]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/commitdiff
scsi: elx: efct: Driver initialization routines
authorJames Smart <jsmart2021@gmail.com>
Tue, 1 Jun 2021 23:54:59 +0000 (16:54 -0700)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 16 Jun 2021 03:39:31 +0000 (23:39 -0400)
Add driver definitions for:

 - Emulex FC Target driver init, attach and hardware setup routines.

Link: https://lore.kernel.org/r/20210601235512.20104-19-jsmart2021@gmail.com
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Co-developed-by: Ram Vegesna <ram.vegesna@broadcom.com>
Signed-off-by: Ram Vegesna <ram.vegesna@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/elx/efct/efct_driver.c [new file with mode: 0644]
drivers/scsi/elx/efct/efct_driver.h [new file with mode: 0644]
drivers/scsi/elx/efct/efct_hw.c [new file with mode: 0644]
drivers/scsi/elx/efct/efct_hw.h
drivers/scsi/elx/efct/efct_xport.c [new file with mode: 0644]
drivers/scsi/elx/efct/efct_xport.h [new file with mode: 0644]

diff --git a/drivers/scsi/elx/efct/efct_driver.c b/drivers/scsi/elx/efct/efct_driver.c
new file mode 100644 (file)
index 0000000..2fa5996
--- /dev/null
@@ -0,0 +1,786 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2021 Broadcom. All Rights Reserved. The term
+ * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.
+ */
+
+#include "efct_driver.h"
+
+#include "efct_hw.h"
+#include "efct_unsol.h"
+#include "efct_scsi.h"
+
+LIST_HEAD(efct_devices);
+
+static int logmask;
+module_param(logmask, int, 0444);
+MODULE_PARM_DESC(logmask, "logging bitmask (default 0)");
+
+static struct libefc_function_template efct_libefc_templ = {
+       .issue_mbox_rqst = efct_issue_mbox_rqst,
+       .send_els = efct_els_hw_srrs_send,
+       .send_bls = efct_efc_bls_send,
+
+       .new_nport = efct_scsi_tgt_new_nport,
+       .del_nport = efct_scsi_tgt_del_nport,
+       .scsi_new_node = efct_scsi_new_initiator,
+       .scsi_del_node = efct_scsi_del_initiator,
+       .hw_seq_free = efct_efc_hw_sequence_free,
+};
+
+static int
+efct_device_init(void)
+{
+       int rc;
+
+       /* driver-wide init for target-server */
+       rc = efct_scsi_tgt_driver_init();
+       if (rc) {
+               pr_err("efct_scsi_tgt_init failed rc=%d\n", rc);
+               return rc;
+       }
+
+       rc = efct_scsi_reg_fc_transport();
+       if (rc) {
+               pr_err("failed to register to FC host\n");
+               return rc;
+       }
+
+       return 0;
+}
+
+static void
+efct_device_shutdown(void)
+{
+       efct_scsi_release_fc_transport();
+
+       efct_scsi_tgt_driver_exit();
+}
+
+static void *
+efct_device_alloc(u32 nid)
+{
+       struct efct *efct = NULL;
+
+       efct = kzalloc_node(sizeof(*efct), GFP_KERNEL, nid);
+       if (!efct)
+               return efct;
+
+       INIT_LIST_HEAD(&efct->list_entry);
+       list_add_tail(&efct->list_entry, &efct_devices);
+
+       return efct;
+}
+
+static void
+efct_teardown_msix(struct efct *efct)
+{
+       u32 i;
+
+       for (i = 0; i < efct->n_msix_vec; i++) {
+               free_irq(pci_irq_vector(efct->pci, i),
+                        &efct->intr_context[i]);
+       }
+
+       pci_free_irq_vectors(efct->pci);
+}
+
+static int
+efct_efclib_config(struct efct *efct, struct libefc_function_template *tt)
+{
+       struct efc *efc;
+       struct sli4 *sli;
+       int rc = 0;
+
+       efc = kzalloc(sizeof(*efc), GFP_KERNEL);
+       if (!efc)
+               return -ENOMEM;
+
+       efct->efcport = efc;
+
+       memcpy(&efc->tt, tt, sizeof(*tt));
+       efc->base = efct;
+       efc->pci = efct->pci;
+
+       efc->def_wwnn = efct_get_wwnn(&efct->hw);
+       efc->def_wwpn = efct_get_wwpn(&efct->hw);
+       efc->enable_tgt = 1;
+       efc->log_level = EFC_LOG_LIB;
+
+       sli = &efct->hw.sli;
+       efc->max_xfer_size = sli->sge_supported_length *
+                            sli_get_max_sgl(&efct->hw.sli);
+       efc->sli = sli;
+       efc->fcfi = efct->hw.fcf_indicator;
+
+       rc = efcport_init(efc);
+       if (rc)
+               efc_log_err(efc, "efcport_init failed\n");
+
+       return rc;
+}
+
+static int efct_request_firmware_update(struct efct *efct);
+
+static const char*
+efct_pci_model(u16 device)
+{
+       switch (device) {
+       case EFCT_DEVICE_LANCER_G6:     return "LPE31004";
+       case EFCT_DEVICE_LANCER_G7:     return "LPE36000";
+       default:                        return "unknown";
+       }
+}
+
+static int
+efct_device_attach(struct efct *efct)
+{
+       u32 rc = 0, i = 0;
+
+       if (efct->attached) {
+               efc_log_err(efct, "Device is already attached\n");
+               return -EIO;
+       }
+
+       snprintf(efct->name, sizeof(efct->name), "[%s%d] ", "fc",
+                efct->instance_index);
+
+       efct->logmask = logmask;
+       efct->filter_def = EFCT_DEFAULT_FILTER;
+       efct->max_isr_time_msec = EFCT_OS_MAX_ISR_TIME_MSEC;
+
+       efct->model = efct_pci_model(efct->pci->device);
+
+       efct->efct_req_fw_upgrade = true;
+
+       /* Allocate transport object and bring online */
+       efct->xport = efct_xport_alloc(efct);
+       if (!efct->xport) {
+               efc_log_err(efct, "failed to allocate transport object\n");
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       rc = efct_xport_attach(efct->xport);
+       if (rc) {
+               efc_log_err(efct, "failed to attach transport object\n");
+               goto xport_out;
+       }
+
+       rc = efct_xport_initialize(efct->xport);
+       if (rc) {
+               efc_log_err(efct, "failed to initialize transport object\n");
+               goto xport_out;
+       }
+
+       rc = efct_efclib_config(efct, &efct_libefc_templ);
+       if (rc) {
+               efc_log_err(efct, "failed to init efclib\n");
+               goto efclib_out;
+       }
+
+       for (i = 0; i < efct->n_msix_vec; i++) {
+               efc_log_debug(efct, "irq %d enabled\n", i);
+               enable_irq(pci_irq_vector(efct->pci, i));
+       }
+
+       efct->attached = true;
+
+       if (efct->efct_req_fw_upgrade)
+               efct_request_firmware_update(efct);
+
+       return rc;
+
+efclib_out:
+       efct_xport_detach(efct->xport);
+xport_out:
+       efct_xport_free(efct->xport);
+       efct->xport = NULL;
+out:
+       return rc;
+}
+
+static int
+efct_device_detach(struct efct *efct)
+{
+       int i;
+
+       if (!efct || !efct->attached) {
+               pr_err("Device is not attached\n");
+               return -EIO;
+       }
+
+       if (efct_xport_control(efct->xport, EFCT_XPORT_SHUTDOWN))
+               efc_log_err(efct, "Transport Shutdown timed out\n");
+
+       for (i = 0; i < efct->n_msix_vec; i++)
+               disable_irq(pci_irq_vector(efct->pci, i));
+
+       efct_xport_detach(efct->xport);
+
+       efct_xport_free(efct->xport);
+       efct->xport = NULL;
+
+       efcport_destroy(efct->efcport);
+       kfree(efct->efcport);
+
+       efct->attached = false;
+
+       return 0;
+}
+
+static void
+efct_fw_write_cb(int status, u32 actual_write_length,
+                u32 change_status, void *arg)
+{
+       struct efct_fw_write_result *result = arg;
+
+       result->status = status;
+       result->actual_xfer = actual_write_length;
+       result->change_status = change_status;
+
+       complete(&result->done);
+}
+
+static int
+efct_firmware_write(struct efct *efct, const u8 *buf, size_t buf_len,
+                   u8 *change_status)
+{
+       int rc = 0;
+       u32 bytes_left;
+       u32 xfer_size;
+       u32 offset;
+       struct efc_dma dma;
+       int last = 0;
+       struct efct_fw_write_result result;
+
+       init_completion(&result.done);
+
+       bytes_left = buf_len;
+       offset = 0;
+
+       dma.size = FW_WRITE_BUFSIZE;
+       dma.virt = dma_alloc_coherent(&efct->pci->dev,
+                                     dma.size, &dma.phys, GFP_DMA);
+       if (!dma.virt)
+               return -ENOMEM;
+
+       while (bytes_left > 0) {
+               if (bytes_left > FW_WRITE_BUFSIZE)
+                       xfer_size = FW_WRITE_BUFSIZE;
+               else
+                       xfer_size = bytes_left;
+
+               memcpy(dma.virt, buf + offset, xfer_size);
+
+               if (bytes_left == xfer_size)
+                       last = 1;
+
+               efct_hw_firmware_write(&efct->hw, &dma, xfer_size, offset,
+                                      last, efct_fw_write_cb, &result);
+
+               if (wait_for_completion_interruptible(&result.done) != 0) {
+                       rc = -ENXIO;
+                       break;
+               }
+
+               if (result.actual_xfer == 0 || result.status != 0) {
+                       rc = -EFAULT;
+                       break;
+               }
+
+               if (last)
+                       *change_status = result.change_status;
+
+               bytes_left -= result.actual_xfer;
+               offset += result.actual_xfer;
+       }
+
+       dma_free_coherent(&efct->pci->dev, dma.size, dma.virt, dma.phys);
+       return rc;
+}
+
+static int
+efct_fw_reset(struct efct *efct)
+{
+       /*
+        * Firmware reset to activate the new firmware.
+        * Function 0 will update and load the new firmware
+        * during attach.
