| Message ID | 1550763193-14128-11-git-send-email-galpress@amazon.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | RDMA/efa: Elastic Fabric Adapter (EFA) driver | expand |
On Thu, Feb 21, 2019 at 05:33:12PM +0200, Gal Pressman wrote: > Add the main EFA module file which takes care of device > probe/initialization/registration/etc. > > Signed-off-by: Gal Pressman <galpress@amazon.com> > --- > drivers/infiniband/hw/efa/efa_main.c | 577 +++++++++++++++++++++++++++++++++++ > 1 file changed, 577 insertions(+) > create mode 100644 drivers/infiniband/hw/efa/efa_main.c > > diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c > new file mode 100644 > index 000000000000..a4ccebc9665d > --- /dev/null > +++ b/drivers/infiniband/hw/efa/efa_main.c > @@ -0,0 +1,577 @@ > +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause Are you sure that this is the license and Linux-OpenIB? > +/* > + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. > + */ > + > +#include <linux/module.h> > +#include <linux/pci.h> > + > +#include <rdma/ib_user_verbs.h> > + > +#include "efa.h" > + > +#define PCI_DEV_ID_EFA_VF 0xefa0 > + > +static const struct pci_device_id efa_pci_tbl[] = { > + { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA_VF) }, > + { } > +}; > + > +MODULE_AUTHOR("Amazon.com, Inc. or its affiliates"); > +MODULE_LICENSE("Dual BSD/GPL"); > +MODULE_DESCRIPTION(DEVICE_NAME); > +MODULE_DEVICE_TABLE(pci, efa_pci_tbl); > + > +#define EFA_REG_BAR 0 > +#define EFA_MEM_BAR 2 > +#define EFA_BASE_BAR_MASK (BIT(EFA_REG_BAR) | BIT(EFA_MEM_BAR)) > + > +#define EFA_AENQ_ENABLED_GROUPS \ > + (BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \ > + BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE)) > + > +static void efa_update_network_attr(struct efa_dev *dev, > + struct efa_com_get_network_attr_result *network_attr) > +{ > + memcpy(dev->addr, network_attr->addr, sizeof(network_attr->addr)); > + dev->mtu = network_attr->mtu; > + > + efa_dbg(&dev->pdev->dev, "Full address %pI6\n", dev->addr); > +} > + > +/* This handler will called for unknown event group or unimplemented handlers */ > +static void unimplemented_aenq_handler(void *data, > + struct efa_admin_aenq_entry *aenq_e) > +{ > + struct efa_dev *dev = (struct efa_dev *)data; > + > + efa_err_rl(&dev->ibdev.dev, > + "Unknown event was received or event with unimplemented handler\n"); > +} > + > +static void efa_keep_alive(void *data, struct efa_admin_aenq_entry *aenq_e) > +{ > + struct efa_dev *dev = (struct efa_dev *)data; > + > + efa_stat_inc(dev, dev->stats.keep_alive_rcvd); > +} > + > +static struct efa_aenq_handlers aenq_handlers = { > + .handlers = { > + [EFA_ADMIN_KEEP_ALIVE] = efa_keep_alive, > + }, > + .unimplemented_handler = unimplemented_aenq_handler > +}; > + > +static void efa_release_bars(struct efa_dev *dev, int bars_mask) > +{ > + struct pci_dev *pdev = dev->pdev; > + int release_bars; > + > + release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & bars_mask; > + pci_release_selected_regions(pdev, release_bars); > +} > + > +static irqreturn_t efa_intr_msix_mgmnt(int irq, void *data) > +{ > + struct efa_dev *dev = (struct efa_dev *)data; > + > + efa_com_admin_q_comp_intr_handler(dev->edev); > + > + /* Don't call the aenq handler before probe is done */ > + if (likely(test_bit(EFA_DEVICE_RUNNING_BIT, &dev->state))) > + efa_com_aenq_intr_handler(dev->edev, data); > + > + return IRQ_HANDLED; > +} > + > +static int efa_request_mgmnt_irq(struct efa_dev *dev) > +{ > + struct efa_irq *irq; > + int err; > + > + irq = &dev->admin_irq; > + err = request_irq(irq->vector, irq->handler, 0, irq->name, > + irq->data); > + if (err) { > + efa_err(&dev->pdev->dev, "Failed to request admin irq (%d)\n", > + err); > + return err; > + } > + > + efa_dbg(&dev->pdev->dev, "Set affinity hint of mgmnt irq to %*pbl (irq vector: %d)\n", > + nr_cpumask_bits, &irq->affinity_hint_mask, irq->vector); > + irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask); > + > + return err; > +} > + > +static void efa_setup_mgmnt_irq(struct efa_dev *dev) > +{ > + u32 cpu; > + > + snprintf(dev->admin_irq.name, EFA_IRQNAME_SIZE, > + "efa-mgmnt@pci:%s", pci_name(dev->pdev)); > + dev->admin_irq.handler = efa_intr_msix_mgmnt; > + dev->admin_irq.data = dev; > + dev->admin_irq.vector = > + pci_irq_vector(dev->pdev, dev->admin_msix_vector_idx); > + cpu = cpumask_first(cpu_online_mask); > + dev->admin_irq.cpu = cpu; > + cpumask_set_cpu(cpu, > + &dev->admin_irq.affinity_hint_mask); > + efa_info(&dev->pdev->dev, "Setup irq:%p vector:%d name:%s\n", > + &dev->admin_irq, > + dev->admin_irq.vector, > + dev->admin_irq.name); > +} > + > +static void efa_free_mgmnt_irq(struct efa_dev *dev) > +{ > + struct efa_irq *irq; > + > + irq = &dev->admin_irq; > + irq_set_affinity_hint(irq->vector, NULL); > + free_irq(irq->vector, irq->data); > +} > + > +static int efa_set_mgmnt_irq(struct efa_dev *dev) > +{ > + efa_setup_mgmnt_irq(dev); > + > + return efa_request_mgmnt_irq(dev); > +} > + > +static int efa_request_doorbell_bar(struct efa_dev *dev) > +{ > + u8 db_bar_idx = dev->dev_attr.db_bar; > + struct pci_dev *pdev = dev->pdev; > + int bars; > + int err; > + > + if (!(BIT(db_bar_idx) & EFA_BASE_BAR_MASK)) { > + bars = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(db_bar_idx); > + > + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); > + if (err) { > + efa_err(&dev->pdev->dev, > + "pci_request_selected_regions for bar %d failed %d\n", > + db_bar_idx, err); > + return err; > + } > + } > + > + dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx); > + dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx); > + > + return 0; > +} > + > +static void efa_release_doorbell_bar(struct efa_dev *dev) > +{ > + if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK)) > + efa_release_bars(dev, BIT(dev->dev_attr.db_bar)); > +} > + > +static void efa_update_hw_hints(struct efa_dev *dev, > + struct efa_com_get_hw_hints_result *hw_hints) > +{ > + struct efa_com_dev *edev = dev->edev; > + > + if (hw_hints->mmio_read_timeout) > + edev->mmio_read.mmio_read_timeout = > + hw_hints->mmio_read_timeout * 1000; > + > + if (hw_hints->poll_interval) > + edev->aq.poll_interval = hw_hints->poll_interval; > + > + if (hw_hints->admin_completion_timeout) > + edev->aq.completion_timeout = > + hw_hints->admin_completion_timeout; > +} > + > +static const struct ib_device_ops efa_dev_ops = { > + .alloc_pd = efa_alloc_pd, > + .alloc_ucontext = efa_alloc_ucontext, > + .create_ah = efa_create_ah, > + .create_cq = efa_create_cq, > + .create_qp = efa_create_qp, > + .dealloc_pd = efa_dealloc_pd, > + .dealloc_ucontext = efa_dealloc_ucontext, > + .dereg_mr = efa_dereg_mr, > + .destroy_ah = efa_destroy_ah, > + .destroy_cq = efa_destroy_cq, > + .destroy_qp = efa_destroy_qp, > + .get_link_layer = efa_port_link_layer, > + .get_port_immutable = efa_get_port_immutable, > + .mmap = efa_mmap, > + .modify_qp = efa_modify_qp, > + .query_device = efa_query_device, > + .query_gid = efa_query_gid, > + .query_pkey = efa_query_pkey, > + .query_port = efa_query_port, > + .query_qp = efa_query_qp, > + .reg_user_mr = efa_reg_mr, > + INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd), ucontext needs this INIT_RDMA_OBJ_SIZE() too. > +}; > + > +static int efa_ib_device_add(struct efa_dev *dev) > +{ > + struct efa_com_get_network_attr_result network_attr; > + struct efa_com_get_hw_hints_result hw_hints; > + struct pci_dev *pdev = dev->pdev; > + int err; > + > + spin_lock_init(&dev->stats_lock); > + > + err = efa_com_get_device_attr(dev->edev, &dev->dev_attr); > + if (err) > + return err; > + > + efa_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n", dev->dev_attr.db_bar); > + err = efa_request_doorbell_bar(dev); > + if (err) > + return err; > + > + err = efa_com_get_network_attr(dev->edev, &network_attr); > + if (err) > + goto err_release_doorbell_bar; > + > + efa_update_network_attr(dev, &network_attr); > + > + err = efa_com_get_hw_hints(dev->edev, &hw_hints); > + if (err) > + goto err_release_doorbell_bar; > + > + efa_update_hw_hints(dev, &hw_hints); > + > + /* Try to enable all the available aenq groups */ > + err = efa_com_set_aenq_config(dev->edev, EFA_AENQ_ENABLED_GROUPS); > + if (err) > + goto err_release_doorbell_bar; > + > + dev->ibdev.owner = THIS_MODULE; > + dev->ibdev.node_type = RDMA_NODE_EFA; > + dev->ibdev.phys_port_cnt = 1; > + dev->ibdev.num_comp_vectors = 1; > + dev->ibdev.dev.parent = &pdev->dev; > + dev->ibdev.uverbs_abi_ver = EFA_UVERBS_ABI_VERSION; > + > + dev->ibdev.uverbs_cmd_mask = > + (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | > + (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) | > + (1ull << IB_USER_VERBS_CMD_QUERY_PORT) | > + (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | > + (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | > + (1ull << IB_USER_VERBS_CMD_REG_MR) | > + (1ull << IB_USER_VERBS_CMD_DEREG_MR) | > + (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | > + (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | > + (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) | > + (1ull << IB_USER_VERBS_CMD_CREATE_QP) | > + (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | > + (1ull << IB_USER_VERBS_CMD_QUERY_QP) | > + (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | > + (1ull << IB_USER_VERBS_CMD_CREATE_AH) | > + (1ull << IB_USER_VERBS_CMD_DESTROY_AH); > + > + dev->ibdev.uverbs_ex_cmd_mask = > + (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE); > + > + dev->ibdev.driver_id = RDMA_DRIVER_EFA; > + ib_set_device_ops(&dev->ibdev, &efa_dev_ops); > + > + err = ib_register_device(&dev->ibdev, "efa_%d"); > + if (err) > + goto err_release_doorbell_bar; > + > + efa_info(&dev->ibdev.dev, "IB device registered\n"); > + > + set_bit(EFA_DEVICE_RUNNING_BIT, &dev->state); > + > + return 0; > + > +err_release_doorbell_bar: > + efa_release_doorbell_bar(dev); > + return err; > +} > + > +static void efa_ib_device_remove(struct efa_dev *dev) > +{ > + > + /* Reset the device only if the device is running. */ > + if (test_bit(EFA_DEVICE_RUNNING_BIT, &dev->state)) > + efa_com_dev_reset(dev->edev, EFA_REGS_RESET_NORMAL); > + > + efa_info(&dev->ibdev.dev, "Unregister ib device\n"); > + ib_unregister_device(&dev->ibdev); > + efa_release_doorbell_bar(dev); > +} > + > +static void efa_disable_msix(struct efa_dev *dev) > +{ > + if (test_and_clear_bit(EFA_MSIX_ENABLED_BIT, &dev->state)) > + pci_free_irq_vectors(dev->pdev); > +} > + > +static int efa_enable_msix(struct efa_dev *dev) > +{ > + int msix_vecs, irq_num; > + > + if (test_bit(EFA_MSIX_ENABLED_BIT, &dev->state)) { > + efa_err(&dev->pdev->dev, "Error, MSI-X is already enabled\n"); > + return -EPERM; > + } > + > + /* Reserve the max msix vectors we might need */ > + msix_vecs = EFA_NUM_MSIX_VEC; > + efa_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n", > + msix_vecs); > + > + dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX; > + irq_num = pci_alloc_irq_vectors(dev->pdev, msix_vecs, > + msix_vecs, PCI_IRQ_MSIX); > + > + if (irq_num < 0) { > + efa_err(&dev->pdev->dev, "Failed to enable MSI-X. irq_num %d\n", > + irq_num); > + return -ENOSPC; > + } > + > + if (irq_num != msix_vecs) { > + efa_warn(&dev->pdev->dev, > + "Allocated %d MSI-X (out of %d requested)\n", > + irq_num, msix_vecs); > + return -ENOSPC; > + } > + > + set_bit(EFA_MSIX_ENABLED_BIT, &dev->state); > + > + return 0; > +} > + > +static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev) > +{ > + int dma_width; > + int err; > + > + err = efa_com_dev_reset(edev, EFA_REGS_RESET_NORMAL); > + if (err) > + return err; > + > + err = efa_com_validate_version(edev); > + if (err) > + return err; > + > + dma_width = efa_com_get_dma_width(edev); > + if (dma_width < 0) { > + err = dma_width; > + return err; > + } > + > + err = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width)); > + if (err) { > + efa_err(&pdev->dev, "pci_set_dma_mask failed 0x%x\n", err); > + return err; > + } > + > + err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width)); > + if (err) { > + efa_err(&pdev->dev, > + "err_pci_set_consistent_dma_mask failed 0x%x\n", > + err); > + return err; > + } > + > + return 0; > +} > + > +static int efa_probe_device(struct pci_dev *pdev) > +{ > + struct efa_com_dev *edev; > + struct efa_dev *dev; > + int bars; > + int err; > + > + err = pci_enable_device_mem(pdev); > + if (err) { > + efa_err(&pdev->dev, "pci_enable_device_mem() failed!\n"); > + return err; > + } > + > + pci_set_master(pdev); > + > + dev = ib_alloc_device(efa_dev, ibdev); It return or NULL or real pointer, > + if (IS_ERR_OR_NULL(dev)) { > + efa_err(&pdev->dev, "Device alloc failed\n"); > + err = dev ? PTR_ERR(dev) : -ENOMEM; > + goto err_disable_device; > + } > + > + edev = kzalloc(sizeof(*edev), GFP_KERNEL); > + if (!edev) { > + err = -ENOMEM; > + goto err_ibdev_destroy; > + } > + > + pci_set_drvdata(pdev, dev); > + edev->dmadev = &pdev->dev; > + dev->edev = edev; > + dev->pdev = pdev; > + > + bars = pci_select_bars(pdev, IORESOURCE_MEM) & EFA_BASE_BAR_MASK; > + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); > + if (err) { > + efa_err(&pdev->dev, "pci_request_selected_regions failed %d\n", > + err); > + goto err_free_efa_dev; > + } > + > + dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR); > + dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR); > + dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR); > + dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR); > + > + edev->reg_bar = devm_ioremap(&pdev->dev, > + dev->reg_bar_addr, > + dev->reg_bar_len); > + if (!edev->reg_bar) { > + efa_err(&pdev->dev, "Failed to remap register bar\n"); > + err = -EFAULT; > + goto err_release_bars; > + } > + > + err = efa_com_mmio_reg_read_init(edev); > + if (err) { > + efa_err(&pdev->dev, "Failed to init readless MMIO\n"); > + goto err_iounmap; > + } > + > + err = efa_device_init(edev, pdev); > + if (err) { > + efa_err(&pdev->dev, "EFA device init failed\n"); > + if (err == -ETIME) > + err = -EPROBE_DEFER; > + goto err_reg_read_destroy; > + } > + > + err = efa_enable_msix(dev); > + if (err) > + goto err_reg_read_destroy; > + > + edev->aq.msix_vector_idx = dev->admin_msix_vector_idx; > + edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx; > + > + err = efa_set_mgmnt_irq(dev); > + if (err) > + goto err_disable_msix; > + > + err = efa_com_admin_init(edev, &aenq_handlers); > + if (err) > + goto err_free_mgmnt_irq; > + > + return 0; > + > +err_free_mgmnt_irq: > + efa_free_mgmnt_irq(dev); > +err_disable_msix: > + efa_disable_msix(dev); > +err_reg_read_destroy: > + efa_com_mmio_reg_read_destroy(edev); > +err_iounmap: > + devm_iounmap(&pdev->dev, edev->reg_bar); > +err_release_bars: > + efa_release_bars(dev, EFA_BASE_BAR_MASK); > +err_free_efa_dev: > + kfree(edev); > +err_ibdev_destroy: > + ib_dealloc_device(&dev->ibdev); > +err_disable_device: > + pci_disable_device(pdev); > + return err; > +} > + > +static void efa_remove_device(struct pci_dev *pdev) > +{ > + struct efa_dev *dev = pci_get_drvdata(pdev); > + struct efa_com_dev *edev; > + > + if (!dev) > + /* > + * This device didn't load properly and its resources > + * already released, nothing to do > + */ How can efa_remove_device() be called if it didn't success efa_add_device? > + return; > + > + edev = dev->edev; > + > + efa_com_admin_destroy(edev); > + efa_free_mgmnt_irq(dev); > + efa_disable_msix(dev); > + efa_com_mmio_reg_read_destroy(edev); > + devm_iounmap(&pdev->dev, edev->reg_bar); > + efa_release_bars(dev, EFA_BASE_BAR_MASK); > + kfree(edev); > + ib_dealloc_device(&dev->ibdev); > + pci_disable_device(pdev); > +} > + > +static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent) > +{ > + struct efa_dev *dev; > + int err; > + > + err = efa_probe_device(pdev); > + if (err) > + return err; > + > + dev = pci_get_drvdata(pdev); > + err = efa_ib_device_add(dev); > + if (err) > + goto err_remove_device; > + > + return 0; > + > +err_remove_device: > + efa_remove_device(pdev); > + return err; > +} > + > +static void efa_remove(struct pci_dev *pdev) > +{ > + struct efa_dev *dev = (struct efa_dev *)pci_get_drvdata(pdev); > + > + efa_ib_device_remove(dev); > + efa_remove_device(pdev); > +} > + > +static struct pci_driver efa_pci_driver = { > + .name = DRV_MODULE_NAME, > + .id_table = efa_pci_tbl, > + .probe = efa_probe, > + .remove = efa_remove, > +}; > + > +static int __init efa_init(void) > +{ > + int err; > + > + err = pci_register_driver(&efa_pci_driver); > + if (err) { > + pr_err("Couldn't register efa driver\n"); > + goto err_register; > + } > + > + return 0; > + > +err_register: > + return err; > +} > + > +static void __exit efa_exit(void) > +{ > + pci_unregister_driver(&efa_pci_driver); > +} > + > +module_init(efa_init); > +module_exit(efa_exit); > -- > 2.7.4 >
On 2/21/2019 9:33 AM, Gal Pressman wrote: > Add the main EFA module file which takes care of device > probe/initialization/registration/etc. > > Signed-off-by: Gal Pressman <galpress@amazon.com> > --- > drivers/infiniband/hw/efa/efa_main.c | 577 +++++++++++++++++++++++++++++++++++ > 1 file changed, 577 insertions(+) > create mode 100644 drivers/infiniband/hw/efa/efa_main.c > > diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c > new file mode 100644 > index 000000000000..a4ccebc9665d > --- /dev/null > +++ b/drivers/infiniband/hw/efa/efa_main.c > @@ -0,0 +1,577 @@ > +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause > +/* > + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. > + */ > + > +#include <linux/module.h> > +#include <linux/pci.h> > + > +#include <rdma/ib_user_verbs.h> > + > +#include "efa.h" > + > +#define PCI_DEV_ID_EFA_VF 0xefa0 > + > +static const struct pci_device_id efa_pci_tbl[] = { > + { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA_VF) }, > + { } > +}; > + > +MODULE_AUTHOR("Amazon.com, Inc. or its affiliates"); > +MODULE_LICENSE("Dual BSD/GPL"); > +MODULE_DESCRIPTION(DEVICE_NAME); > +MODULE_DEVICE_TABLE(pci, efa_pci_tbl); > + > +#define EFA_REG_BAR 0 > +#define EFA_MEM_BAR 2 > +#define EFA_BASE_BAR_MASK (BIT(EFA_REG_BAR) | BIT(EFA_MEM_BAR)) > + > +#define EFA_AENQ_ENABLED_GROUPS \ > + (BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \ > + BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE)) > + > +static void efa_update_network_attr(struct efa_dev *dev, > + struct efa_com_get_network_attr_result *network_attr) > +{ > + memcpy(dev->addr, network_attr->addr, sizeof(network_attr->addr)); > + dev->mtu = network_attr->mtu; > + > + efa_dbg(&dev->pdev->dev, "Full address %pI6\n", dev->addr); > +} > + > +/* This handler will called for unknown event group or unimplemented handlers */ > +static void unimplemented_aenq_handler(void *data, > + struct efa_admin_aenq_entry *aenq_e) > +{ > + struct efa_dev *dev = (struct efa_dev *)data; > + > + efa_err_rl(&dev->ibdev.dev, > + "Unknown event was received or event with unimplemented handler\n"); > +} > + > +static void efa_keep_alive(void *data, struct efa_admin_aenq_entry *aenq_e) > +{ > + struct efa_dev *dev = (struct efa_dev *)data; > + > + efa_stat_inc(dev, dev->stats.keep_alive_rcvd); > +} > + > +static struct efa_aenq_handlers aenq_handlers = { > + .handlers = { > + [EFA_ADMIN_KEEP_ALIVE] = efa_keep_alive, > + }, > + .unimplemented_handler = unimplemented_aenq_handler > +}; > + > +static void efa_release_bars(struct efa_dev *dev, int bars_mask) > +{ > + struct pci_dev *pdev = dev->pdev; > + int release_bars; > + > + release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & bars_mask; > + pci_release_selected_regions(pdev, release_bars); > +} > + > +static irqreturn_t efa_intr_msix_mgmnt(int irq, void *data) > +{ > + struct efa_dev *dev = (struct efa_dev *)data; > + > + efa_com_admin_q_comp_intr_handler(dev->edev); > + > + /* Don't call the aenq handler before probe is done */ > + if (likely(test_bit(EFA_DEVICE_RUNNING_BIT, &dev->state))) > + efa_com_aenq_intr_handler(dev->edev, data); > + > + return IRQ_HANDLED; > +} > + > +static int efa_request_mgmnt_irq(struct efa_dev *dev) > +{ > + struct efa_irq *irq; > + int err; > + > + irq = &dev->admin_irq; > + err = request_irq(irq->vector, irq->handler, 0, irq->name, > + irq->data); > + if (err) { > + efa_err(&dev->pdev->dev, "Failed to request admin irq (%d)\n", > + err); > + return err; > + } > + > + efa_dbg(&dev->pdev->dev, "Set affinity hint of mgmnt irq to %*pbl (irq vector: %d)\n", > + nr_cpumask_bits, &irq->affinity_hint_mask, irq->vector); > + irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask); > + > + return err; > +} > + > +static void efa_setup_mgmnt_irq(struct efa_dev *dev) > +{ > + u32 cpu; > + > + snprintf(dev->admin_irq.name, EFA_IRQNAME_SIZE, > + "efa-mgmnt@pci:%s", pci_name(dev->pdev)); > + dev->admin_irq.handler = efa_intr_msix_mgmnt; > + dev->admin_irq.data = dev; > + dev->admin_irq.vector = > + pci_irq_vector(dev->pdev, dev->admin_msix_vector_idx); > + cpu = cpumask_first(cpu_online_mask); > + dev->admin_irq.cpu = cpu; > + cpumask_set_cpu(cpu, > + &dev->admin_irq.affinity_hint_mask); > + efa_info(&dev->pdev->dev, "Setup irq:%p vector:%d name:%s\n", > + &dev->admin_irq, > + dev->admin_irq.vector, > + dev->admin_irq.name); > +} > + > +static void efa_free_mgmnt_irq(struct efa_dev *dev) > +{ > + struct efa_irq *irq; > + > + irq = &dev->admin_irq; > + irq_set_affinity_hint(irq->vector, NULL); > + free_irq(irq->vector, irq->data); > +} > + > +static int efa_set_mgmnt_irq(struct efa_dev *dev) > +{ > + efa_setup_mgmnt_irq(dev); > + > + return efa_request_mgmnt_irq(dev); > +} > + > +static int efa_request_doorbell_bar(struct efa_dev *dev) > +{ > + u8 db_bar_idx = dev->dev_attr.db_bar; > + struct pci_dev *pdev = dev->pdev; > + int bars; > + int err; > + > + if (!