
+	unsigned int urlen;
+	int ret;
+	hdr = (struct iscsi_cmd_rsp *)task->hdr;
+	hdr->opcode = io_task->resp_opcode;
+	hdr->exp_cmdsn = be32_to_cpu(psol->
+			dw[offsetof(struct amap_sol_cqe, i_exp_cmd_sn) / 32]
+			&  SOL_EXP_CMD_SN_MASK);
+	hdr->max_cmdsn = be32_to_cpu((psol->
+			dw[offsetof(struct amap_sol_cqe, i_exp_cmd_sn) / 32]
+			&  SOL_EXP_CMD_SN_MASK) +
+			((psol->dw[offsetof(struct amap_sol_cqe, i_cmd_wnd)
+				/ 32]&  SOL_CMD_WND_MASK)>>  24) - 1);
+	hdr->response = ((psol->dw[offsetof(struct amap_sol_cqe, i_resp) / 32]
+						&  SOL_RESP_MASK)>>  16);
+	hdr->cmd_status = ((psol->dw[offsetof(struct amap_sol_cqe, i_sts) / 32]
+						&  SOL_STS_MASK)>>  8);
+	hdr->flags = ((psol->dw[offsetof(struct amap_sol_cqe, i_flags) / 32]
+					&  SOL_FLAGS_MASK)>>  24) | 0x80;
+	hdr->itt = io_task->itt;
+	hdr->rsvd1 = 0xFFFFFFFF;
+
+	if (hdr->flags&  0x06) {
+		hdr->residual_count = (psol->
+				dw[offsetof(struct amap_sol_cqe, i_res_cnt) /
+				32]&  SOL_RES_CNT_MASK);
+		if (hdr->flags&  0x02) {
+			urlen = be32_to_cpu(io_task->resp_len);
+			urlen = urlen - hdr->residual_count;
+			io_task->resp_len = cpu_to_be32(urlen);
+		} else
+			SE_DEBUG(DBG_LVL_1, "\nOVERRUN\n");
+
+		hdr->bi_residual_count = 0x00;
+		hdr->residual_count =  cpu_to_be32(psol->
+				dw[offsetof(struct amap_sol_cqe, i_res_cnt)
+					/ 32]&  SOL_RES_CNT_MASK);
+	} else
+		hdr->residual_count = 0;
+	if (io_task->resp_len&  0xffffff00) {
+		hdr->dlength[2] = (io_task->resp_len&  0xff000000)>>  24;
+		hdr->dlength[1] = (io_task->resp_len&  0x00ff0000)>>  16;
+		hdr->dlength[0] = (io_task->resp_len&  0x0000ff00)>>  8;
+	}
+
+	if (hdr->opcode == ISCSI_OP_SCSI_DATA_IN)
+		goto io_done;
+
+	if (hdr->cmd_status == SAM_STAT_CHECK_CONDITION) {
+		conn->data = (unsigned char *)(sts_bhs->sense_info +
+					sizeof(unsigned short));
+		sense_len =
+		    cpu_to_be16((unsigned short)(sts_bhs->sense_info[0]));
+	} else {
+		conn->data = NULL;

You should not be messing with the conn->data pointer. I think you just want some local char * that set the sense buffer to and then pass that to __iscsi_complete_pdu



We should also add some code to libiscsi to make it not preallocate a buffer for your driver since you do not use it.