+        */
+       if (timer_pending(&efct->xport->stats_timer))
+               del_timer(&efct->xport->stats_timer);
+
+       if (efct_hw_reset(&efct->hw, EFCT_HW_RESET_FIRMWARE)) {
+               efc_log_info(efct, "failed to reset firmware\n");
+               return -EIO;
+       }
+
+       efc_log_info(efct, "successfully reset firmware.Now resetting port\n");
+
+       efct_device_detach(efct);
+       return efct_device_attach(efct);
+}
+
+static int
+efct_request_firmware_update(struct efct *efct)
+{
+       int rc = 0;
+       u8 file_name[256], fw_change_status = 0;
+       const struct firmware *fw;
+       struct efct_hw_grp_hdr *fw_image;
+
+       snprintf(file_name, 256, "%s.grp", efct->model);
+
+       rc = request_firmware(&fw, file_name, &efct->pci->dev);
+       if (rc) {
+               efc_log_debug(efct, "Firmware file(%s) not found.\n", file_name);
+               return rc;
+       }
+
+       fw_image = (struct efct_hw_grp_hdr *)fw->data;
+
+       if (!strncmp(efct->hw.sli.fw_name[0], fw_image->revision,
+                    strnlen(fw_image->revision, 16))) {
+               efc_log_debug(efct,
+                             "Skip update. Firmware is already up to date.\n");
+               goto exit;
+       }
+
+       efc_log_info(efct, "Firmware update is initiated. %s -> %s\n",
+                    efct->hw.sli.fw_name[0], fw_image->revision);
+
+       rc = efct_firmware_write(efct, fw->data, fw->size, &fw_change_status);
+       if (rc) {
+               efc_log_err(efct, "Firmware update failed. rc = %d\n", rc);
+               goto exit;
+       }
+
+       efc_log_info(efct, "Firmware updated successfully\n");
+       switch (fw_change_status) {
+       case 0x00:
+               efc_log_info(efct, "New firmware is active.\n");
+               break;
+       case 0x01:
+               efc_log_info(efct,
+                       "System reboot needed to activate the new firmware\n");
+               break;
+       case 0x02:
+       case 0x03:
+               efc_log_info(efct,
+                            "firmware reset to activate the new firmware\n");
+               efct_fw_reset(efct);
+               break;
+       default:
+               efc_log_info(efct, "Unexected value change_status:%d\n",
+                            fw_change_status);
+               break;
+       }
+
+exit:
+       release_firmware(fw);
+
+       return rc;
+}
+
+static void
+efct_device_free(struct efct *efct)
+{
+       if (efct) {
+               list_del(&efct->list_entry);
+               kfree(efct);
+       }
+}
+
+static int
+efct_device_interrupts_required(struct efct *efct)
+{
+       int rc;
+
+       rc = efct_hw_setup(&efct->hw, efct, efct->pci);
+       if (rc < 0)
+               return rc;
+
+       return efct->hw.config.n_eq;
+}
+
+static irqreturn_t
+efct_intr_thread(int irq, void *handle)
+{
+       struct efct_intr_context *intr_ctx = handle;
+       struct efct *efct = intr_ctx->efct;
+
+       efct_hw_process(&efct->hw, intr_ctx->index, efct->max_isr_time_msec);
+       return IRQ_HANDLED;
+}
+
+static irqreturn_t
+efct_intr_msix(int irq, void *handle)
+{
+       return IRQ_WAKE_THREAD;
+}
+
+static int
+efct_setup_msix(struct efct *efct, u32 num_intrs)
+{
+       int rc = 0, i;
+
+       if (!pci_find_capability(efct->pci, PCI_CAP_ID_MSIX)) {
+               dev_err(&efct->pci->dev,
+                       "%s : MSI-X not available\n", __func__);
+               return -EIO;
+       }
+
+       efct->n_msix_vec = num_intrs;
+
+       rc = pci_alloc_irq_vectors(efct->pci, num_intrs, num_intrs,
+                                  PCI_IRQ_MSIX | PCI_IRQ_AFFINITY);
+
+       if (rc < 0) {
+               dev_err(&efct->pci->dev, "Failed to alloc irq : %d\n", rc);
+               return rc;
+       }
+
+       for (i = 0; i < num_intrs; i++) {
+               struct efct_intr_context *intr_ctx = NULL;
+
+               intr_ctx = &efct->intr_context[i];
+               intr_ctx->efct = efct;
+               intr_ctx->index = i;
+
+               rc = request_threaded_irq(pci_irq_vector(efct->pci, i),
+                                         efct_intr_msix, efct_intr_thread, 0,
+                                         EFCT_DRIVER_NAME, intr_ctx);
+               if (rc) {
+                       dev_err(&efct->pci->dev,
+                               "Failed to register %d vector: %d\n", i, rc);
+                       goto out;
+               }
+       }
+
+       return rc;
+
+out:
+       while (--i >= 0)
+               free_irq(pci_irq_vector(efct->pci, i),
+                        &efct->intr_context[i]);
+
+       pci_free_irq_vectors(efct->pci);
+       return rc;
+}
+
+static struct pci_device_id efct_pci_table[] = {
+       {PCI_DEVICE(EFCT_VENDOR_ID, EFCT_DEVICE_LANCER_G6), 0},
+       {PCI_DEVICE(EFCT_VENDOR_ID, EFCT_DEVICE_LANCER_G7), 0},
+       {}      /* terminate list */
+};
+
+static int
+efct_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+       struct efct *efct = NULL;
+       int rc;
+       u32 i, r;
+       int num_interrupts = 0;
+       int nid;
+
+       dev_info(&pdev->dev, "%s\n", EFCT_DRIVER_NAME);
+
+       rc = pci_enable_device_mem(pdev);
+       if (rc)
+               return rc;
+
+       pci_set_master(pdev);
+
+       rc = pci_set_mwi(pdev);
+       if (rc) {
+               dev_info(&pdev->dev, "pci_set_mwi returned %d\n", rc);
+               goto mwi_out;
+       }
+
+       rc = pci_request_regions(pdev, EFCT_DRIVER_NAME);
+       if (rc) {
+               dev_err(&pdev->dev, "pci_request_regions failed %d\n", rc);
+               goto req_regions_out;
+       }
+
+       /* Fetch the Numa node id for this device */
+       nid = dev_to_node(&pdev->dev);
+       if (nid < 0) {
+               dev_err(&pdev->dev, "Warning Numa node ID is %d\n", nid);
+               nid = 0;
+       }
+
+       /* Allocate efct */
+       efct = efct_device_alloc(nid);
+       if (!efct) {
+               dev_err(&pdev->dev, "Failed to allocate efct\n");
+               rc = -ENOMEM;
+               goto alloc_out;
+       }
+
+       efct->pci = pdev;
+       efct->numa_node = nid;
+
+       /* Map all memory BARs */
+       for (i = 0, r = 0; i < EFCT_PCI_MAX_REGS; i++) {
+               if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
+                       efct->reg[r] = ioremap(pci_resource_start(pdev, i),
+                                              pci_resource_len(pdev, i));
+                       r++;
+               }
+
+               /*
+                * If the 64-bit attribute is set, both this BAR and the
+                * next form the complete address. Skip processing the
+                * next BAR.
+                */
+               if (pci_resource_flags(pdev, i) & IORESOURCE_MEM_64)
+                       i++;
+       }
+
+       pci_set_drvdata(pdev, efct);
+
+       if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) != 0 ||
+           pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
+               dev_warn(&pdev->dev, "trying DMA_BIT_MASK(32)\n");
+               if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0 ||
+                   pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
+                       dev_err(&pdev->dev, "setting DMA_BIT_MASK failed\n");
+                       rc = -1;
+                       goto dma_mask_out;
+               }
+       }
+
+       num_interrupts = efct_device_interrupts_required(efct);
+       if (num_interrupts < 0) {
+               efc_log_err(efct, "efct_device_interrupts_required failed\n");
+               rc = -1;
+               goto dma_mask_out;
+       }
+
+       /*
+        * Initialize MSIX interrupts, note,
+        * efct_setup_msix() enables the interrupt
+        */
+       rc = efct_setup_msix(efct, num_interrupts);
+       if (rc) {
+               dev_err(&pdev->dev, "Can't setup msix\n");
+               goto dma_mask_out;
+       }
+       /* Disable interrupt for now */
+       for (i = 0; i < efct->n_msix_vec; i++) {
+               efc_log_debug(efct, "irq %d disabled\n", i);
+               disable_irq(pci_irq_vector(efct->pci, i));
+       }
+
+       rc = efct_device_attach(efct);
+       if (rc)
+               goto attach_out;
+
+       return 0;
+
+attach_out:
+       efct_teardown_msix(efct);
+dma_mask_out:
+       pci_set_drvdata(pdev, NULL);
+
+       for (i = 0; i < EFCT_PCI_MAX_REGS; i++) {
+               if (efct->reg[i])
+                       iounmap(efct->reg[i]);
+       }
+       efct_device_free(efct);
+alloc_out:
+       pci_release_regions(pdev);
+req_regions_out:
+       pci_clear_mwi(pdev);
+mwi_out:
+       pci_disable_device(pdev);
+       return rc;
+}
+
+static void
+efct_pci_remove(struct pci_dev *pdev)
+{
+       struct efct *efct = pci_get_drvdata(pdev);
+       u32 i;
+
+       if (!efct)
+               return;
+
+       efct_device_detach(efct);
+
+       efct_teardown_msix(efct);
+
+       for (i = 0; i < EFCT_PCI_MAX_REGS; i++) {
+               if (efct->reg[i])
+                       iounmap(efct->reg[i]);
+       }
+
+       pci_set_drvdata(pdev, NULL);
+
+       efct_device_free(efct);
+
+       pci_release_regions(pdev);
+
+       pci_disable_device(pdev);
+}
+
+static void
+efct_device_prep_for_reset(struct efct *efct, struct pci_dev *pdev)
+{
+       if (efct) {
+               efc_log_debug(efct,
+                             "PCI channel disable preparing for reset\n");
+               efct_device_detach(efct);
+               /* Disable interrupt and pci device */
+               efct_teardown_msix(efct);
+       }
+       pci_disable_device(pdev);
+}
+
+static void
+efct_device_prep_for_recover(struct efct *efct)
+{
+       if (efct) {
+               efc_log_debug(efct, "PCI channel preparing for recovery\n");
+               efct_hw_io_abort_all(&efct->hw);
+       }
+}
+
+/**
+ * efct_pci_io_error_detected - method for handling PCI I/O error
+ * @pdev: pointer to PCI device.
+ * @state: the current PCI connection state.
+ *
+ * This routine is registered to the PCI subsystem for error handling. This
+ * function is called by the PCI subsystem after a PCI bus error affecting
+ * this device has been detected. When this routine is invoked, it dispatches
+ * device error detected handling routine, which will perform the proper
+ * error detected operation.
+ *
+ * Return codes
+ * PCI_ERS_RESULT_NEED_RESET - need to reset before recovery
+ * PCI_ERS_RESULT_DISCONNECT - device could not be recovered
+ */
+static pci_ers_result_t
+efct_pci_io_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
+{
+       struct efct *efct = pci_get_drvdata(pdev);
+       pci_ers_result_t rc;
+
+       switch (state) {
+       case pci_channel_io_normal:
+               efct_device_prep_for_recover(efct);
+               rc = PCI_ERS_RESULT_CAN_RECOVER;
+               break;
+       case pci_channel_io_frozen:
+               efct_device_prep_for_reset(efct, pdev);
+               rc = PCI_ERS_RESULT_NEED_RESET;
+               break;
+       case pci_channel_io_perm_failure:
+               efct_device_detach(efct);
+               rc = PCI_ERS_RESULT_DISCONNECT;
+               break;
+       default:
+               efc_log_debug(efct, "Unknown PCI error state:0x%x\n", state);
+               efct_device_prep_for_reset(efct, pdev);
+               rc = PCI_ERS_RESULT_NEED_RESET;
+               break;
+       }
+
+       return rc;
+}
+
+static pci_ers_result_t
+efct_pci_io_slot_reset(struct pci_dev *pdev)
+{
+       int rc;
+       struct efct *efct = pci_get_drvdata(pdev);
+
+       rc = pci_enable_device_mem(pdev);
+       if (rc) {
+               efc_log_err(efct, "failed to enable PCI device after reset\n");
+               return PCI_ERS_RESULT_DISCONNECT;
+       }
+
+       /*
+        * As the new kernel behavior of pci_restore_state() API call clears
+        * device saved_state flag, need to save the restored state again.
+        */
+
+       pci_save_state(pdev);
+
+       pci_set_master(pdev);
+
+       rc = efct_setup_msix(efct, efct->n_msix_vec);
+       if (rc)
+               efc_log_err(efct, "rc %d returned, IRQ allocation failed\n",
+                           rc);
+
+       /* Perform device reset */
+       efct_device_detach(efct);
+       /* Bring device to online*/
+       efct_device_attach(efct);
+
+       return PCI_ERS_RESULT_RECOVERED;
+}
+
+static void
+efct_pci_io_resume(struct pci_dev *pdev)
+{
+       struct efct *efct = pci_get_drvdata(pdev);
+
+       /* Perform device reset */
+       efct_device_detach(efct);
+       /* Bring device to online*/
+       efct_device_attach(efct);
+}
+
+MODULE_DEVICE_TABLE(pci, efct_pci_table);
+
+static struct pci_error_handlers efct_pci_err_handler = {
+       .error_detected = efct_pci_io_error_detected,
+       .slot_reset = efct_pci_io_slot_reset,
+       .resume = efct_pci_io_resume,
+};
+
+static struct pci_driver efct_pci_driver = {
+       .name           = EFCT_DRIVER_NAME,
+       .id_table       = efct_pci_table,
+       .probe          = efct_pci_probe,
+       .remove         = efct_pci_remove,
+       .err_handler    = &efct_pci_err_handler,
+};
+
+static
+int __init efct_init(void)
+{
+       int rc;
+
+       rc = efct_device_init();
+       if (rc) {
+               pr_err("efct_device_init failed rc=%d\n", rc);
+               return rc;
+       }
+
+       rc = pci_register_driver(&efct_pci_driver);
+       if (rc) {
+               pr_err("pci_register_driver failed rc=%d\n", rc);
+               efct_device_shutdown();
+       }
+
+       return rc;
+}
+
+static void __exit efct_exit(void)
+{
+       pci_unregister_driver(&efct_pci_driver);
+       efct_device_shutdown();
+}
+
+module_init(efct_init);
+module_exit(efct_exit);
+MODULE_VERSION(EFCT_DRIVER_VERSION);
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Broadcom");
diff --git a/drivers/scsi/elx/efct/efct_driver.h b/drivers/scsi/elx/efct/efct_driver.h
new file mode 100644 (file)
index 0000000..dab8eac
--- /dev/null
@@ -0,0 +1,109 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2021 Broadcom. All Rights Reserved. The term
+ * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.