(BIT(db_bar_idx) & EFA_BASE_BAR_MASK)) { > + bars = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(db_bar_idx); > + > + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); > + if (err) { > + efa_err(&dev->pdev->dev, > + "pci_request_selected_regions for bar %d failed %d\n", > + db_bar_idx, err); > + return err; > + } > + } > + > + dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx); > + dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx); > + > + return 0; > +} > + > +static void efa_release_doorbell_bar(struct efa_dev *dev) > +{ > + if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK)) > + efa_release_bars(dev, BIT(dev->dev_attr.db_bar)); > +} > + > +static void efa_update_hw_hints(struct efa_dev *dev, > + struct efa_com_get_hw_hints_result *hw_hints) > +{ > + struct efa_com_dev *edev = dev->edev; > + > + if (hw_hints->mmio_read_timeout) > + edev->mmio_read.mmio_read_timeout = > + hw_hints->mmio_read_timeout * 1000; > + > + if (hw_hints->poll_interval) > + edev->aq.poll_interval = hw_hints->poll_interval; > + > + if (hw_hints->admin_completion_timeout) > + edev->aq.completion_timeout = > + hw_hints->admin_completion_timeout; > +} > + > +static const struct ib_device_ops efa_dev_ops = { > + .alloc_pd = efa_alloc_pd, > + .alloc_ucontext = efa_alloc_ucontext, > + .create_ah = efa_create_ah, > + .create_cq = efa_create_cq, > + .create_qp = efa_create_qp, > + .dealloc_pd = efa_dealloc_pd, > + .dealloc_ucontext = efa_dealloc_ucontext, > + .dereg_mr = efa_dereg_mr, > + .destroy_ah = efa_destroy_ah, > + .destroy_cq = efa_destroy_cq, > + .destroy_qp = efa_destroy_qp, > + .get_link_layer = efa_port_link_layer, > + .get_port_immutable = efa_get_port_immutable, > + .mmap = efa_mmap, > + .modify_qp = efa_modify_qp, > + .query_device = efa_query_device, > + .query_gid = efa_query_gid, > + .query_pkey = efa_query_pkey, > + .query_port = efa_query_port, > + .query_qp = efa_query_qp, > + .reg_user_mr = efa_reg_mr, > + INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd), > +}; > + > +static int efa_ib_device_add(struct efa_dev *dev) > +{ > + struct efa_com_get_network_attr_result network_attr; > + struct efa_com_get_hw_hints_result hw_hints; > + struct pci_dev *pdev = dev->pdev; > + int err; > + > + spin_lock_init(&dev->stats_lock); > + > + err = efa_com_get_device_attr(dev->edev, &dev->dev_attr); > + if (err) > + return err; > + > + efa_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n", dev->dev_attr.db_bar); > + err = efa_request_doorbell_bar(dev); > + if (err) > + return err; > + > + err = efa_com_get_network_attr(dev->edev, &network_attr); > + if (err) > + goto err_release_doorbell_bar; > + > + efa_update_network_attr(dev, &network_attr); > + > + err = efa_com_get_hw_hints(dev->edev, &hw_hints); > + if (err) > + goto err_release_doorbell_bar; > + > + efa_update_hw_hints(dev, &hw_hints); > + > + /* Try to enable all the available aenq groups */ > + err = efa_com_set_aenq_config(dev->edev, EFA_AENQ_ENABLED_GROUPS); > + if (err) > + goto err_release_doorbell_bar; > + > + dev->ibdev.owner = THIS_MODULE; > + dev->ibdev.node_type = RDMA_NODE_EFA; > + dev->ibdev.phys_port_cnt = 1; > + dev->ibdev.num_comp_vectors = 1; > + dev->ibdev.dev.parent = &pdev->dev; > + dev->ibdev.uverbs_abi_ver = EFA_UVERBS_ABI_VERSION; > + > + dev->ibdev.uverbs_cmd_mask = > + (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | > + (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) | > + (1ull << IB_USER_VERBS_CMD_QUERY_PORT) | > + (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | > + (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | > + (1ull << IB_USER_VERBS_CMD_REG_MR) | > + (1ull << IB_USER_VERBS_CMD_DEREG_MR) | > + (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | > + (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | > + (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) | > + (1ull << IB_USER_VERBS_CMD_CREATE_QP) | > + (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | > + (1ull << IB_USER_VERBS_CMD_QUERY_QP) | > + (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | > + (1ull << IB_USER_VERBS_CMD_CREATE_AH) | > + (1ull << IB_USER_VERBS_CMD_DESTROY_AH); > + > + dev->ibdev.uverbs_ex_cmd_mask = > + (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE); > + > + dev->ibdev.driver_id = RDMA_DRIVER_EFA; > + ib_set_device_ops(&dev->ibdev, &efa_dev_ops); > + > + err = ib_register_device(&dev->ibdev, "efa_%d"); > + if (err) > + goto err_release_doorbell_bar; > + > + efa_info(&dev->ibdev.dev, "IB device registered\n"); > + > + set_bit(EFA_DEVICE_RUNNING_BIT, &dev->state); > + > + return 0; > + > +err_release_doorbell_bar: > + efa_release_doorbell_bar(dev); > + return err; > +} > + > +static void efa_ib_device_remove(struct efa_dev *dev) > +{ > + > + /* Reset the device only if the device is running. */ > + if (test_bit(EFA_DEVICE_RUNNING_BIT, &dev->state)) > + efa_com_dev_reset(dev->edev, EFA_REGS_RESET_NORMAL); > + > + efa_info(&dev->ibdev.dev, "Unregister ib device\n"); > + ib_unregister_device(&dev->ibdev); > + efa_release_doorbell_bar(dev); > +} > + > +static void efa_disable_msix(struct efa_dev *dev) > +{ > + if (test_and_clear_bit(EFA_MSIX_ENABLED_BIT, &dev->state)) > + pci_free_irq_vectors(dev->pdev); > +} > + > +static int efa_enable_msix(struct efa_dev *dev) > +{ > + int msix_vecs, irq_num; > + > + if (test_bit(EFA_MSIX_ENABLED_BIT, &dev->state)) { > + efa_err(&dev->pdev->dev, "Error, MSI-X is already enabled\n"); > + return -EPERM; > + } > + > + /* Reserve the max msix vectors we might need */ > + msix_vecs = EFA_NUM_MSIX_VEC; > + efa_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n", > + msix_vecs); > + > + dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX; > + irq_num = pci_alloc_irq_vectors(dev->pdev, msix_vecs, > + msix_vecs, PCI_IRQ_MSIX); > + > + if (irq_num < 0) { > + efa_err(&dev->pdev->dev, "Failed to enable MSI-X. irq_num %d\n", > + irq_num); > + return -ENOSPC; > + } > + > + if (irq_num != msix_vecs) { > + efa_warn(&dev->pdev->dev, > + "Allocated %d MSI-X (out of %d requested)\n", > + irq_num, msix_vecs); > + return -ENOSPC; > + } > + > + set_bit(EFA_MSIX_ENABLED_BIT, &dev->state); > + > + return 0; > +} > + > +static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev) > +{ > + int dma_width; > + int err; > + > + err = efa_com_dev_reset(edev, EFA_REGS_RESET_NORMAL); > + if (err) > + return err; > + > + err = efa_com_validate_version(edev); > + if (err) > + return err; > + > + dma_width = efa_com_get_dma_width(edev); > + if (dma_width < 0) { > + err = dma_width; > + return err; > + } > + > + err = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width)); > + if (err) { > + efa_err(&pdev->dev, "pci_set_dma_mask failed 0x%x\n", err); > + return err; > + } > + > + err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width)); > + if (err) { > + efa_err(&pdev->dev, > + "err_pci_set_consistent_dma_mask failed 0x%x\n", > + err); > + return err; > + } > + > + return 0; > +} > + > +static int efa_probe_device(struct pci_dev *pdev) > +{ > + struct efa_com_dev *edev; > + struct efa_dev *dev; > + int bars; > + int err; > + > + err = pci_enable_device_mem(pdev); > + if (err) { > + efa_err(&pdev->dev, "pci_enable_device_mem() failed!