+	}
+io_done:
+	if (io_task->cmd_bhs.iscsi_hdr.flags&  ISCSI_FLAG_CMD_READ) {
+		conn->datain_pdus_cnt += 0;
+		if (psol->dw[offsetof(struct amap_sol_cqe, i_res_cnt) / 32]
+				&  SOL_RES_CNT_MASK)
+			 conn->rxdata_octets += (psol->
+			      dw[offsetof(struct amap_sol_cqe, i_res_cnt) / 32]
+							&  SOL_RES_CNT_MASK);
+	} else {
+		conn->dataout_pdus_cnt += 0;
+		conn->txdata_octets += 0;
+	}
+	scsi_dma_unmap(io_task->scsi_cmnd);
+
+	spin_lock(&phba->sgl_lock);
+	free_io_sgl_handle(phba, io_task->psgl_handle);
+	spin_unlock(&phba->sgl_lock);
+	ret = __iscsi_complete_pdu(conn, (struct iscsi_hdr *)hdr, conn->data,
+			     sense_len);
+	if (ret)
+		SE_DEBUG(DBG_LVL_1, "__iscsi_complete_pdu Failed %d \n", ret);
+}
+
+static void
+be_complete_logout(struct beiscsi_conn *beiscsi_conn,
+		   struct iscsi_task *task, struct sol_cqe *psol)
+{
+	struct iscsi_logout_rsp *hdr;
+	struct beiscsi_io_task *io_task = task->dd_data;
+	struct iscsi_conn *conn = beiscsi_conn->conn;
+	struct beiscsi_hba *phba = beiscsi_conn->phba;
+
+	hdr = (struct iscsi_logout_rsp *)task->hdr;
+	hdr->t2wait = 5;
+	hdr->t2retain = 0;
+	hdr->flags = ((psol->dw[offsetof(struct amap_sol_cqe, i_flags) / 32]
+					&  SOL_FLAGS_MASK)>>  24) | 0x80;
+	hdr->response = (psol->dw[offsetof(struct amap_sol_cqe, i_resp) /
+		32]&  SOL_RESP_MASK);
+	hdr->exp_cmdsn = cpu_to_be32(psol->
+			dw[offsetof(struct amap_sol_cqe, i_exp_cmd_sn) / 32]
+						&  SOL_EXP_CMD_SN_MASK);
+	hdr->max_cmdsn = be32_to_cpu((psol->
+			 dw[offsetof(struct amap_sol_cqe, i_exp_cmd_sn) / 32]
+						&  SOL_EXP_CMD_SN_MASK) +
+			((psol->dw[offsetof(struct amap_sol_cqe, i_cmd_wnd)
+				/ 32]&  SOL_CMD_WND_MASK)>>  24) - 1);
+	hdr->hlength = 0;
+	hdr->itt = io_task->itt;
+	spin_lock(&phba->sgl_lock);
+	free_eh_sgl_handle(phba, io_task->psgl_handle);
+	spin_unlock(&phba->sgl_lock);
+	__iscsi_complete_pdu(conn, (struct iscsi_hdr *)hdr, NULL, 0);
+}
+
+static void
+be_complete_tmf(struct beiscsi_conn *beiscsi_conn,
+		struct iscsi_task *task, struct sol_cqe *psol)
+{
+	struct iscsi_tm_rsp *hdr;
+	struct beiscsi_io_task *io_task = task->dd_data;
+	struct iscsi_conn *conn = beiscsi_conn->conn;
+	struct beiscsi_hba *phba = beiscsi_conn->phba;
+
+	hdr = (struct iscsi_tm_rsp *)task->hdr;
+	hdr->flags = ((psol->
+	      dw[offsetof(struct amap_sol_cqe, i_flags) / 32]
+				&  SOL_FLAGS_MASK)>>  24) | 0x80;
+	hdr->response =
+	    (psol->
+	     dw[offsetof(struct amap_sol_cqe, i_resp) /
+		32]&  SOL_RESP_MASK);
+	hdr->exp_cmdsn =
+	    cpu_to_be32(psol->
+			dw[offsetof(struct amap_sol_cqe, i_exp_cmd_sn) / 32]
+						&  SOL_EXP_CMD_SN_MASK);
+	hdr->max_cmdsn =
+	    be32_to_cpu((psol->
+			 dw[offsetof(struct amap_sol_cqe, i_exp_cmd_sn) / 32]
+						&  SOL_EXP_CMD_SN_MASK) +
+			((psol->dw[offsetof(struct amap_sol_cqe, i_cmd_wnd)
+			  / 32]&  SOL_CMD_WND_MASK)>>  24) - 1);
+	hdr->itt = io_task->itt;
+	spin_lock(&phba->sgl_lock);
+	free_eh_sgl_handle(phba, io_task->old_task->psgl_handle);
+	free_eh_sgl_handle(phba, io_task->psgl_handle);
+	spin_unlock(&phba->sgl_lock);
+	__iscsi_complete_pdu(conn, (struct iscsi_hdr *)hdr, NULL, 0);