+ */
+
+#if !defined(__EFCT_DRIVER_H__)
+#define __EFCT_DRIVER_H__
+
+/***************************************************************************
+ * OS specific includes
+ */
+#include <stdarg.h>
+#include <linux/module.h>
+#include <linux/debugfs.h>
+#include <linux/firmware.h>
+#include "../include/efc_common.h"
+#include "../libefc/efclib.h"
+#include "efct_hw.h"
+#include "efct_io.h"
+#include "efct_xport.h"
+
+#define EFCT_DRIVER_NAME                       "efct"
+#define EFCT_DRIVER_VERSION                    "1.0.0.0"
+
+/* EFCT_DEFAULT_FILTER-
+ * MRQ filter to segregate the IO flow.
+ */
+#define EFCT_DEFAULT_FILTER                    "0x01ff22ff,0,0,0"
+
+/* EFCT_OS_MAX_ISR_TIME_MSEC -
+ * maximum time driver code should spend in an interrupt
+ * or kernel thread context without yielding
+ */
+#define EFCT_OS_MAX_ISR_TIME_MSEC              1000
+
+#define EFCT_FC_MAX_SGL                                64
+#define EFCT_FC_DIF_SEED                       0
+
+/* Watermark */
+#define EFCT_WATERMARK_HIGH_PCT                        90
+#define EFCT_WATERMARK_LOW_PCT                 80
+#define EFCT_IO_WATERMARK_PER_INITIATOR                8
+
+#define EFCT_PCI_MAX_REGS                      6
+#define MAX_PCI_INTERRUPTS                     16
+
+struct efct_intr_context {
+       struct efct             *efct;
+       u32                     index;
+};
+
+struct efct {
+       struct pci_dev                  *pci;
+       void __iomem                    *reg[EFCT_PCI_MAX_REGS];
+
+       u32                             n_msix_vec;
+       bool                            attached;
+       bool                            soft_wwn_enable;
+       u8                              efct_req_fw_upgrade;
+       struct efct_intr_context        intr_context[MAX_PCI_INTERRUPTS];
+       u32                             numa_node;
+
+       char                            name[EFC_NAME_LENGTH];
+       u32                             instance_index;
+       struct list_head                list_entry;
+       struct efct_scsi_tgt            tgt_efct;
+       struct efct_xport               *xport;
+       struct efc                      *efcport;
+       struct Scsi_Host                *shost;
+       int                             logmask;
+       u32                             max_isr_time_msec;
+
+       const char                      *desc;
+
+       const char                      *model;
+
+       struct efct_hw                  hw;
+
+       u32                             rq_selection_policy;
+       char                            *filter_def;
+       int                             topology;
+
+       /* Look up for target node */
+       struct xarray                   lookup;
+
+       /*
+        * Target IO timer value:
+        * Zero: target command timeout disabled.
+        * Non-zero: Timeout value, in seconds, for target commands
+        */
+       u32                             target_io_timer_sec;
+
+       int                             speed;
+       struct dentry                   *sess_debugfs_dir;
+};
+
+#define FW_WRITE_BUFSIZE               (64 * 1024)
+
+struct efct_fw_write_result {
+       struct completion done;
+       int status;
+       u32 actual_xfer;
+       u32 change_status;
+};
+
+extern struct list_head                        efct_devices;
+
+#endif /* __EFCT_DRIVER_H__ */
diff --git a/drivers/scsi/elx/efct/efct_hw.c b/drivers/scsi/elx/efct/efct_hw.c
new file mode 100644 (file)
index 0000000..e726b93
--- /dev/null
@@ -0,0 +1,1153 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2021 Broadcom. All Rights Reserved. The term
+ * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.
+ */
+
+#include "efct_driver.h"
+#include "efct_hw.h"
+#include "efct_unsol.h"
+
+struct efct_mbox_rqst_ctx {
+       int (*callback)(struct efc *efc, int status, u8 *mqe, void *arg);
+       void *arg;
+};
+
+static int
+efct_hw_link_event_init(struct efct_hw *hw)
+{
+       hw->link.status = SLI4_LINK_STATUS_MAX;
+       hw->link.topology = SLI4_LINK_TOPO_NONE;
+       hw->link.medium = SLI4_LINK_MEDIUM_MAX;
+       hw->link.speed = 0;
+       hw->link.loop_map = NULL;
+       hw->link.fc_id = U32_MAX;
+
+       return 0;
+}
+
+static int
+efct_hw_read_max_dump_size(struct efct_hw *hw)
+{
+       u8 buf[SLI4_BMBX_SIZE];
+       struct efct *efct = hw->os;
+       int rc = 0;
+       struct sli4_rsp_cmn_set_dump_location *rsp;
+
+       /* attempt to detemine the dump size for function 0 only. */
+       if (PCI_FUNC(efct->pci->devfn) != 0)
+               return rc;
+
+       if (sli_cmd_common_set_dump_location(&hw->sli, buf, 1, 0, NULL, 0))
+               return -EIO;
+
+       rsp = (struct sli4_rsp_cmn_set_dump_location *)
+             (buf + offsetof(struct sli4_cmd_sli_config, payload.embed));
+
+       rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
+       if (rc != 0) {
+               efc_log_debug(hw->os, "set dump location cmd failed\n");
+               return rc;
+       }
+
+       hw->dump_size =
+         le32_to_cpu(rsp->buffer_length_dword) & SLI4_CMN_SET_DUMP_BUFFER_LEN;
+
+       efc_log_debug(hw->os, "Dump size %x\n", hw->dump_size);
+
+       return rc;
+}
+
+static int
+__efct_read_topology_cb(struct efct_hw *hw, int status, u8 *mqe, void *arg)
+{
+       struct sli4_cmd_read_topology *read_topo =
+                               (struct sli4_cmd_read_topology *)mqe;
+       u8 speed;
+       struct efc_domain_record drec = {0};
+       struct efct *efct = hw->os;
+
+       if (status || le16_to_cpu(read_topo->hdr.status)) {
+               efc_log_debug(hw->os, "bad status cqe=%#x mqe=%#x\n", status,
+                             le16_to_cpu(read_topo->hdr.status));
+               return -EIO;
+       }
+
+       switch (le32_to_cpu(read_topo->dw2_attentype) &
+               SLI4_READTOPO_ATTEN_TYPE) {
+       case SLI4_READ_TOPOLOGY_LINK_UP:
+               hw->link.status = SLI4_LINK_STATUS_UP;
+               break;
+       case SLI4_READ_TOPOLOGY_LINK_DOWN:
+               hw->link.status = SLI4_LINK_STATUS_DOWN;
+               break;
+       case SLI4_READ_TOPOLOGY_LINK_NO_ALPA:
+               hw->link.status = SLI4_LINK_STATUS_NO_ALPA;
+               break;
+       default:
+               hw->link.status = SLI4_LINK_STATUS_MAX;
+               break;
+       }
+
+       switch (read_topo->topology) {
+       case SLI4_READ_TOPO_NON_FC_AL:
+               hw->link.topology = SLI4_LINK_TOPO_NON_FC_AL;
+               break;
+       case SLI4_READ_TOPO_FC_AL:
+               hw->link.topology = SLI4_LINK_TOPO_FC_AL;
+               if (hw->link.status == SLI4_LINK_STATUS_UP)
+                       hw->link.loop_map = hw->loop_map.virt;
+               hw->link.fc_id = read_topo->acquired_al_pa;
+               break;
+       default:
+               hw->link.topology = SLI4_LINK_TOPO_MAX;
+               break;
+       }
+
+       hw->link.medium = SLI4_LINK_MEDIUM_FC;
+
+       speed = (le32_to_cpu(read_topo->currlink_state) &
+                SLI4_READTOPO_LINKSTATE_SPEED) >> 8;
+       switch (speed) {
+       case SLI4_READ_TOPOLOGY_SPEED_1G:
+               hw->link.speed =  1 * 1000;
+               break;
+       case SLI4_READ_TOPOLOGY_SPEED_2G:
+               hw->link.speed =  2 * 1000;
+               break;
+       case SLI4_READ_TOPOLOGY_SPEED_4G:
+               hw->link.speed =  4 * 1000;
+               break;
+       case SLI4_READ_TOPOLOGY_SPEED_8G:
+               hw->link.speed =  8 * 1000;
+               break;
+       case SLI4_READ_TOPOLOGY_SPEED_16G:
+               hw->link.speed = 16 * 1000;
+               break;
+       case SLI4_READ_TOPOLOGY_SPEED_32G:
+               hw->link.speed = 32 * 1000;
+               break;
+       case SLI4_READ_TOPOLOGY_SPEED_64G:
+               hw->link.speed = 64 * 1000;
+               break;
+       case SLI4_READ_TOPOLOGY_SPEED_128G:
+               hw->link.speed = 128 * 1000;
+               break;
+       }
+
+       drec.speed = hw->link.speed;
+       drec.fc_id = hw->link.fc_id;
+       drec.is_nport = true;
+       efc_domain_cb(efct->efcport, EFC_HW_DOMAIN_FOUND, &drec);
+
+       return 0;
+}
+
+static int
+efct_hw_cb_link(void *ctx, void *e)
+{
+       struct efct_hw *hw = ctx;
+       struct sli4_link_event *event = e;
+       struct efc_domain *d = NULL;
+       int rc = 0;
+       struct efct *efct = hw->os;
+
+       efct_hw_link_event_init(hw);
+
+       switch (event->status) {
+       case SLI4_LINK_STATUS_UP:
+
+               hw->link = *event;
+               efct->efcport->link_status = EFC_LINK_STATUS_UP;
+
+               if (event->topology == SLI4_LINK_TOPO_NON_FC_AL) {
+                       struct efc_domain_record drec = {0};
+
+                       efc_log_info(hw->os, "Link Up, NPORT, speed is %d\n",
+                                    event->speed);
+                       drec.speed = event->speed;
+                       drec.fc_id = event->fc_id;
+                       drec.is_nport = true;
+                       efc_domain_cb(efct->efcport, EFC_HW_DOMAIN_FOUND,
+                                     &drec);
+               } else if (event->topology == SLI4_LINK_TOPO_FC_AL) {
+                       u8 buf[SLI4_BMBX_SIZE];
+
+                       efc_log_info(hw->os, "Link Up, LOOP, speed is %d\n",
+                                    event->speed);
+
+                       if (!sli_cmd_read_topology(&hw->sli, buf,
+                                                  &hw->loop_map)) {
+                               rc = efct_hw_command(hw, buf, EFCT_CMD_NOWAIT,
+                                               __efct_read_topology_cb, NULL);
+                       }
+
+                       if (rc)
+                               efc_log_debug(hw->os, "READ_TOPOLOGY failed\n");
+               } else {
+                       efc_log_info(hw->os, "%s(%#x), speed is %d\n",
+                                    "Link Up, unsupported topology ",
+                                    event->topology, event->speed);
+               }
+               break;
+       case SLI4_LINK_STATUS_DOWN:
+               efc_log_info(hw->os, "Link down\n");
+
+               hw->link.status = event->status;
+               efct->efcport->link_status = EFC_LINK_STATUS_DOWN;
+
+               d = efct->efcport->domain;
+               if (d)
+                       efc_domain_cb(efct->efcport, EFC_HW_DOMAIN_LOST, d);
+               break;
+       default:
+               efc_log_debug(hw->os, "unhandled link status %#x\n",
+                             event->status);
+               break;
+       }
+
+       return 0;
+}
+
+int
+efct_hw_setup(struct efct_hw *hw, void *os, struct pci_dev *pdev)
+{
+       u32 i, max_sgl, cpus;
+
+       if (hw->hw_setup_called)
+               return 0;
+
+       /*
+        * efct_hw_init() relies on NULL pointers indicating that a structure
+        * needs allocation. If a structure is non-NULL, efct_hw_init() won't
+        * free/realloc that memory
+        */
+       memset(hw, 0, sizeof(struct efct_hw));
+
+       hw->hw_setup_called = true;
+
+       hw->os = os;
+
+       mutex_init(&hw->bmbx_lock);
+       spin_lock_init(&hw->cmd_lock);
+       INIT_LIST_HEAD(&hw->cmd_head);
+       INIT_LIST_HEAD(&hw->cmd_pending);
+       hw->cmd_head_count = 0;
+
+       /* Create mailbox command ctx pool */
+       hw->cmd_ctx_pool = mempool_create_kmalloc_pool(EFCT_CMD_CTX_POOL_SZ,
+                                       sizeof(struct efct_command_ctx));
+       if (!hw->cmd_ctx_pool) {
+               efc_log_err(hw->os, "failed to allocate mailbox buffer pool\n");
+               return -EIO;
+       }
+
+       /* Create mailbox request ctx pool for library callback */
+       hw->mbox_rqst_pool = mempool_create_kmalloc_pool(EFCT_CMD_CTX_POOL_SZ,
+                                       sizeof(struct efct_mbox_rqst_ctx));
+       if (!hw->mbox_rqst_pool) {
+               efc_log_err(hw->os, "failed to allocate mbox request pool\n");
+               return -EIO;
+       }
+
+       spin_lock_init(&hw->io_lock);
+       INIT_LIST_HEAD(&hw->io_inuse);
+       INIT_LIST_HEAD(&hw->io_free);
+       INIT_LIST_HEAD(&hw->io_wait_free);
+
+       atomic_set(&hw->io_alloc_failed_count, 0);
+
+       hw->config.speed = SLI4_LINK_SPEED_AUTO_16_8_4;
+       if (sli_setup(&hw->sli, hw->os, pdev, ((struct efct *)os)->reg)) {
+               efc_log_err(hw->os, "SLI setup failed\n");
+               return -EIO;
+       }
+
+       efct_hw_link_event_init(hw);
+
+       sli_callback(&hw->sli, SLI4_CB_LINK, efct_hw_cb_link, hw);
+
+       /*
+        * Set all the queue sizes to the maximum allowed.