\n"); > + return err; > + } > + > + pci_set_master(pdev); > + > + dev = ib_alloc_device(efa_dev, ibdev); > + if (IS_ERR_OR_NULL(dev)) { > + efa_err(&pdev->dev, "Device alloc failed\n"); > + err = dev ? PTR_ERR(dev) : -ENOMEM; > + goto err_disable_device; > + } > + > + edev = kzalloc(sizeof(*edev), GFP_KERNEL); > + if (!edev) { > + err = -ENOMEM; > + goto err_ibdev_destroy; > + } > + > + pci_set_drvdata(pdev, dev); > + edev->dmadev = &pdev->dev; > + dev->edev = edev; > + dev->pdev = pdev; > + > + bars = pci_select_bars(pdev, IORESOURCE_MEM) & EFA_BASE_BAR_MASK; > + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); > + if (err) { > + efa_err(&pdev->dev, "pci_request_selected_regions failed %d\n", > + err); > + goto err_free_efa_dev; > + } > + > + dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR); > + dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR); > + dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR); > + dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR); > + > + edev->reg_bar = devm_ioremap(&pdev->dev, > + dev->reg_bar_addr, > + dev->reg_bar_len); > + if (!edev->reg_bar) { > + efa_err(&pdev->dev, "Failed to remap register bar\n"); > + err = -EFAULT; > + goto err_release_bars; > + } > + > + err = efa_com_mmio_reg_read_init(edev); > + if (err) { > + efa_err(&pdev->dev, "Failed to init readless MMIO\n"); > + goto err_iounmap; > + } > + > + err = efa_device_init(edev, pdev); > + if (err) { > + efa_err(&pdev->dev, "EFA device init failed\n"); > + if (err == -ETIME) > + err = -EPROBE_DEFER; > + goto err_reg_read_destroy; > + } > + > + err = efa_enable_msix(dev); > + if (err) > + goto err_reg_read_destroy; > + > + edev->aq.msix_vector_idx = dev->admin_msix_vector_idx; > + edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx; > + > + err = efa_set_mgmnt_irq(dev); > + if (err) > + goto err_disable_msix; > + > + err = efa_com_admin_init(edev, &aenq_handlers); > + if (err) > + goto err_free_mgmnt_irq; > + > + return 0; > + > +err_free_mgmnt_irq: > + efa_free_mgmnt_irq(dev); > +err_disable_msix: > + efa_disable_msix(dev); > +err_reg_read_destroy: > + efa_com_mmio_reg_read_destroy(edev); > +err_iounmap: > + devm_iounmap(&pdev->dev, edev->reg_bar); > +err_release_bars: > + efa_release_bars(dev, EFA_BASE_BAR_MASK); > +err_free_efa_dev: > + kfree(edev); > +err_ibdev_destroy: > + ib_dealloc_device(&dev->ibdev); > +err_disable_device: > + pci_disable_device(pdev); > + return err; > +} > + > +static void efa_remove_device(struct pci_dev *pdev) > +{ > + struct efa_dev *dev = pci_get_drvdata(pdev); > + struct efa_com_dev *edev; > + > + if (!dev) > + /* > + * This device didn't load properly and its resources > + * already released, nothing to do > + */ > + return; > + > + edev = dev->edev; > + > + efa_com_admin_destroy(edev); > + efa_free_mgmnt_irq(dev); > + efa_disable_msix(dev); > + efa_com_mmio_reg_read_destroy(edev); > + devm_iounmap(&pdev->dev, edev->reg_bar); > + efa_release_bars(dev, EFA_BASE_BAR_MASK); > + kfree(edev); > + ib_dealloc_device(&dev->ibdev); > + pci_disable_device(pdev); > +} > + > +static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent) > +{ > + struct efa_dev *dev; > + int err; > + > + err = efa_probe_device(pdev); > + if (err) > + return err; > + > + dev = pci_get_drvdata(pdev); > + err = efa_ib_device_add(dev); > + if (err) > + goto err_remove_device; > + > + return 0; > + > +err_remove_device: > + efa_remove_device(pdev); > + return err; > +} > + > +static void efa_remove(struct pci_dev *pdev) > +{ > + struct efa_dev *dev = (struct efa_dev *)pci_get_drvdata(pdev); > + > + efa_ib_device_remove(dev); > + efa_remove_device(pdev); > +} > + > +static struct pci_driver efa_pci_driver = { > + .name = DRV_MODULE_NAME, > + .id_table = efa_pci_tbl, > + .probe = efa_probe, > + .remove = efa_remove, > +}; > + > +static int __init efa_init(void) > +{ > + int err; > + > + err = pci_register_driver(&efa_pci_driver); > + if (err) { > + pr_err("Couldn't register efa driver\n"); > + goto err_register; > + } > + > + return 0; > + > +err_register: > + return err; > +} > + > +static void __exit efa_exit(void) > +{ > + pci_unregister_driver(&efa_pci_driver); > +} > + > +module_init(efa_init); > +module_exit(efa_exit); Reviewed-by: Steve Wise <swise@opengridcomputing.com>
On 26-Feb-19 21:28, Leon Romanovsky wrote: > On Thu, Feb 21, 2019 at 05:33:12PM +0200, Gal Pressman wrote: >> Add the main EFA module file which takes care of device >> probe/initialization/registration/etc. >> >> Signed-off-by: Gal Pressman <galpress@amazon.com> >> --- >> drivers/infiniband/hw/efa/efa_main.c | 577 +++++++++++++++++++++++++++++++++++ >> 1 file changed, 577 insertions(+) >> create mode 100644 drivers/infiniband/hw/efa/efa_main.c >> >> diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c >> new file mode 100644 >> index 000000000000..a4ccebc9665d >> --- /dev/null >> +++ b/drivers/infiniband/hw/efa/efa_main.c >> @@ -0,0 +1,577 @@ >> +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause > > Are you sure that this is the license and Linux-OpenIB? What do you mean? > >> +/* >> + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. >> + */ >> + >> + >> +static const struct ib_device_ops efa_dev_ops = { >> + .alloc_pd = efa_alloc_pd, >> + .alloc_ucontext = efa_alloc_ucontext, >> + .create_ah = efa_create_ah, >> + .create_cq = efa_create_cq, >> + .create_qp = efa_create_qp, >> + .dealloc_pd = efa_dealloc_pd, >> + .dealloc_ucontext = efa_dealloc_ucontext, >> + .dereg_mr = efa_dereg_mr, >> + .destroy_ah = efa_destroy_ah, >> + .destroy_cq = efa_destroy_cq, >> + .destroy_qp = efa_destroy_qp, >> + .get_link_layer = efa_port_link_layer, >> + .get_port_immutable = efa_get_port_immutable, >> + .mmap = efa_mmap, >> + .modify_qp = efa_modify_qp, >> + .query_device = efa_query_device, >> + .query_gid = efa_query_gid, >> + .query_pkey = efa_query_pkey, >> + .query_port = efa_query_port, >> + .query_qp = efa_query_qp, >> + .reg_user_mr = efa_reg_mr, >> + INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd), > > ucontext needs this INIT_RDMA_OBJ_SIZE() too. Trying to keep up with these changes :). > >> +}; >> + >> +static int efa_probe_device(struct pci_dev *pdev) >> +{ >> + struct efa_com_dev *edev; >> + struct efa_dev *dev; >> + int bars; >> + int err; >> + >> + err = pci_enable_device_mem(pdev); >> + if (err) { >> + efa_err(&pdev->dev, "pci_enable_device_mem() failed!