+}
+
+/* This handles only NOPIN Resp , Req are processed by def pdu */
+static void
+be_complete_nopin_resp(struct beiscsi_conn *beiscsi_conn,
+		       struct iscsi_task *task, struct sol_cqe *psol)
+{
+	struct iscsi_nopin *hdr;
+	struct beiscsi_io_task *io_task = task->dd_data;
+	struct iscsi_conn *conn = beiscsi_conn->conn;
+	struct beiscsi_hba *phba = beiscsi_conn->phba;
+
+	hdr = (struct iscsi_nopin *)task->hdr;
+	hdr->flags = ((psol->
+	      dw[offsetof(struct amap_sol_cqe, i_flags) / 32]
+				&  SOL_FLAGS_MASK)>>  24) | 0x80;
+	hdr->exp_cmdsn =
+	    cpu_to_be32(psol->dw[offsetof(struct amap_sol_cqe, i_exp_cmd_sn)
+			    / 32]&  SOL_EXP_CMD_SN_MASK);
+	hdr->max_cmdsn =
+	    be32_to_cpu((psol->dw[offsetof(struct amap_sol_cqe, i_exp_cmd_sn)
+			/ 32]&  SOL_EXP_CMD_SN_MASK) +
+			((psol->dw[offsetof(struct amap_sol_cqe, i_cmd_wnd)
+			/ 32]&  SOL_CMD_WND_MASK)>>  24) - 1);
+	hdr->itt = io_task->itt;
+	hdr->opcode = io_task->resp_opcode;
+	spin_lock(&phba->sgl_lock);
+	free_eh_sgl_handle(phba, io_task->psgl_handle);
+	spin_unlock(&phba->sgl_lock);
+	__iscsi_complete_pdu(conn, (struct iscsi_hdr *)hdr, NULL, 0);
+}
+
+void
+hwi_complete_cmd(struct beiscsi_conn *beiscsi_conn,
+		 struct beiscsi_hba *phba, struct sol_cqe *psol)
+{
+	struct hwi_wrb_context *pwrb_context;
+	struct wrb_handle *pwrb_handle;
+	struct iscsi_wrb *pwrb = NULL;
+	struct hwi_controller *phwi_ctrlr;
+	struct iscsi_task *task;
+	struct beiscsi_io_task *io_task;
+	struct iscsi_conn *conn = beiscsi_conn->conn;
+	struct iscsi_session *session = conn->session;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+
+	pwrb_context =&phwi_ctrlr->
+	    wrb_context[((psol->dw[offsetof(struct amap_sol_cqe, cid) / 32]
+					&  SOL_CID_MASK)>>  6)];
+	pwrb_handle = (struct wrb_handle *)pwrb_context->
+	    pwrb_handle_basestd[((psol->
+				  dw[offsetof(struct amap_sol_cqe, wrb_index)
+				     / 32]&  SOL_WRB_INDEX_MASK)>>  16)];
+
+	task = (struct iscsi_task *)pwrb_handle->pio_handle;
+	io_task = task->dd_data;
+	spin_lock_bh(&session->lock);
+	pwrb = pwrb_handle->pwrb;
+	switch ((pwrb->
+		 dw[offsetof(struct amap_iscsi_wrb, type) / 32]
+		&  WRB_TYPE_MASK)>>  28) {
+	case HWH_TYPE_IO:
+	case HWH_TYPE_IO_RD:
+		if ((task->hdr->opcode&  ISCSI_OPCODE_MASK) ==
+		    ISCSI_OP_NOOP_OUT) {
+			be_complete_nopin_resp(beiscsi_conn, task, psol);
+		} else {
+			be_complete_io(beiscsi_conn, task, psol);
+		}
+		break;
+
+	case HWH_TYPE_LOGOUT:
+		be_complete_logout(beiscsi_conn, task, psol);
+		break;
+
+	case HWH_TYPE_LOGIN:
+		SE_DEBUG(DBG_LVL_1,
+			 "\t\t No HWH_TYPE_LOGIN Expected in hwi_complete_cmd"
+			 "- Solicited path \n");
+		break;
+
+	case HWH_TYPE_TMF:
+		be_complete_tmf(beiscsi_conn, task, psol);
+		break;
+
+	case HWH_TYPE_NOP:
+		/* This is only in case initiator sent out a NOPOut and gets
+		 * back a NOPIn response. Async NOPIns end up in default PDU
+		 * queue
+		 */
+		be_complete_nopin_resp(beiscsi_conn, task, psol);
+		break;
+
+	default:
+		dev_warn(&phba->pcidev->dev,
+			 "wrb_index 0x%x CID 0x%x\n",
+			 ((psol->
+			   dw[offsetof(struct amap_iscsi_wrb, type) / 32]