+        */
+       for (i = 0; i < ARRAY_SIZE(hw->num_qentries); i++)
+               hw->num_qentries[i] = hw->sli.qinfo.max_qentries[i];
+       /*
+        * Adjust the size of the WQs so that the CQ is twice as big as
+        * the WQ to allow for 2 completions per IO. This allows us to
+        * handle multi-phase as well as aborts.
+        */
+       hw->num_qentries[SLI4_QTYPE_WQ] = hw->num_qentries[SLI4_QTYPE_CQ] / 2;
+
+       /*
+        * The RQ assignment for RQ pair mode.
+        */
+
+       hw->config.rq_default_buffer_size = EFCT_HW_RQ_SIZE_PAYLOAD;
+       hw->config.n_io = hw->sli.ext[SLI4_RSRC_XRI].size;
+
+       cpus = num_possible_cpus();
+       hw->config.n_eq = cpus > EFCT_HW_MAX_NUM_EQ ? EFCT_HW_MAX_NUM_EQ : cpus;
+
+       max_sgl = sli_get_max_sgl(&hw->sli) - SLI4_SGE_MAX_RESERVED;
+       max_sgl = (max_sgl > EFCT_FC_MAX_SGL) ? EFCT_FC_MAX_SGL : max_sgl;
+       hw->config.n_sgl = max_sgl;
+
+       (void)efct_hw_read_max_dump_size(hw);
+
+       return 0;
+}
+
+static void
+efct_logfcfi(struct efct_hw *hw, u32 j, u32 i, u32 id)
+{
+       efc_log_info(hw->os,
+                    "REG_FCFI: filter[%d] %08X -> RQ[%d] id=%d\n",
+                    j, hw->config.filter_def[j], i, id);
+}
+
+static inline void
+efct_hw_init_free_io(struct efct_hw_io *io)
+{
+       /*
+        * Set io->done to NULL, to avoid any callbacks, should
+        * a completion be received for one of these IOs
+        */
+       io->done = NULL;
+       io->abort_done = NULL;
+       io->status_saved = false;
+       io->abort_in_progress = false;
+       io->type = 0xFFFF;
+       io->wq = NULL;
+}
+
+static u8 efct_hw_iotype_is_originator(u16 io_type)
+{
+       switch (io_type) {
+       case EFCT_HW_FC_CT:
+       case EFCT_HW_ELS_REQ:
+               return 0;
+       default:
+               return -EIO;
+       }
+}
+
+static void
+efct_hw_io_restore_sgl(struct efct_hw *hw, struct efct_hw_io *io)
+{
+       /* Restore the default */
+       io->sgl = &io->def_sgl;
+       io->sgl_count = io->def_sgl_count;
+}
+
+static void
+efct_hw_wq_process_io(void *arg, u8 *cqe, int status)
+{
+       struct efct_hw_io *io = arg;
+       struct efct_hw *hw = io->hw;
+       struct sli4_fc_wcqe *wcqe = (void *)cqe;
+       u32     len = 0;
+       u32 ext = 0;
+
+       /* clear xbusy flag if WCQE[XB] is clear */
+       if (io->xbusy && (wcqe->flags & SLI4_WCQE_XB) == 0)
+               io->xbusy = false;
+
+       /* get extended CQE status */
+       switch (io->type) {
+       case EFCT_HW_BLS_ACC:
+       case EFCT_HW_BLS_RJT:
+               break;
+       case EFCT_HW_ELS_REQ:
+               sli_fc_els_did(&hw->sli, cqe, &ext);
+               len = sli_fc_response_length(&hw->sli, cqe);
+               break;
+       case EFCT_HW_ELS_RSP:
+       case EFCT_HW_FC_CT_RSP:
+               break;
+       case EFCT_HW_FC_CT:
+               len = sli_fc_response_length(&hw->sli, cqe);
+               break;
+       case EFCT_HW_IO_TARGET_WRITE:
+               len = sli_fc_io_length(&hw->sli, cqe);
+               break;
+       case EFCT_HW_IO_TARGET_READ:
+               len = sli_fc_io_length(&hw->sli, cqe);
+               break;
+       case EFCT_HW_IO_TARGET_RSP:
+               break;
+       case EFCT_HW_IO_DNRX_REQUEUE:
+               /* release the count for re-posting the buffer */
+               /* efct_hw_io_free(hw, io); */
+               break;
+       default:
+               efc_log_err(hw->os, "unhandled io type %#x for XRI 0x%x\n",
+                           io->type, io->indicator);
+               break;
+       }
+       if (status) {
+               ext = sli_fc_ext_status(&hw->sli, cqe);
+               /*
+                * If we're not an originator IO, and XB is set, then issue
+                * abort for the IO from within the HW
+                */
+               if ((!efct_hw_iotype_is_originator(io->type)) &&
+                   wcqe->flags & SLI4_WCQE_XB) {
+                       int rc;
+
+                       efc_log_debug(hw->os, "aborting xri=%#x tag=%#x\n",
+                                     io->indicator, io->reqtag);
+
+                       /*
+                        * Because targets may send a response when the IO
+                        * completes using the same XRI, we must wait for the
+                        * XRI_ABORTED CQE to issue the IO callback
+                        */
+                       rc = efct_hw_io_abort(hw, io, false, NULL, NULL);
+                       if (rc == 0) {
+                               /*
+                                * latch status to return after abort is
+                                * complete
+                                */
+                               io->status_saved = true;
+                               io->saved_status = status;
+                               io->saved_ext = ext;
+                               io->saved_len = len;
+                               goto exit_efct_hw_wq_process_io;
+                       } else if (rc == -EINPROGRESS) {
+                               /*
+                                * Already being aborted by someone else (ABTS
+                                * perhaps). Just return original
+                                * error.
+                                */
+                               efc_log_debug(hw->os, "%s%#x tag=%#x\n",
+                                             "abort in progress xri=",
+                                             io->indicator, io->reqtag);
+
+                       } else {
+                               /* Failed to abort for some other reason, log
+                                * error
+                                */
+                               efc_log_debug(hw->os, "%s%#x tag=%#x rc=%d\n",
+                                             "Failed to abort xri=",
+                                             io->indicator, io->reqtag, rc);
+                       }
+               }
+       }
+
+       if (io->done) {
+               efct_hw_done_t done = io->done;
+
+               io->done = NULL;
+
+               if (io->status_saved) {
+                       /* use latched status if exists */
+                       status = io->saved_status;
+                       len = io->saved_len;
+                       ext = io->saved_ext;
+                       io->status_saved = false;
+               }
+
+               /* Restore default SGL */
+               efct_hw_io_restore_sgl(hw, io);
+               done(io, len, status, ext, io->arg);
+       }
+
+exit_efct_hw_wq_process_io:
+       return;
+}
+
+static int
+efct_hw_setup_io(struct efct_hw *hw)
+{
+       u32     i = 0;
+       struct efct_hw_io       *io = NULL;
+       uintptr_t       xfer_virt = 0;
+       uintptr_t       xfer_phys = 0;
+       u32     index;
+       bool new_alloc = true;
+       struct efc_dma *dma;
+       struct efct *efct = hw->os;
+
+       if (!hw->io) {
+               hw->io = kmalloc_array(hw->config.n_io, sizeof(io), GFP_KERNEL);
+               if (!hw->io)
+                       return -ENOMEM;
+
+               memset(hw->io, 0, hw->config.n_io * sizeof(io));
+
+               for (i = 0; i < hw->config.n_io; i++) {
+                       hw->io[i] = kzalloc(sizeof(*io), GFP_KERNEL);
+                       if (!hw->io[i])
+                               goto error;
+               }
+
+               /* Create WQE buffs for IO */
+               hw->wqe_buffs = kzalloc((hw->config.n_io * hw->sli.wqe_size),
+                                       GFP_KERNEL);
+               if (!hw->wqe_buffs) {
+                       kfree(hw->io);
+                       return -ENOMEM;
+               }
+
+       } else {
+               /* re-use existing IOs, including SGLs */
+               new_alloc = false;
+       }
+
+       if (new_alloc) {
+               dma = &hw->xfer_rdy;
+               dma->size = sizeof(struct fcp_txrdy) * hw->config.n_io;
+               dma->virt = dma_alloc_coherent(&efct->pci->dev,
+                                              dma->size, &dma->phys, GFP_DMA);
+               if (!dma->virt)
+                       return -ENOMEM;
+       }
+       xfer_virt = (uintptr_t)hw->xfer_rdy.virt;
+       xfer_phys = hw->xfer_rdy.phys;
+
+       /* Initialize the pool of HW IO objects */
+       for (i = 0; i < hw->config.n_io; i++) {
+               struct hw_wq_callback *wqcb;
+
+               io = hw->io[i];
+
+               /* initialize IO fields */
+               io->hw = hw;
+
+               /* Assign a WQE buff */
+               io->wqe.wqebuf = &hw->wqe_buffs[i * hw->sli.wqe_size];
+
+               /* Allocate the request tag for this IO */
+               wqcb = efct_hw_reqtag_alloc(hw, efct_hw_wq_process_io, io);
+               if (!wqcb) {
+                       efc_log_err(hw->os, "can't allocate request tag\n");
+                       return -ENOSPC;
+               }
+               io->reqtag = wqcb->instance_index;
+
+               /* Now for the fields that are initialized on each free */
+               efct_hw_init_free_io(io);
+
+               /* The XB flag isn't cleared on IO free, so init to zero */
+               io->xbusy = 0;
+
+               if (sli_resource_alloc(&hw->sli, SLI4_RSRC_XRI,
+                                      &io->indicator, &index)) {
+                       efc_log_err(hw->os,
+                                   "sli_resource_alloc failed @ %d\n", i);
+                       return -ENOMEM;
+               }
+
+               if (new_alloc) {
+                       dma = &io->def_sgl;
+                       dma->size = hw->config.n_sgl *
+                                       sizeof(struct sli4_sge);
+                       dma->virt = dma_alloc_coherent(&efct->pci->dev,
+                                                      dma->size, &dma->phys,
+                                                      GFP_DMA);
+                       if (!dma->virt) {
+                               efc_log_err(hw->os, "dma_alloc fail %d\n", i);
+                               memset(&io->def_sgl, 0,
+                                      sizeof(struct efc_dma));
+                               return -ENOMEM;
+                       }
+               }
+               io->def_sgl_count = hw->config.n_sgl;
+               io->sgl = &io->def_sgl;
+               io->sgl_count = io->def_sgl_count;
+
+               if (hw->xfer_rdy.size) {
+                       io->xfer_rdy.virt = (void *)xfer_virt;
+                       io->xfer_rdy.phys = xfer_phys;
+                       io->xfer_rdy.size = sizeof(struct fcp_txrdy);
+
+                       xfer_virt += sizeof(struct fcp_txrdy);
+                       xfer_phys += sizeof(struct fcp_txrdy);
+               }
+       }
+
+       return 0;
+error:
+       for (i = 0; i < hw->config.n_io && hw->io[i]; i++) {
+               kfree(hw->io[i]);
+               hw->io[i] = NULL;
+       }
+
+       kfree(hw->io);
+       hw->io = NULL;
+
+       return -ENOMEM;
+}
+
+static int
+efct_hw_init_prereg_io(struct efct_hw *hw)
+{
+       u32 i, idx = 0;
+       struct efct_hw_io *io = NULL;
+       u8 cmd[SLI4_BMBX_SIZE];
+       int rc = 0;
+       u32 n_rem;
+       u32 n = 0;
+       u32 sgls_per_request = 256;
+       struct efc_dma **sgls = NULL;
+       struct efc_dma req;
+       struct efct *efct = hw->os;
+
+       sgls = kmalloc_array(sgls_per_request, sizeof(*sgls), GFP_KERNEL);
+       if (!sgls)
+               return -ENOMEM;
+
+       memset(&req, 0, sizeof(struct efc_dma));
+       req.size = 32 + sgls_per_request * 16;
+       req.virt = dma_alloc_coherent(&efct->pci->dev, req.size, &req.phys,
+                                     GFP_DMA);
+       if (!req.virt) {
+               kfree(sgls);
+               return -ENOMEM;
+       }
+
+       for (n_rem = hw->config.n_io; n_rem; n_rem -= n) {
+               /* Copy address of SGL's into local sgls[] array, break
+                * out if the xri is not contiguous.
+                */
+               u32 min = (sgls_per_request < n_rem) ? sgls_per_request : n_rem;
+
+               for (n = 0; n < min; n++) {
+                       /* Check that we have contiguous xri values */
+                       if (n > 0) {
+                               if (hw->io[idx + n]->indicator !=
+                                   hw->io[idx + n - 1]->indicator + 1)
+                                       break;
+                       }
+
+                       sgls[n] = hw->io[idx + n]->sgl;
+               }
+
+               if (sli_cmd_post_sgl_pages(&hw->sli, cmd,
+                               hw->io[idx]->indicator, n, sgls, NULL, &req)) {
+                       rc = -EIO;
+                       break;
+               }
+
+               rc = efct_hw_command(hw, cmd, EFCT_CMD_POLL, NULL, NULL);
+               if (rc) {
+                       efc_log_err(hw->os, "SGL post failed, rc=%d\n", rc);
+                       break;
+               }
+
+               /* Add to tail if successful */
+               for (i = 0; i < n; i++, idx++) {
+                       io = hw->io[idx];
+                       io->state = EFCT_HW_IO_STATE_FREE;
+                       INIT_LIST_HEAD(&io->list_entry);
+                       list_add_tail(&io->list_entry, &hw->io_free);
+               }
+       }
+
+       dma_free_coherent(&efct->pci->dev, req.size, req.virt, req.phys);
+       memset(&req, 0, sizeof(struct efc_dma));
+       kfree(sgls);
+
+       return rc;
+}
+
+static int
+efct_hw_init_io(struct efct_hw *hw)
+{
+       u32 i, idx = 0;
+       bool prereg = false;
+       struct efct_hw_io *io = NULL;
+       int rc = 0;
+
+       prereg = hw->sli.params.sgl_pre_registered;
+
+       if (prereg)
+               return efct_hw_init_prereg_io(hw);
+
+       for (i = 0; i < hw->config.n_io; i++, idx++) {
+               io = hw->io[idx];
+               io->state = EFCT_HW_IO_STATE_FREE;
+               INIT_LIST_HEAD(&io->list_entry);
+               list_add_tail(&io->list_entry, &hw->io_free);
+       }
+
+       return rc;
+}
+
+static int
+efct_hw_config_set_fdt_xfer_hint(struct efct_hw *hw, u32 fdt_xfer_hint)
+{
+       int rc = 0;
+       u8 buf[SLI4_BMBX_SIZE];
+       struct sli4_rqst_cmn_set_features_set_fdt_xfer_hint param;
+
+       memset(&param, 0, sizeof(param));
+       param.fdt_xfer_hint = cpu_to_le32(fdt_xfer_hint);
+       /* build the set_features command */
+       sli_cmd_common_set_features(&hw->sli, buf,
+               SLI4_SET_FEATURES_SET_FTD_XFER_HINT, sizeof(param), &param);
+
+       rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
+       if (rc)
+               efc_log_warn(hw->os, "set FDT hint %d failed: %d\n",
+                            fdt_xfer_hint, rc);
+       else
+               efc_log_info(hw->os, "Set FTD transfer hint to %d\n",
+                            le32_to_cpu(param.fdt_xfer_hint));
+
+       return rc;
+}
+
+static int
+efct_hw_config_rq(struct efct_hw *hw)
+{
+       u32 min_rq_count, i, rc;
+       struct sli4_cmd_rq_cfg rq_cfg[SLI4_CMD_REG_FCFI_NUM_RQ_CFG];
+       u8 buf[SLI4_BMBX_SIZE];
+
+       efc_log_info(hw->os, "using REG_FCFI standard\n");
+
+       /*
+        * Set the filter match/mask values from hw's
+        * filter_def values
+        */
+       for (i = 0; i < SLI4_CMD_REG_FCFI_NUM_RQ_CFG; i++) {
+               rq_cfg[i].rq_id = cpu_to_le16(0xffff);
+               rq_cfg[i].r_ctl_mask = (u8)hw->config.filter_def[i];
+               rq_cfg[i].r_ctl_match = (u8)(hw->config.filter_def[i] >> 8);
+               rq_cfg[i].type_mask = (u8)(hw->config.filter_def[i] >> 16);
+               rq_cfg[i].type_match = (u8)(hw->config.filter_def[i] >> 24);
+       }
+
+       /*
+        * Update the rq_id's of the FCF configuration
+        * (don't update more than the number of rq_cfg
+        * elements)
+        */
+       min_rq_count = (hw->hw_rq_count < SLI4_CMD_REG_FCFI_NUM_RQ_CFG) ?
+                       hw->hw_rq_count : SLI4_CMD_REG_FCFI_NUM_RQ_CFG;
+       for (i = 0; i < min_rq_count; i++) {
+               struct hw_rq *rq = hw->hw_rq[i];
+               u32 j;
+
+               for (j = 0; j < SLI4_CMD_REG_FCFI_NUM_RQ_CFG; j++) {
+                       u32 mask = (rq->filter_mask != 0) ?
+                               rq->filter_mask : 1;
+
+                       if (!(mask & (1U << j)))
+                               continue;
+
+                       rq_cfg[i].rq_id = cpu_to_le16(rq->hdr->id);
+                       efct_logfcfi(hw, j, i, rq->hdr->id);
+               }
+       }
+
+       rc = -EIO;
+       if (!sli_cmd_reg_fcfi(&hw->sli, buf, 0, rq_cfg))
+               rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
+
+       if (rc != 0) {
+               efc_log_err(hw->os, "FCFI registration failed\n");
+               return rc;
+       }
+       hw->fcf_indicator =
+               le16_to_cpu(((struct sli4_cmd_reg_fcfi *)buf)->fcfi);
+
+       return rc;
+}
+
+static int
+efct_hw_config_mrq(struct efct_hw *hw, u8 mode, u16 fcf_index)
+{
+       u8 buf[SLI4_BMBX_SIZE], mrq_bitmask = 0;
+       struct hw_rq *rq;
+       struct sli4_cmd_reg_fcfi_mrq *rsp = NULL;
+       struct sli4_cmd_rq_cfg rq_filter[SLI4_CMD_REG_FCFI_MRQ_NUM_RQ_CFG];
+       u32 rc, i;
+
+       if (mode == SLI4_CMD_REG_FCFI_SET_FCFI_MODE)
+               goto issue_cmd;
+
+       /* Set the filter match/mask values from hw's filter_def values */
+       for (i = 0; i < SLI4_CMD_REG_FCFI_NUM_RQ_CFG; i++) {
+               rq_filter[i].rq_id = cpu_to_le16(0xffff);
+               rq_filter[i].type_mask = (u8)hw->config.filter_def[i];
+               rq_filter[i].type_match = (u8)(hw->config.filter_def[i] >> 8);
+               rq_filter[i].r_ctl_mask = (u8)(hw->config.filter_def[i] >> 16);
+               rq_filter[i].r_ctl_match = (u8)(hw->config.filter_def[i] >> 24);
+       }
+
+       rq = hw->hw_rq[0];
+       rq_filter[0].rq_id = cpu_to_le16(rq->hdr->id);
+       rq_filter[1].rq_id = cpu_to_le16(rq->hdr->id);
+
+       mrq_bitmask = 0x2;
+issue_cmd:
+       efc_log_debug(hw->os, "Issue reg_fcfi_mrq count:%d policy:%d mode:%d\n",
+                     hw->hw_rq_count, hw->config.rq_selection_policy, mode);
+       /* Invoke REG_FCFI_MRQ */
+       rc = sli_cmd_reg_fcfi_mrq(&hw->sli, buf, mode, fcf_index,
+                                 hw->config.rq_selection_policy, mrq_bitmask,
+                                 hw->hw_mrq_count, rq_filter);
+       if (rc) {
+               efc_log_err(hw->os, "sli_cmd_reg_fcfi_mrq() failed\n");
+               return -EIO;
+       }
+
+       rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
+
+       rsp = (struct sli4_cmd_reg_fcfi_mrq *)buf;
+
+       if ((rc) || (le16_to_cpu(rsp->hdr.status))) {
+               efc_log_err(hw->os, "FCFI MRQ reg failed. cmd=%x status=%x\n",
+                           rsp->hdr.command, le16_to_cpu(rsp->hdr.status));
+               return -EIO;
+       }
+
+       if (mode == SLI4_CMD_REG_FCFI_SET_FCFI_MODE)
+               hw->fcf_indicator = le16_to_cpu(rsp->fcfi);
+
+       return 0;
+}
+
+static void
+efct_hw_queue_hash_add(struct efct_queue_hash *hash,
+                      u16 id, u16 index)
+{
+       u32 hash_index = id & (EFCT_HW_Q_HASH_SIZE - 1);
+
+       /*
+        * Since the hash is always bigger than the number of queues, then we
+        * never have to worry about an infinite loop.
+        */
+       while (hash[hash_index].in_use)
+               hash_index = (hash_index + 1) & (EFCT_HW_Q_HASH_SIZE - 1);
+
+       /* not used, claim the entry */
+       hash[hash_index].id = id;
+       hash[hash_index].in_use = true;
+       hash[hash_index].index = index;
+}
+
+static int
+efct_hw_config_sli_port_health_check(struct efct_hw *hw, u8 query, u8 enable)
+{
+       int rc = 0;
+       u8 buf[SLI4_BMBX_SIZE];
+       struct sli4_rqst_cmn_set_features_health_check param;
+       u32 health_check_flag = 0;
+
+       memset(&param, 0, sizeof(param));
+
+       if (enable)
+               health_check_flag |= SLI4_RQ_HEALTH_CHECK_ENABLE;
+
+       if (query)
+               health_check_flag |= SLI4_RQ_HEALTH_CHECK_QUERY;
+
+       param.health_check_dword = cpu_to_le32(health_check_flag);
+
+       /* build the set_features command */
+       sli_cmd_common_set_features(&hw->sli, buf,
+               SLI4_SET_FEATURES_SLI_PORT_HEALTH_CHECK, sizeof(param), &param);
+
+       rc = efct_hw_command(hw, buf, EFCT_CMD_POLL, NULL, NULL);
+       if (rc)
+               efc_log_err(hw->os, "efct_hw_command returns %d\n", rc);
+       else
+               efc_log_debug(hw->os, "SLI Port Health Check is enabled\n");
+
+       return rc;
+}
+
+int
+efct_hw_init(struct efct_hw *hw)
+{
+       int rc;
+       u32 i = 0;
+       int rem_count;
+       unsigned long flags = 0;
+       struct efct_hw_io *temp;
+       struct efc_dma *dma;
+
+       /*
+        * Make sure the command lists are empty. If this is start-of-day,
+        * they'll be empty since they were just initialized in efct_hw_setup.