+						&  SOL_WRB_INDEX_MASK)>>  16),
+			 ((psol->
+			   dw[offsetof(struct amap_sol_cqe, cid) / 32]
+						&  SOL_CID_MASK)>>  6));
+		break;
+	}
+
+	free_wrb_handle(phba, pwrb_context, pwrb_handle);
+	spin_unlock_bh(&session->lock);
+}
+
+void
+hwi_complete_drvr_msgs(struct beiscsi_conn *beiscsi_conn,
+		       struct beiscsi_hba *phba, struct sol_cqe *psol)
+{
+	struct hwi_wrb_context *pwrb_context;
+	struct wrb_handle *pwrb_handle;
+	struct hwi_controller *phwi_ctrlr;
+	struct iscsi_conn *conn = beiscsi_conn->conn;
+	struct iscsi_session *session = conn->session;
+
+	phwi_ctrlr = GET_HWI_CONTROLLER_WS(phba);
+	pwrb_context =
+	&phwi_ctrlr->
+	    wrb_context[((psol->
+			  dw[offsetof(struct amap_sol_cqe, cid) / 32]
+						&  SOL_CID_MASK)>>  6)];
+	pwrb_handle =
+	    (struct wrb_handle *)pwrb_context->
+	    pwrb_handle_basestd[((psol->
+				  dw[offsetof(struct amap_sol_cqe, wrb_index)
+				     / 32]&  SOL_WRB_INDEX_MASK)>>  16)];
+	spin_lock_bh(&session->lock);
+	free_wrb_handle(phba, pwrb_context, pwrb_handle);
+	spin_unlock_bh(&session->lock);
+
+}
+
+/*
+ *  This is the remove function reported in struct pci_driver.
+ */
+void beiscsi_remove(struct pci_dev *pcidev)
+{
+	struct beiscsi_hba *phba = NULL;
+
+	phba = (struct beiscsi_hba *)pci_get_drvdata(pcidev);
+	if (!phba) {
+		dev_err(&pcidev->dev, "beiscsi_remove called with no phba \n");
+		return;
+	}
+	hwi_disable_intr(phba);
+	if (phba->pcidev->irq)
+		free_irq(phba->pcidev->irq, phba);
+
+	/*
+	 * Lastly stop the hardware
+	 */
+
+	beiscsi_clean_port(phba);
+	if (blk_iopoll_enabled)
+		blk_iopoll_disable(&phba->iopoll);
+	beiscsi_unmap_pci_function(phba);
+	pci_dev_put(phba->pcidev);
+	pci_disable_device(phba->pcidev);
+	beiscsi_free_mem(phba);
+	iscsi_host_remove(phba->shost);
+	iscsi_host_free(phba->shost);



I think we are leaking lots of memory here. There seem to be a lot more kmalloc/kzallocs than there are kfrees.


+}
+
+static int __init beiscsi_module_init(void)
+{
+	int ret;
+
+	beiscsi_scsi_transport =
+	    iscsi_register_transport(&beiscsi_iscsi_transport);
+	if (!beiscsi_scsi_transport) {
+		SE_DEBUG(DBG_LVL_1,
+			 "beiscsi_module_init - Unable to  register beiscsi"
+			 "transport.\n");
+		ret = -ENOMEM;
+	}
+	SE_DEBUG(DBG_LVL_8, "In beiscsi_module_init, tt=%p \n",
+		&beiscsi_iscsi_transport);
+
+	ret = pci_register_driver(&beiscsi_pci_driver);
+	if (ret) {
+		SE_DEBUG(DBG_LVL_1,
+			 "beiscsi_module_init - Unable to  register"
+			 "beiscsi pci driver.\n");
+		goto unregister_iscsi_transport;
+	}
+	return 0;
+
+unregister_iscsi_transport:
+	iscsi_unregister_transport(&beiscsi_iscsi_transport);
+	return ret;
+}
+
+static void __exit beiscsi_module_exit(void)
+{
+	pci_unregister_driver(&beiscsi_pci_driver);
+	iscsi_unregister_transport(&beiscsi_iscsi_transport);
+}
+
+module_init(beiscsi_module_init);
+module_exit(beiscsi_module_exit);

--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [SCSI Target Devel]     [Linux SCSI Target Infrastructure]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Linux IIO]     [Samba]     [Device Mapper]
  Powered by Linux