+        * If we've just gone through a reset, the command and command pending
+        * lists should have been cleaned up as part of the reset
+        * (efct_hw_reset()).
+        */
+       spin_lock_irqsave(&hw->cmd_lock, flags);
+       if (!list_empty(&hw->cmd_head)) {
+               spin_unlock_irqrestore(&hw->cmd_lock, flags);
+               efc_log_err(hw->os, "command found on cmd list\n");
+               return -EIO;
+       }
+       if (!list_empty(&hw->cmd_pending)) {
+               spin_unlock_irqrestore(&hw->cmd_lock, flags);
+               efc_log_err(hw->os, "command found on pending list\n");
+               return -EIO;
+       }
+       spin_unlock_irqrestore(&hw->cmd_lock, flags);
+
+       /* Free RQ buffers if prevously allocated */
+       efct_hw_rx_free(hw);
+
+       /*
+        * The IO queues must be initialized here for the reset case. The
+        * efct_hw_init_io() function will re-add the IOs to the free list.
+        * The cmd_head list should be OK since we free all entries in
+        * efct_hw_command_cancel() that is called in the efct_hw_reset().
+        */
+
+       /* If we are in this function due to a reset, there may be stale items
+        * on lists that need to be removed.  Clean them up.
+        */
+       rem_count = 0;
+       while ((!list_empty(&hw->io_wait_free))) {
+               rem_count++;
+               temp = list_first_entry(&hw->io_wait_free, struct efct_hw_io,
+                                       list_entry);
+               list_del_init(&temp->list_entry);
+       }
+       if (rem_count > 0)
+               efc_log_debug(hw->os, "rmvd %d items from io_wait_free list\n",
+                             rem_count);
+
+       rem_count = 0;
+       while ((!list_empty(&hw->io_inuse))) {
+               rem_count++;
+               temp = list_first_entry(&hw->io_inuse, struct efct_hw_io,
+                                       list_entry);
+               list_del_init(&temp->list_entry);
+       }
+       if (rem_count > 0)
+               efc_log_debug(hw->os, "rmvd %d items from io_inuse list\n",
+                             rem_count);
+
+       rem_count = 0;
+       while ((!list_empty(&hw->io_free))) {
+               rem_count++;
+               temp = list_first_entry(&hw->io_free, struct efct_hw_io,
+                                       list_entry);
+               list_del_init(&temp->list_entry);
+       }
+       if (rem_count > 0)
+               efc_log_debug(hw->os, "rmvd %d items from io_free list\n",
+                             rem_count);
+
+       /* If MRQ not required, Make sure we dont request feature. */
+       if (hw->config.n_rq == 1)
+               hw->sli.features &= (~SLI4_REQFEAT_MRQP);
+
+       if (sli_init(&hw->sli)) {
+               efc_log_err(hw->os, "SLI failed to initialize\n");
+               return -EIO;
+       }
+
+       if (hw->sliport_healthcheck) {
+               rc = efct_hw_config_sli_port_health_check(hw, 0, 1);
+               if (rc != 0) {
+                       efc_log_err(hw->os, "Enable port Health check fail\n");
+                       return rc;
+               }
+       }
+
+       /*
+        * Set FDT transfer hint, only works on Lancer
+        */
+       if (hw->sli.if_type == SLI4_INTF_IF_TYPE_2) {
+               /*
+                * Non-fatal error. In particular, we can disregard failure to
+                * set EFCT_HW_FDT_XFER_HINT on devices with legacy firmware
+                * that do not support EFCT_HW_FDT_XFER_HINT feature.
+                */
+               efct_hw_config_set_fdt_xfer_hint(hw, EFCT_HW_FDT_XFER_HINT);
+       }
+
+       /* zero the hashes */
+       memset(hw->cq_hash, 0, sizeof(hw->cq_hash));
+       efc_log_debug(hw->os, "Max CQs %d, hash size = %d\n",
+                     EFCT_HW_MAX_NUM_CQ, EFCT_HW_Q_HASH_SIZE);
+
+       memset(hw->rq_hash, 0, sizeof(hw->rq_hash));
+       efc_log_debug(hw->os, "Max RQs %d, hash size = %d\n",
+                     EFCT_HW_MAX_NUM_RQ, EFCT_HW_Q_HASH_SIZE);
+
+       memset(hw->wq_hash, 0, sizeof(hw->wq_hash));
+       efc_log_debug(hw->os, "Max WQs %d, hash size = %d\n",
+                     EFCT_HW_MAX_NUM_WQ, EFCT_HW_Q_HASH_SIZE);
+
+       rc = efct_hw_init_queues(hw);
+       if (rc)
+               return rc;
+
+       rc = efct_hw_map_wq_cpu(hw);
+       if (rc)
+               return rc;
+
+       /* Allocate and p_st RQ buffers */
+       rc = efct_hw_rx_allocate(hw);
+       if (rc) {
+               efc_log_err(hw->os, "rx_allocate failed\n");
+               return rc;
+       }
+
+       rc = efct_hw_rx_post(hw);
+       if (rc) {
+               efc_log_err(hw->os, "WARNING - error posting RQ buffers\n");
+               return rc;
+       }
+
+       if (hw->config.n_eq == 1) {
+               rc = efct_hw_config_rq(hw);
+               if (rc) {
+                       efc_log_err(hw->os, "config rq failed %d\n", rc);
+                       return rc;
+               }
+       } else {
+               rc = efct_hw_config_mrq(hw, SLI4_CMD_REG_FCFI_SET_FCFI_MODE, 0);
+               if (rc != 0) {
+                       efc_log_err(hw->os, "REG_FCFI_MRQ FCFI reg failed\n");
+                       return rc;
+               }
+
+               rc = efct_hw_config_mrq(hw, SLI4_CMD_REG_FCFI_SET_MRQ_MODE, 0);
+               if (rc != 0) {
+                       efc_log_err(hw->os, "REG_FCFI_MRQ MRQ reg failed\n");
+                       return rc;
+               }
+       }
+
+       /*
+        * Allocate the WQ request tag pool, if not previously allocated
+        * (the request tag value is 16 bits, thus the pool allocation size
+        * of 64k)
+        */
+       hw->wq_reqtag_pool = efct_hw_reqtag_pool_alloc(hw);
+       if (!hw->wq_reqtag_pool) {
+               efc_log_err(hw->os, "efct_hw_reqtag_init failed %d\n", rc);
+               return rc;
+       }
+
+       rc = efct_hw_setup_io(hw);
+       if (rc) {
+               efc_log_err(hw->os, "IO allocation failure\n");
+               return rc;
+       }
+
+       rc = efct_hw_init_io(hw);
+       if (rc) {
+               efc_log_err(hw->os, "IO initialization failure\n");
+               return rc;
+       }
+
+       dma = &hw->loop_map;
+       dma->size = SLI4_MIN_LOOP_MAP_BYTES;
+       dma->virt = dma_alloc_coherent(&hw->os->pci->dev, dma->size, &dma->phys,
+                                      GFP_DMA);
+       if (!dma->virt)
+               return -EIO;
+
+       /*
+        * Arming the EQ allows (e.g.) interrupts when CQ completions write EQ
+        * entries
+        */
+       for (i = 0; i < hw->eq_count; i++)
+               sli_queue_arm(&hw->sli, &hw->eq[i], true);
+
+       /*
+        * Initialize RQ hash
+        */
+       for (i = 0; i < hw->rq_count; i++)
+               efct_hw_queue_hash_add(hw->rq_hash, hw->rq[i].id, i);
+
+       /*
+        * Initialize WQ hash
+        */
+       for (i = 0; i < hw->wq_count; i++)
+               efct_hw_queue_hash_add(hw->wq_hash, hw->wq[i].id, i);
+
+       /*
+        * Arming the CQ allows (e.g.) MQ completions to write CQ entries
+        */
+       for (i = 0; i < hw->cq_count; i++) {
+               efct_hw_queue_hash_add(hw->cq_hash, hw->cq[i].id, i);
+               sli_queue_arm(&hw->sli, &hw->cq[i], true);
+       }
+
+       /* Set RQ process limit*/
+       for (i = 0; i < hw->hw_rq_count; i++) {
+               struct hw_rq *rq = hw->hw_rq[i];
+
+               hw->cq[rq->cq->instance].proc_limit = hw->config.n_io / 2;
+       }
+
+       /* record the fact that the queues are functional */
+       hw->state = EFCT_HW_STATE_ACTIVE;
+       /*
+        * Allocate a HW IOs for send frame.
+        */
+       hw->hw_wq[0]->send_frame_io = efct_hw_io_alloc(hw);
+       if (!hw->hw_wq[0]->send_frame_io)
+               efc_log_err(hw->os, "alloc for send_frame_io failed\n");
+
+       /* Initialize send frame sequence id */
+       atomic_set(&hw->send_frame_seq_id, 0);
+
+       return 0;
+}
+
+int
+efct_hw_parse_filter(struct efct_hw *hw, void *value)
+{
+       int rc = 0;
+       char *p = NULL;
+       char *token;
+       u32 idx = 0;
+
+       for (idx = 0; idx < ARRAY_SIZE(hw->config.filter_def); idx++)
+               hw->config.filter_def[idx] = 0;
+
+       p = kstrdup(value, GFP_KERNEL);
+       if (!p || !*p) {
+               efc_log_err(hw->os, "p is NULL\n");
+               return -ENOMEM;
+       }
+
+       idx = 0;
+       while ((token = strsep(&p, ",")) && *token) {
+               if (kstrtou32(token, 0, &hw->config.filter_def[idx++]))
+                       efc_log_err(hw->os, "kstrtoint failed\n");
+
+               if (!p || !*p)
+                       break;
+
+               if (idx == ARRAY_SIZE(hw->config.filter_def))
+                       break;
+       }
+       kfree(p);
+
+       return rc;
+}
+
+u64
+efct_get_wwnn(struct efct_hw *hw)
+{
+       struct sli4 *sli = &hw->sli;
+       u8 p[8];
+
+       memcpy(p, sli->wwnn, sizeof(p));
+       return get_unaligned_be64(p);
+}
+
+u64
+efct_get_wwpn(struct efct_hw *hw)
+{
+       struct sli4 *sli = &hw->sli;
+       u8 p[8];
+
+       memcpy(p, sli->wwpn, sizeof(p));
+       return get_unaligned_be64(p);
+}
index d5f1d8c052091222ace641969541eadb6d69bce6..6c8ed6a59f5ea5d63638c96bd39bac634db35121 100644 (file)
@@ -583,4 +583,19 @@ struct efct_hw_grp_hdr {
        u8                      revision[32];
 };
 
+static inline int
+efct_hw_get_link_speed(struct efct_hw *hw) {
+       return hw->link.speed;
+}
+
+int
+efct_hw_setup(struct efct_hw *hw, void *os, struct pci_dev *pdev);
+int efct_hw_init(struct efct_hw *hw);
+int
+efct_hw_parse_filter(struct efct_hw *hw, void *value);
+uint64_t
+efct_get_wwnn(struct efct_hw *hw);
+uint64_t
+efct_get_wwpn(struct efct_hw *hw);
+
 #endif /* __EFCT_H__ */
diff --git a/drivers/scsi/elx/efct/efct_xport.c b/drivers/scsi/elx/efct/efct_xport.c
new file mode 100644 (file)
index 0000000..58dad8a
--- /dev/null
@@ -0,0 +1,500 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (C) 2021 Broadcom. All Rights Reserved. The term
+ * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.
+ */
+
+#include "efct_driver.h"
+#include "efct_unsol.h"
+
+static struct dentry *efct_debugfs_root;
+static atomic_t efct_debugfs_count;
+
+static struct scsi_host_template efct_template = {
+       .module                 = THIS_MODULE,
+       .name                   = EFCT_DRIVER_NAME,
+       .supported_mode         = MODE_TARGET,
+};
+
+/* globals */
+static struct fc_function_template efct_xport_functions;
+static struct fc_function_template efct_vport_functions;
+
+static struct scsi_transport_template *efct_xport_fc_tt;
+static struct scsi_transport_template *efct_vport_fc_tt;
+
+struct efct_xport *
+efct_xport_alloc(struct efct *efct)
+{
+       struct efct_xport *xport;
+
+       xport = kzalloc(sizeof(*xport), GFP_KERNEL);
+       if (!xport)
+               return xport;
+
+       xport->efct = efct;
+       return xport;
+}
+
+static int
+efct_xport_init_debugfs(struct efct *efct)
+{
+       /* Setup efct debugfs root directory */
+       if (!efct_debugfs_root) {
+               efct_debugfs_root = debugfs_create_dir("efct", NULL);
+               atomic_set(&efct_debugfs_count, 0);
+               if (!efct_debugfs_root) {
+                       efc_log_err(efct, "failed to create debugfs entry\n");
+                       goto debugfs_fail;
+               }
+       }
+
+       /* Create a directory for sessions in root */
+       if (!efct->sess_debugfs_dir) {
+               efct->sess_debugfs_dir = debugfs_create_dir("sessions", NULL);
+               if (!efct->sess_debugfs_dir) {
+                       efc_log_err(efct,
+                                   "failed to create debugfs entry for sessions\n");
+                       goto debugfs_fail;
+               }
+               atomic_inc(&efct_debugfs_count);
+       }
+
+       return 0;
+
+debugfs_fail:
+       return -EIO;
+}
+
+static void efct_xport_delete_debugfs(struct efct *efct)
+{
+       /* Remove session debugfs directory */
+       debugfs_remove(efct->sess_debugfs_dir);
+       efct->sess_debugfs_dir = NULL;
+       atomic_dec(&efct_debugfs_count);
+
+       if (atomic_read(&efct_debugfs_count) == 0) {
+               /* remove root debugfs directory */
+               debugfs_remove(efct_debugfs_root);
+               efct_debugfs_root = NULL;
+       }
+}
+
+int
+efct_xport_attach(struct efct_xport *xport)
+{
+       struct efct *efct = xport->efct;
+       int rc;
+
+       rc = efct_hw_setup(&efct->hw, efct, efct->pci);
+       if (rc) {
+               efc_log_err(efct, "%s: Can't setup hardware\n", efct->desc);
+               return rc;
+       }
+
+       efct_hw_parse_filter(&efct->hw, (void *)efct->filter_def);
+
+       xport->io_pool = efct_io_pool_create(efct, efct->hw.config.n_sgl);
+       if (!xport->io_pool) {
+               efc_log_err(efct, "Can't allocate IO pool\n");
+               return -ENOMEM;
+       }
+
+       return 0;
+}
+
+static void
+efct_xport_link_stats_cb(int status, u32 num_counters,
+                        struct efct_hw_link_stat_counts *counters, void *arg)
+{
+       union efct_xport_stats_u *result = arg;
+
+       result->stats.link_stats.link_failure_error_count =
+               counters[EFCT_HW_LINK_STAT_LINK_FAILURE_COUNT].counter;
+       result->stats.link_stats.loss_of_sync_error_count =
+               counters[EFCT_HW_LINK_STAT_LOSS_OF_SYNC_COUNT].counter;
+       result->stats.link_stats.primitive_sequence_error_count =
+               counters[EFCT_HW_LINK_STAT_PRIMITIVE_SEQ_COUNT].counter;
+       result->stats.link_stats.invalid_transmission_word_error_count =
+               counters[EFCT_HW_LINK_STAT_INVALID_XMIT_WORD_COUNT].counter;
+       result->stats.link_stats.crc_error_count =
+               counters[EFCT_HW_LINK_STAT_CRC_COUNT].counter;
+
+       complete(&result->stats.done);
+}
+
+static void
+efct_xport_host_stats_cb(int status, u32 num_counters,
+                        struct efct_hw_host_stat_counts *counters, void *arg)
+{
+       union efct_xport_stats_u *result = arg;
+
+       result->stats.host_stats.transmit_kbyte_count =
+               counters[EFCT_HW_HOST_STAT_TX_KBYTE_COUNT].counter;
+       result->stats.host_stats.receive_kbyte_count =
+               counters[EFCT_HW_HOST_STAT_RX_KBYTE_COUNT].counter;
+       result->stats.host_stats.transmit_frame_count =
+               counters[EFCT_HW_HOST_STAT_TX_FRAME_COUNT].counter;
+       result->stats.host_stats.receive_frame_count =
+               counters[EFCT_HW_HOST_STAT_RX_FRAME_COUNT].counter;
+
+       complete(&result->stats.done);
+}
+
+static void
+efct_xport_async_link_stats_cb(int status, u32 num_counters,
+                              struct efct_hw_link_stat_counts *counters,
+                              void *arg)
+{
+       union efct_xport_stats_u *result = arg;
+
+       result->stats.link_stats.link_failure_error_count =
+               counters[EFCT_HW_LINK_STAT_LINK_FAILURE_COUNT].counter;
+       result->stats.link_stats.loss_of_sync_error_count =
+               counters[EFCT_HW_LINK_STAT_LOSS_OF_SYNC_COUNT].counter;
+       result->stats.link_stats.primitive_sequence_error_count =
+               counters[EFCT_HW_LINK_STAT_PRIMITIVE_SEQ_COUNT].counter;
+       result->stats.link_stats.invalid_transmission_word_error_count =
+               counters[EFCT_HW_LINK_STAT_INVALID_XMIT_WORD_COUNT].counter;
+       result->stats.link_stats.crc_error_count =
+               counters[EFCT_HW_LINK_STAT_CRC_COUNT].counter;
+}
+
+static void
+efct_xport_async_host_stats_cb(int status, u32 num_counters,
+                              struct efct_hw_host_stat_counts *counters,
+                              void *arg)
+{
+       union efct_xport_stats_u *result = arg;
+
+       result->stats.host_stats.transmit_kbyte_count =
+               counters[EFCT_HW_HOST_STAT_TX_KBYTE_COUNT].counter;
+       result->stats.host_stats.receive_kbyte_count =
+               counters[EFCT_HW_HOST_STAT_RX_KBYTE_COUNT].counter;
+       result->stats.host_stats.transmit_frame_count =
+               counters[EFCT_HW_HOST_STAT_TX_FRAME_COUNT].counter;
+       result->stats.host_stats.receive_frame_count =
+               counters[EFCT_HW_HOST_STAT_RX_FRAME_COUNT].counter;
+}
+
+static void
+efct_xport_config_stats_timer(struct efct *efct);
+
+static void
+efct_xport_stats_timer_cb(struct timer_list *t)
+{
+       struct efct_xport *xport = from_timer(xport, t, stats_timer);
+       struct efct *efct = xport->efct;
+
+       efct_xport_config_stats_timer(efct);
+}
+
+static void
+efct_xport_config_stats_timer(struct efct *efct)
+{
+       u32 timeout = 3 * 1000;
+       struct efct_xport *xport = NULL;
+
+       if (!efct) {
+               pr_err("%s: failed to locate EFCT device\n", __func__);
+               return;
+       }
+
+       xport = efct->xport;
+       efct_hw_get_link_stats(&efct->hw, 0, 0, 0,
+                              efct_xport_async_link_stats_cb,
+                              &xport->fc_xport_stats);
+       efct_hw_get_host_stats(&efct->hw, 0, efct_xport_async_host_stats_cb,
+                              &xport->fc_xport_stats);
+
+       timer_setup(&xport->stats_timer,
+                   &efct_xport_stats_timer_cb, 0);
+       mod_timer(&xport->stats_timer,
+                 jiffies + msecs_to_jiffies(timeout));
+}
+
+int
+efct_xport_initialize(struct efct_xport *xport)
+{
+       struct efct *efct = xport->efct;
+       int rc = 0;
+
+       /* Initialize io lists */
+       spin_lock_init(&xport->io_pending_lock);
+       INIT_LIST_HEAD(&xport->io_pending_list);
+       atomic_set(&xport->io_active_count, 0);
+       atomic_set(&xport->io_pending_count, 0);
+       atomic_set(&xport->io_total_free, 0);
+       atomic_set(&xport->io_total_pending, 0);
+       atomic_set(&xport->io_alloc_failed_count, 0);
+       atomic_set(&xport->io_pending_recursing, 0);
+
+       rc = efct_hw_init(&efct->hw);
+       if (rc) {
+               efc_log_err(efct, "efct_hw_init failure\n");
+               goto out;
+       }
+
+       rc = efct_scsi_tgt_new_device(efct);
+       if (rc) {
+               efc_log_err(efct, "failed to initialize target\n");
+               goto hw_init_out;
+       }
+
+       rc = efct_scsi_new_device(efct);
+       if (rc) {
+               efc_log_err(efct, "failed to initialize initiator\n");
+               goto tgt_dev_out;
+       }
+
+       /* Get FC link and host statistics perodically*/
+       efct_xport_config_stats_timer(efct);
+
+       efct_xport_init_debugfs(efct);
+
+       return rc;
+
+tgt_dev_out:
+       efct_scsi_tgt_del_device(efct);
+
+hw_init_out:
+       efct_hw_teardown(&efct->hw);
+out:
+       return rc;
+}
+
+int
+efct_xport_status(struct efct_xport *xport, enum efct_xport_status cmd,
+                 union efct_xport_stats_u *result)
+{
+       int rc = 0;
+       struct efct *efct = NULL;
+       union efct_xport_stats_u value;
+
+       efct = xport->efct;
+
+       switch (cmd) {
+       case EFCT_XPORT_CONFIG_PORT_STATUS:
+               if (xport->configured_link_state == 0) {
+                       /*
+                        * Initial state is offline. configured_link_state is
+                        * set to online explicitly when port is brought online
+                        */
+                       xport->configured_link_state = EFCT_XPORT_PORT_OFFLINE;
+               }
+               result->value = xport->configured_link_state;
+               break;
+
+       case EFCT_XPORT_PORT_STATUS:
+               /* Determine port status based on link speed. */
+               value.value = efct_hw_get_link_speed(&efct->hw);
+               if (value.value == 0)
+                       result->value = EFCT_XPORT_PORT_OFFLINE;
+               else
+                       result->value = EFCT_XPORT_PORT_ONLINE;
+               break;
+
+       case EFCT_XPORT_LINK_SPEED:
+               result->value = efct_hw_get_link_speed(&efct->hw);
+               break;
+
+       case EFCT_XPORT_LINK_STATISTICS:
+               memcpy((void *)result, &efct->xport->fc_xport_stats,
+                      sizeof(union efct_xport_stats_u));
+               break;
+       case EFCT_XPORT_LINK_STAT_RESET: {
+               /* Create a completion to synchronize the stat reset process */
+               init_completion(&result->stats.done);
+
+               /* First reset the link stats */
+               rc = efct_hw_get_link_stats(&efct->hw, 0, 1, 1,
+                                           efct_xport_link_stats_cb, result);
+               if (rc)
+                       break;
+
+               /* Wait for completion to be signaled when the cmd completes */
+               if (wait_for_completion_interruptible(&result->stats.done)) {
+                       /* Undefined failure */
+                       efc_log_debug(efct, "sem wait failed\n");
+                       rc = -EIO;
+                       break;
+               }
+
+               /* Next reset the host stats */
+               rc = efct_hw_get_host_stats(&efct->hw, 1,
+                                           efct_xport_host_stats_cb, result);
+
+               if (rc)
+                       break;
+
+               /* Wait for completion to be signaled when the cmd completes */
+               if (wait_for_completion_interruptible(&result->stats.done)) {
+                       /* Undefined failure */
+                       efc_log_debug(efct, "sem wait failed\n");
+                       rc = -EIO;
+                       break;
+               }
+               break;
+       }
+       default:
+               rc = -EIO;
+               break;
+       }
+
+       return rc;
+}
+
+static int
+efct_get_link_supported_speeds(struct efct *efct)
+{
+       u32 supported_speeds = 0;
+       u32 link_module_type, i;
+       struct {
+               u32 lmt_speed;
+               u32 speed;
+       } supported_speed_list[] = {
+               {SLI4_LINK_MODULE_TYPE_1GB, FC_PORTSPEED_1GBIT},
+               {SLI4_LINK_MODULE_TYPE_2GB, FC_PORTSPEED_2GBIT},
+               {SLI4_LINK_MODULE_TYPE_4GB, FC_PORTSPEED_4GBIT},
+               {SLI4_LINK_MODULE_TYPE_8GB, FC_PORTSPEED_8GBIT},
+               {SLI4_LINK_MODULE_TYPE_16GB, FC_PORTSPEED_16GBIT},
+               {SLI4_LINK_MODULE_TYPE_32GB, FC_PORTSPEED_32GBIT},
+               {SLI4_LINK_MODULE_TYPE_64GB, FC_PORTSPEED_64GBIT},
+               {SLI4_LINK_MODULE_TYPE_128GB, FC_PORTSPEED_128GBIT},
+       };
+
+       link_module_type = sli_get_lmt(&efct->hw.sli);
+
+       /* populate link supported speeds */
+       for (i = 0; i < ARRAY_SIZE(supported_speed_list); i++) {
+               if (link_module_type & supported_speed_list[i].lmt_speed)
+                       supported_speeds |= supported_speed_list[i].speed;
+       }
+
+       return supported_speeds;
+}
+
+int
+efct_scsi_new_device(struct efct *efct)
+{
+       struct Scsi_Host *shost = NULL;
+       int error = 0;
+       struct efct_vport *vport = NULL;
+
+       shost = scsi_host_alloc(&efct_template, sizeof(*vport));
+       if (!shost) {
+               efc_log_err(efct, "failed to allocate Scsi_Host struct\n");
+               return -ENOMEM;
+       }
+
+       /* save shost to initiator-client context */
+       efct->shost = shost;
+
+       /* save efct information to shost LLD-specific space */
+       vport = (struct efct_vport *)shost->hostdata;
+       vport->efct = efct;
+
+       /*
+        * Set initial can_queue value to the max SCSI IOs. This is the maximum
+        * global queue depth (as opposed to the per-LUN queue depth --
+        * .cmd_per_lun This may need to be adjusted for I+T mode.
+        */
+       shost->can_queue = efct->hw.config.n_io;
+       shost->max_cmd_len = 16; /* 16-byte CDBs */
+       shost->max_id = 0xffff;
+       shost->max_lun = 0xffffffff;
+
+       /*
+        * can only accept (from mid-layer) as many SGEs as we've
+        * pre-registered
+        */
+       shost->sg_tablesize = sli_get_max_sgl(&efct->hw.sli);
+
+       /* attach FC Transport template to shost */
+       shost->transportt = efct_xport_fc_tt;
+       efc_log_debug(efct, "transport template=%p\n", efct_xport_fc_tt);
+
+       /* get pci_dev structure and add host to SCSI ML */
+       error = scsi_add_host_with_dma(shost, &efct->pci->dev,
+                                      &efct->pci->dev);
+       if (error) {
+               efc_log_debug(efct, "failed scsi_add_host_with_dma\n");
+               return -EIO;
+       }
+
+       /* Set symbolic name for host port */
+       snprintf(fc_host_symbolic_name(shost),
+                sizeof(fc_host_symbolic_name(shost)),
+                    "Emulex %s FV%s DV%s", efct->model,
+                    efct->hw.sli.fw_name[0], EFCT_DRIVER_VERSION);
+
+       /* Set host port supported classes */
+       fc_host_supported_classes(shost) = FC_COS_CLASS3;
+
+       fc_host_supported_speeds(shost) = efct_get_link_supported_speeds(efct);
+
+       fc_host_node_name(shost) = efct_get_wwnn(&efct->hw);
+       fc_host_port_name(shost) = efct_get_wwpn(&efct->hw);
+       fc_host_max_npiv_vports(shost) = 128;
+
+       return 0;
+}
+
+struct scsi_transport_template *
+efct_attach_fc_transport(void)
+{
+       struct scsi_transport_template *efct_fc_template = NULL;
+
+       efct_fc_template = fc_attach_transport(&efct_xport_functions);
+
+       if (!efct_fc_template)
+               pr_err("failed to attach EFCT with fc transport\n");
+
+       return efct_fc_template;
+}
+
+struct scsi_transport_template *
+efct_attach_vport_fc_transport(void)
+{
+       struct scsi_transport_template *efct_fc_template = NULL;
+
+       efct_fc_template = fc_attach_transport(&efct_vport_functions);
+
+       if (!efct_fc_template)
+               pr_err("failed to attach EFCT with fc transport\n");
+
+       return efct_fc_template;
+}
+
+int
+efct_scsi_reg_fc_transport(void)
+{
+       /* attach to appropriate scsi_tranport_* module */
+       efct_xport_fc_tt = efct_attach_fc_transport();
+       if (!efct_xport_fc_tt) {
+               pr_err("%s: failed to attach to scsi_transport_*", __func__);
+               return -EIO;
+       }
+
+       efct_vport_fc_tt = efct_attach_vport_fc_transport();
+       if (!efct_vport_fc_tt) {
+               pr_err("%s: failed to attach to scsi_transport_*", __func__);
+               efct_release_fc_transport(efct_xport_fc_tt);
+               efct_xport_fc_tt = NULL;
+               return -EIO;
+       }
+
+       return 0;
+}
+
+void
+efct_scsi_release_fc_transport(void)
+{
+       /* detach from scsi_transport_* */
+       efct_release_fc_transport(efct_xport_fc_tt);
+       efct_xport_fc_tt = NULL;
+       if (efct_vport_fc_tt)
+               efct_release_fc_transport(efct_vport_fc_tt);
+
+       efct_vport_fc_tt = NULL;
+}
diff --git a/drivers/scsi/elx/efct/efct_xport.h b/drivers/scsi/elx/efct/efct_xport.h
new file mode 100644 (file)
index 0000000..89f3c20
--- /dev/null
@@ -0,0 +1,186 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2021 Broadcom. All Rights Reserved. The term
+ * “Broadcom” refers to Broadcom Inc. and/or its subsidiaries.
+ */
+
+#if !defined(__EFCT_XPORT_H__)
+#define __EFCT_XPORT_H__
+
+enum efct_xport_ctrl {
+       EFCT_XPORT_PORT_ONLINE = 1,
+       EFCT_XPORT_PORT_OFFLINE,
+       EFCT_XPORT_SHUTDOWN,
+       EFCT_XPORT_POST_NODE_EVENT,
+       EFCT_XPORT_WWNN_SET,
+       EFCT_XPORT_WWPN_SET,
+};
+
+enum efct_xport_status {
+       EFCT_XPORT_PORT_STATUS,
+       EFCT_XPORT_CONFIG_PORT_STATUS,
+       EFCT_XPORT_LINK_SPEED,
+       EFCT_XPORT_IS_SUPPORTED_LINK_SPEED,
+       EFCT_XPORT_LINK_STATISTICS,
+       EFCT_XPORT_LINK_STAT_RESET,
+       EFCT_XPORT_IS_QUIESCED
+};
+
+struct efct_xport_link_stats {
+       bool            rec;
+       bool            gec;
+       bool            w02of;
+       bool            w03of;
+       bool            w04of;
+       bool            w05of;
+       bool            w06of;
+       bool            w07of;
+       bool            w08of;
+       bool            w09of;
+       bool            w10of;
+       bool            w11of;
+       bool            w12of;
+       bool            w13of;
+       bool            w14of;
+       bool            w15of;
+       bool            w16of;
+       bool            w17of;
+       bool            w18of;
+       bool            w19of;
+       bool            w20of;
+       bool            w21of;
+       bool            clrc;
+       bool            clof1;
+       u32             link_failure_error_count;
+       u32             loss_of_sync_error_count;
+       u32             loss_of_signal_error_count;
+       u32             primitive_sequence_error_count;
+       u32             invalid_transmission_word_error_count;
+       u32             crc_error_count;
+       u32             primitive_sequence_event_timeout_count;
+       u32             elastic_buffer_overrun_error_count;
+       u32             arbitration_fc_al_timeout_count;
+       u32             advertised_receive_bufftor_to_buffer_credit;
+       u32             current_receive_buffer_to_buffer_credit;
+       u32             advertised_transmit_buffer_to_buffer_credit;
+       u32             current_transmit_buffer_to_buffer_credit;
+       u32             received_eofa_count;
+       u32             received_eofdti_count;
+       u32             received_eofni_count;
+       u32             received_soff_count;
+       u32             received_dropped_no_aer_count;
+       u32             received_dropped_no_available_rpi_resources_count;
+       u32             received_dropped_no_available_xri_resources_count;
+};
+
+struct efct_xport_host_stats {
+       bool            cc;
+       u32             transmit_kbyte_count;
+       u32             receive_kbyte_count;
+       u32             transmit_frame_count;
+       u32             receive_frame_count;
+       u32             transmit_sequence_count;
+       u32             receive_sequence_count;
+       u32             total_exchanges_originator;
+       u32             total_exchanges_responder;
+       u32             receive_p_bsy_count;
+       u32             receive_f_bsy_count;
+       u32             dropped_frames_due_to_no_rq_buffer_count;
+       u32             empty_rq_timeout_count;
+       u32             dropped_frames_due_to_no_xri_count;
+       u32             empty_xri_pool_count;
+};
+
+struct efct_xport_host_statistics {
+       struct completion               done;
+       struct efct_xport_link_stats    link_stats;
+       struct efct_xport_host_stats    host_stats;
+};
+
+union efct_xport_stats_u {
+       u32     value;
+       struct efct_xport_host_statistics stats;
+};
+
+struct efct_xport_fcp_stats {
+       u64             input_bytes;
+       u64             output_bytes;
+       u64             input_requests;
+       u64             output_requests;
+       u64             control_requests;
+};
+
+struct efct_xport {
+       struct efct             *efct;
+       /* wwpn requested by user for primary nport */
+       u64                     req_wwpn;
+       /* wwnn requested by user for primary nport */
+       u64                     req_wwnn;
+
+       /* Nodes */
+       /* number of allocated nodes */
+       u32                     nodes_count;
+       /* used to track how often IO pool is empty */
+       atomic_t                io_alloc_failed_count;
+       /* array of pointers to nodes */
+       struct efc_node         **nodes;
+
+       /* Io pool and counts */
+       /* pointer to IO pool */
+       struct efct_io_pool     *io_pool;
+       /* lock for io_pending_list */
+       spinlock_t              io_pending_lock;
+       /* list of IOs waiting for HW resources
+        *  lock: xport->io_pending_lock
+        *  link: efct_io_s->io_pending_link
+        */
+       struct list_head        io_pending_list;
+       /* count of totals IOS allocated */
+       atomic_t                io_total_alloc;
+       /* count of totals IOS free'd */
+       atomic_t                io_total_free;
+       /* count of totals IOS that were pended */
+       atomic_t                io_total_pending;
+       /* count of active IOS */
+       atomic_t                io_active_count;
+       /* count of pending IOS */
+       atomic_t                io_pending_count;
+       /* non-zero if efct_scsi_check_pending is executing */
+       atomic_t                io_pending_recursing;
+
+       /* Port */
+       /* requested link state */
+       u32                     configured_link_state;
+
+       /* Timer for Statistics */
+       struct timer_list       stats_timer;
+       union efct_xport_stats_u fc_xport_stats;
+       struct efct_xport_fcp_stats fcp_stats;
+};
+
+struct efct_rport_data {
+       struct efc_node         *node;
+};
+
+struct efct_xport *
+efct_xport_alloc(struct efct *efct);
+int
+efct_xport_attach(struct efct_xport *xport);
+int
+efct_xport_initialize(struct efct_xport *xport);
+void
+efct_xport_detach(struct efct_xport *xport);
+int
+efct_xport_control(struct efct_xport *xport, enum efct_xport_ctrl cmd, ...);
+int
+efct_xport_status(struct efct_xport *xport, enum efct_xport_status cmd,
+                 union efct_xport_stats_u *result);
+void
+efct_xport_free(struct efct_xport *xport);
+
+struct scsi_transport_template *efct_attach_fc_transport(void);
+struct scsi_transport_template *efct_attach_vport_fc_transport(void);
+void
+efct_release_fc_transport(struct scsi_transport_template *transport_template);
+
+#endif /* __EFCT_XPORT_H__ */