\n"); >> + return err; >> + } >> + >> + pci_set_master(pdev); >> + >> + dev = ib_alloc_device(efa_dev, ibdev); > > It return or NULL or real pointer, ACK. > >> + if (IS_ERR_OR_NULL(dev)) { >> + efa_err(&pdev->dev, "Device alloc failed\n"); >> + err = dev ? PTR_ERR(dev) : -ENOMEM; >> + goto err_disable_device; >> + } >> + >> + edev = kzalloc(sizeof(*edev), GFP_KERNEL); >> + if (!edev) { >> + err = -ENOMEM; >> + goto err_ibdev_destroy; >> + } >> + >> + pci_set_drvdata(pdev, dev); >> + edev->dmadev = &pdev->dev; >> + dev->edev = edev; >> + dev->pdev = pdev; >> + >> + bars = pci_select_bars(pdev, IORESOURCE_MEM) & EFA_BASE_BAR_MASK; >> + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); >> + if (err) { >> + efa_err(&pdev->dev, "pci_request_selected_regions failed %d\n", >> + err); >> + goto err_free_efa_dev; >> + } >> + >> + dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR); >> + dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR); >> + dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR); >> + dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR); >> + >> + edev->reg_bar = devm_ioremap(&pdev->dev, >> + dev->reg_bar_addr, >> + dev->reg_bar_len); >> + if (!edev->reg_bar) { >> + efa_err(&pdev->dev, "Failed to remap register bar\n"); >> + err = -EFAULT; >> + goto err_release_bars; >> + } >> + >> + err = efa_com_mmio_reg_read_init(edev); >> + if (err) { >> + efa_err(&pdev->dev, "Failed to init readless MMIO\n"); >> + goto err_iounmap; >> + } >> + >> + err = efa_device_init(edev, pdev); >> + if (err) { >> + efa_err(&pdev->dev, "EFA device init failed\n"); >> + if (err == -ETIME) >> + err = -EPROBE_DEFER; >> + goto err_reg_read_destroy; >> + } >> + >> + err = efa_enable_msix(dev); >> + if (err) >> + goto err_reg_read_destroy; >> + >> + edev->aq.msix_vector_idx = dev->admin_msix_vector_idx; >> + edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx; >> + >> + err = efa_set_mgmnt_irq(dev); >> + if (err) >> + goto err_disable_msix; >> + >> + err = efa_com_admin_init(edev, &aenq_handlers); >> + if (err) >> + goto err_free_mgmnt_irq; >> + >> + return 0; >> + >> +err_free_mgmnt_irq: >> + efa_free_mgmnt_irq(dev); >> +err_disable_msix: >> + efa_disable_msix(dev); >> +err_reg_read_destroy: >> + efa_com_mmio_reg_read_destroy(edev); >> +err_iounmap: >> + devm_iounmap(&pdev->dev, edev->reg_bar); >> +err_release_bars: >> + efa_release_bars(dev, EFA_BASE_BAR_MASK); >> +err_free_efa_dev: >> + kfree(edev); >> +err_ibdev_destroy: >> + ib_dealloc_device(&dev->ibdev); >> +err_disable_device: >> + pci_disable_device(pdev); >> + return err; >> +} >> + >> +static void efa_remove_device(struct pci_dev *pdev) >> +{ >> + struct efa_dev *dev = pci_get_drvdata(pdev); >> + struct efa_com_dev *edev; >> + >> + if (!dev) >> + /* >> + * This device didn't load properly and its resources >> + * already released, nothing to do >> + */ > > How can efa_remove_device() be called if it didn't success efa_add_device? It can't, removing. > >> + return; >> + >> + edev = dev->edev; >> + >> + efa_com_admin_destroy(edev); >> + efa_free_mgmnt_irq(dev); >> + efa_disable_msix(dev); >> + efa_com_mmio_reg_read_destroy(edev); >> + devm_iounmap(&pdev->dev, edev->reg_bar); >> + efa_release_bars(dev, EFA_BASE_BAR_MASK); >> + kfree(edev); >> + ib_dealloc_device(&dev->ibdev); >> + pci_disable_device(pdev); >> +} >> + >> +static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent) >> +{ >> + struct efa_dev *dev; >> + int err; >> + >> + err = efa_probe_device(pdev); >> + if (err) >> + return err; >> + >> + dev = pci_get_drvdata(pdev); >> + err = efa_ib_device_add(dev); >> + if (err) >> + goto err_remove_device; >> + >> + return 0; >> + >> +err_remove_device: >> + efa_remove_device(pdev); >> + return err; >> +} >> + >> +static void efa_remove(struct pci_dev *pdev) >> +{ >> + struct efa_dev *dev = (struct efa_dev *)pci_get_drvdata(pdev); >> + >> + efa_ib_device_remove(dev); >> + efa_remove_device(pdev); >> +} >> + >> +static struct pci_driver efa_pci_driver = { >> + .name = DRV_MODULE_NAME, >> + .id_table = efa_pci_tbl, >> + .probe = efa_probe, >> + .remove = efa_remove, >> +}; >> + >> +static int __init efa_init(void) >> +{ >> + int err; >> + >> + err = pci_register_driver(&efa_pci_driver); >> + if (err) { >> + pr_err("Couldn't register efa driver\n"); >> + goto err_register; >> + } >> + >> + return 0; >> + >> +err_register: >> + return err; >> +} >> + >> +static void __exit efa_exit(void) >> +{ >> + pci_unregister_driver(&efa_pci_driver); >> +} >> + >> +module_init(efa_init); >> +module_exit(efa_exit); >> -- >> 2.7.4 >>
On 26-Feb-19 23:48, Steve Wise wrote: > > On 2/21/2019 9:33 AM, Gal Pressman wrote: >> Add the main EFA module file which takes care of device >> probe/initialization/registration/etc. >> >> Signed-off-by: Gal Pressman <galpress@amazon.com> >> --- >> drivers/infiniband/hw/efa/efa_main.c | 577 +++++++++++++++++++++++++++++++++++ >> 1 file changed, 577 insertions(+) >> create mode 100644 drivers/infiniband/hw/efa/efa_main.c >> >> diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c >> new file mode 100644 >> index 000000000000..a4ccebc9665d >> --- /dev/null >> +++ b/drivers/infiniband/hw/efa/efa_main.c >> @@ -0,0 +1,577 @@ >> +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause >> +/* >> + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. >> + */ >> + >> +#include <linux/module.h> >> +#include <linux/pci.h> >> + >> +#include <rdma/ib_user_verbs.h> >> + >> +#include "efa.h" >> + >> +#define PCI_DEV_ID_EFA_VF 0xefa0 >> + >> +static const struct pci_device_id efa_pci_tbl[] = { >> + { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA_VF) }, >> + { } >> +}; >> + >> +MODULE_AUTHOR("Amazon.com, Inc. or its affiliates"); >> +MODULE_LICENSE("Dual BSD/GPL"); >> +MODULE_DESCRIPTION(DEVICE_NAME); >> +MODULE_DEVICE_TABLE(pci, efa_pci_tbl); >> + >> +#define EFA_REG_BAR 0 >> +#define EFA_MEM_BAR 2 >> +#define EFA_BASE_BAR_MASK (BIT(EFA_REG_BAR) | BIT(EFA_MEM_BAR)) >> + >> +#define EFA_AENQ_ENABLED_GROUPS \ >> + (BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \ >> + BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE)) >> + >> +static void efa_update_network_attr(struct efa_dev *dev, >> + struct efa_com_get_network_attr_result *network_attr) >> +{ >> + memcpy(dev->addr, network_attr->addr, sizeof(network_attr->addr)); >> + dev->mtu = network_attr->mtu; >> + >> + efa_dbg(&dev->pdev->dev, "Full address %pI6\n", dev->addr); >> +} >> + >> +/* This handler will called for unknown event group or unimplemented handlers */ >> +static void unimplemented_aenq_handler(void *data, >> + struct efa_admin_aenq_entry *aenq_e) >> +{ >> + struct efa_dev *dev = (struct efa_dev *)data; >> + >> + efa_err_rl(&dev->ibdev.dev, >> + "Unknown event was received or event with unimplemented handler\n"); >> +} >> + >> +static void efa_keep_alive(void *data, struct efa_admin_aenq_entry *aenq_e) >> +{ >> + struct efa_dev *dev = (struct efa_dev *)data; >> + >> + efa_stat_inc(dev, dev->stats.keep_alive_rcvd); >> +} >> + >> +static struct efa_aenq_handlers aenq_handlers = { >> + .handlers = { >> + [EFA_ADMIN_KEEP_ALIVE] = efa_keep_alive, >> + }, >> + .unimplemented_handler = unimplemented_aenq_handler >> +}; >> + >> +static void efa_release_bars(struct efa_dev *dev, int bars_mask) >> +{ >> + struct pci_dev *pdev = dev->pdev; >> + int release_bars; >> + >> + release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & bars_mask; >> + pci_release_selected_regions(pdev, release_bars); >> +} >> + >> +static irqreturn_t efa_intr_msix_mgmnt(int irq, void *data) >> +{ >> + struct efa_dev *dev = (struct efa_dev *)data; >> + >> + efa_com_admin_q_comp_intr_handler(dev->edev); >> + >> + /* Don't call the aenq handler before probe is done */ >> + if (likely(test_bit(EFA_DEVICE_RUNNING_BIT, &dev->state))) >> + efa_com_aenq_intr_handler(dev->edev, data); >> + >> + return IRQ_HANDLED; >> +} >> + >> +static int efa_request_mgmnt_irq(struct efa_dev *dev) >> +{ >> + struct efa_irq *irq; >> + int err; >> + >> + irq = &dev->admin_irq; >> + err = request_irq(irq->vector, irq->handler, 0, irq->name, >> + irq->data); >> + if (err) { >> + efa_err(&dev->pdev->dev, "Failed to request admin irq (%d)\n", >> + err); >> + return err; >> + } >> + >> + efa_dbg(&dev->pdev->dev, "Set affinity hint of mgmnt irq to %*pbl (irq vector: %d)\n", >> + nr_cpumask_bits, &irq->affinity_hint_mask, irq->vector); >> + irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask); >> + >> + return err; >> +} >> + >> +static void efa_setup_mgmnt_irq(struct efa_dev *dev) >> +{ >> + u32 cpu; >> + >> + snprintf(dev->admin_irq.name, EFA_IRQNAME_SIZE, >> + "efa-mgmnt@pci:%s", pci_name(dev->pdev)); >> + dev->admin_irq.handler = efa_intr_msix_mgmnt; >> + dev->admin_irq.data = dev; >> + dev->admin_irq.vector = >> + pci_irq_vector(dev->pdev, dev->admin_msix_vector_idx); >> + cpu = cpumask_first(cpu_online_mask); >> + dev->admin_irq.cpu = cpu; >> + cpumask_set_cpu(cpu, >> + &dev->admin_irq.affinity_hint_mask); >> + efa_info(&dev->pdev->dev, "Setup irq:%p vector:%d name:%s\n", >> + &dev->admin_irq, >> + dev->admin_irq.vector, >> + dev->admin_irq.name); >> +} >> + >> +static void efa_free_mgmnt_irq(struct efa_dev *dev) >> +{ >> + struct efa_irq *irq; >> + >> + irq = &dev->admin_irq; >> + irq_set_affinity_hint(irq->vector, NULL); >> + free_irq(irq->vector, irq->data); >> +} >> + >> +static int efa_set_mgmnt_irq(struct efa_dev *dev) >> +{ >> + efa_setup_mgmnt_irq(dev); >> + >> + return efa_request_mgmnt_irq(dev); >> +} >> + >> +static int efa_request_doorbell_bar(struct efa_dev *dev) >> +{ >> + u8 db_bar_idx = dev->dev_attr.db_bar; >> + struct pci_dev *pdev = dev->pdev; >> + int bars; >> + int err; >> + >> + if (!(BIT(db_bar_idx) & EFA_BASE_BAR_MASK)) { >> + bars = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(db_bar_idx); >> + >> + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); >> + if (err) { >> + efa_err(&dev->pdev->dev, >> + "pci_request_selected_regions for bar %d failed %d\n", >> + db_bar_idx, err); >> + return err; >> + } >> + } >> + >> + dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx); >> + dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx); >> + >> + return 0; >> +} >> + >> +static void efa_release_doorbell_bar(struct efa_dev *dev) >> +{ >> + if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK)) >> + efa_release_bars(dev, BIT(dev->dev_attr.db_bar)); >> +} >> + >> +static void efa_update_hw_hints(struct efa_dev *dev, >> + struct efa_com_get_hw_hints_result *hw_hints) >> +{ >> + struct efa_com_dev *edev = dev->edev; >> + >> + if (hw_hints->mmio_read_timeout) >> + edev->mmio_read.mmio_read_timeout = >> + hw_hints->mmio_read_timeout * 1000; >> + >> + if (hw_hints->poll_interval) >> + edev->aq.poll_interval = hw_hints->poll_interval; >> + >> + if (hw_hints->admin_completion_timeout) >> + edev->aq.completion_timeout = >> + hw_hints->admin_completion_timeout; >> +} >> + >> +static const struct ib_device_ops efa_dev_ops = { >> + .alloc_pd = efa_alloc_pd, >> + .alloc_ucontext = efa_alloc_ucontext, >> + .create_ah = efa_create_ah, >> + .create_cq = efa_create_cq, >> + .create_qp = efa_create_qp, >> + .dealloc_pd = efa_dealloc_pd, >> + .dealloc_ucontext = efa_dealloc_ucontext, >> + .dereg_mr = efa_dereg_mr, >> + .destroy_ah = efa_destroy_ah, >> + .destroy_cq = efa_destroy_cq, >> + .destroy_qp = efa_destroy_qp, >> + .get_link_layer = efa_port_link_layer, >> + .get_port_immutable = efa_get_port_immutable, >> + .mmap = efa_mmap, >> + .modify_qp = efa_modify_qp, >> + .query_device = efa_query_device, >> + .query_gid = efa_query_gid, >> + .query_pkey = efa_query_pkey, >> + .query_port = efa_query_port, >> + .query_qp = efa_query_qp, >> + .reg_user_mr = efa_reg_mr, >> + INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd), >> +}; >> + >> +static int efa_ib_device_add(struct efa_dev *dev) >> +{ >> + struct efa_com_get_network_attr_result network_attr; >> + struct efa_com_get_hw_hints_result hw_hints; >> + struct pci_dev *pdev = dev->pdev; >> + int err; >> + >> + spin_lock_init(&dev->stats_lock); >> + >> + err = efa_com_get_device_attr(dev->edev, &dev->dev_attr); >> + if (err) >> + return err; >> + >> + efa_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n", dev->dev_attr.db_bar); >> + err = efa_request_doorbell_bar(dev); >> + if (err) >> + return err; >> + >> + err = efa_com_get_network_attr(dev->edev, &network_attr); >> + if (err) >> + goto err_release_doorbell_bar; >> + >> + efa_update_network_attr(dev, &network_attr); >> + >> + err = efa_com_get_hw_hints(dev->edev, &hw_hints); >> + if (err) >> + goto err_release_doorbell_bar; >> + >> + efa_update_hw_hints(dev, &hw_hints); >> + >> + /* Try to enable all the available aenq groups */ >> + err = efa_com_set_aenq_config(dev->edev, EFA_AENQ_ENABLED_GROUPS); >> + if (err) >> + goto err_release_doorbell_bar; >> + >> + dev->ibdev.owner = THIS_MODULE; >> + dev->ibdev.node_type = RDMA_NODE_EFA; >> + dev->ibdev.phys_port_cnt = 1; >> + dev->ibdev.num_comp_vectors = 1; >> + dev->ibdev.dev.parent = &pdev->dev; >> + dev->ibdev.uverbs_abi_ver = EFA_UVERBS_ABI_VERSION; >> + >> + dev->ibdev.uverbs_cmd_mask = >> + (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | >> + (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) | >> + (1ull << IB_USER_VERBS_CMD_QUERY_PORT) | >> + (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | >> + (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | >> + (1ull << IB_USER_VERBS_CMD_REG_MR) | >> + (1ull << IB_USER_VERBS_CMD_DEREG_MR) | >> + (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | >> + (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | >> + (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) | >> + (1ull << IB_USER_VERBS_CMD_CREATE_QP) | >> + (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | >> + (1ull << IB_USER_VERBS_CMD_QUERY_QP) | >> + (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | >> + (1ull << IB_USER_VERBS_CMD_CREATE_AH) | >> + (1ull << IB_USER_VERBS_CMD_DESTROY_AH); >> + >> + dev->ibdev.uverbs_ex_cmd_mask = >> + (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE); >> + >> + dev->ibdev.driver_id = RDMA_DRIVER_EFA; >> + ib_set_device_ops(&dev->ibdev, &efa_dev_ops); >> + >> + err = ib_register_device(&dev->ibdev, "efa_%d"); >> + if (err) >> + goto err_release_doorbell_bar; >> + >> + efa_info(&dev->ibdev.dev, "IB device registered\n"); >> + >> + set_bit(EFA_DEVICE_RUNNING_BIT, &dev->state); >> + >> + return 0; >> + >> +err_release_doorbell_bar: >> + efa_release_doorbell_bar(dev); >> + return err; >> +} >> + >> +static void efa_ib_device_remove(struct efa_dev *dev) >> +{ >> + >> + /* Reset the device only if the device is running. */ >> + if (test_bit(EFA_DEVICE_RUNNING_BIT, &dev->state)) >> + efa_com_dev_reset(dev->edev, EFA_REGS_RESET_NORMAL); >> + >> + efa_info(&dev->ibdev.dev, "Unregister ib device\n"); >> + ib_unregister_device(&dev->ibdev); >> + efa_release_doorbell_bar(dev); >> +} >> + >> +static void efa_disable_msix(struct efa_dev *dev) >> +{ >> + if (test_and_clear_bit(EFA_MSIX_ENABLED_BIT, &dev->state)) >> + pci_free_irq_vectors(dev->pdev); >> +} >> + >> +static int efa_enable_msix(struct efa_dev *dev) >> +{ >> + int msix_vecs, irq_num; >> + >> + if (test_bit(EFA_MSIX_ENABLED_BIT, &dev->state)) { >> + efa_err(&dev->pdev->dev, "Error, MSI-X is already enabled\n"); >> + return -EPERM; >> + } >> + >> + /* Reserve the max msix vectors we might need */ >> + msix_vecs = EFA_NUM_MSIX_VEC; >> + efa_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n", >> + msix_vecs); >> + >> + dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX; >> + irq_num = pci_alloc_irq_vectors(dev->pdev, msix_vecs, >> + msix_vecs, PCI_IRQ_MSIX); >> + >> + if (irq_num < 0) { >> + efa_err(&dev->pdev->dev, "Failed to enable MSI-X. irq_num %d\n", >> + irq_num); >> + return -ENOSPC; >> + } >> + >> + if (irq_num != msix_vecs) { >> + efa_warn(&dev->pdev->dev, >> + "Allocated %d MSI-X (out of %d requested)\n", >> + irq_num, msix_vecs); >> + return -ENOSPC; >> + } >> + >> + set_bit(EFA_MSIX_ENABLED_BIT, &dev->state); >> + >> + return 0; >> +} >> + >> +static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev) >> +{ >> + int dma_width; >> + int err; >> + >> + err = efa_com_dev_reset(edev, EFA_REGS_RESET_NORMAL); >> + if (err) >> + return err; >> + >> + err = efa_com_validate_version(edev); >> + if (err) >> + return err; >> + >> + dma_width = efa_com_get_dma_width(edev); >> + if (dma_width < 0) { >> + err = dma_width; >> + return err; >> + } >> + >> + err = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width)); >> + if (err) { >> + efa_err(&pdev->dev, "pci_set_dma_mask failed 0x%x\n", err); >> + return err; >> + } >> + >> + err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width)); >> + if (err) { >> + efa_err(&pdev->dev, >> + "err_pci_set_consistent_dma_mask failed 0x%x\n", >> + err); >> + return err; >> + } >> + >> + return 0; >> +} >> + >> +static int efa_probe_device(struct pci_dev *pdev) >> +{ >> + struct efa_com_dev *edev; >> + struct efa_dev *dev; >> + int bars; >> + int err; >> + >> + err = pci_enable_device_mem(pdev); >> + if (err) { >> + efa_err(&pdev->dev, "pci_enable_device_mem() failed!\n"); >> + return err; >> + } >> + >> + pci_set_master(pdev); >> + >> + dev = ib_alloc_device(efa_dev, ibdev); >> + if (IS_ERR_OR_NULL(dev)) { >> + efa_err(&pdev->dev, "Device alloc failed\n"); >> + err = dev ? PTR_ERR(dev) : -ENOMEM; >> + goto err_disable_device; >> + } >> + >> + edev = kzalloc(sizeof(*edev), GFP_KERNEL); >> + if (!edev) { >> + err = -ENOMEM; >> + goto err_ibdev_destroy; >> + } >> + >> + pci_set_drvdata(pdev, dev); >> + edev->dmadev = &pdev->dev; >> + dev->edev = edev; >> + dev->pdev = pdev; >> + >> + bars = pci_select_bars(pdev, IORESOURCE_MEM) & EFA_BASE_BAR_MASK; >> + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); >> + if (err) { >> + efa_err(&pdev->dev, "pci_request_selected_regions failed %d\n", >> + err); >> + goto err_free_efa_dev; >> + } >> + >> + dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR); >> + dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR); >> + dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR); >> + dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR); >> + >> + edev->reg_bar = devm_ioremap(&pdev->dev, >> + dev->reg_bar_addr, >> + dev->reg_bar_len); >> + if (!edev->reg_bar) { >> + efa_err(&pdev->dev, "Failed to remap register bar\n"); >> + err = -EFAULT; >> + goto err_release_bars; >> + } >> + >> + err = efa_com_mmio_reg_read_init(edev); >> + if (err) { >> + efa_err(&pdev->dev, "Failed to init readless MMIO\n"); >> + goto err_iounmap; >> + } >> + >> + err = efa_device_init(edev, pdev); >> + if (err) { >> + efa_err(&pdev->dev, "EFA device init failed\n"); >> + if (err == -ETIME) >> + err = -EPROBE_DEFER; >> + goto err_reg_read_destroy; >> + } >> + >> + err = efa_enable_msix(dev); >> + if (err) >> + goto err_reg_read_destroy; >> + >> + edev->aq.msix_vector_idx = dev->admin_msix_vector_idx; >> + edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx; >> + >> + err = efa_set_mgmnt_irq(dev); >> + if (err) >> + goto err_disable_msix; >> + >> + err = efa_com_admin_init(edev, &aenq_handlers); >> + if (err) >> + goto err_free_mgmnt_irq; >> + >> + return 0; >> + >> +err_free_mgmnt_irq: >> + efa_free_mgmnt_irq(dev); >> +err_disable_msix: >> + efa_disable_msix(dev); >> +err_reg_read_destroy: >> + efa_com_mmio_reg_read_destroy(edev); >> +err_iounmap: >> + devm_iounmap(&pdev->dev, edev->reg_bar); >> +err_release_bars: >> + efa_release_bars(dev, EFA_BASE_BAR_MASK); >> +err_free_efa_dev: >> + kfree(edev); >> +err_ibdev_destroy: >> + ib_dealloc_device(&dev->ibdev); >> +err_disable_device: >> + pci_disable_device(pdev); >> + return err; >> +} >> + >> +static void efa_remove_device(struct pci_dev *pdev) >> +{ >> + struct efa_dev *dev = pci_get_drvdata(pdev); >> + struct efa_com_dev *edev; >> + >> + if (!dev) >> + /* >> + * This device didn't load properly and its resources >> + * already released, nothing to do >> + */ >> + return; >> + >> + edev = dev->edev; >> + >> + efa_com_admin_destroy(edev); >> + efa_free_mgmnt_irq(dev); >> + efa_disable_msix(dev); >> + efa_com_mmio_reg_read_destroy(edev); >> + devm_iounmap(&pdev->dev, edev->reg_bar); >> + efa_release_bars(dev, EFA_BASE_BAR_MASK); >> + kfree(edev); >> + ib_dealloc_device(&dev->ibdev); >> + pci_disable_device(pdev); >> +} >> + >> +static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent) >> +{ >> + struct efa_dev *dev; >> + int err; >> + >> + err = efa_probe_device(pdev); >> + if (err) >> + return err; >> + >> + dev = pci_get_drvdata(pdev); >> + err = efa_ib_device_add(dev); >> + if (err) >> + goto err_remove_device; >> + >> + return 0; >> + >> +err_remove_device: >> + efa_remove_device(pdev); >> + return err; >> +} >> + >> +static void efa_remove(struct pci_dev *pdev) >> +{ >> + struct efa_dev *dev = (struct efa_dev *)pci_get_drvdata(pdev); >> + >> + efa_ib_device_remove(dev); >> + efa_remove_device(pdev); >> +} >> + >> +static struct pci_driver efa_pci_driver = { >> + .name = DRV_MODULE_NAME, >> + .id_table = efa_pci_tbl, >> + .probe = efa_probe, >> + .remove = efa_remove, >> +}; >> + >> +static int __init efa_init(void) >> +{ >> + int err; >> + >> + err = pci_register_driver(&efa_pci_driver); >> + if (err) { >> + pr_err("Couldn't register efa driver\n"); >> + goto err_register; >> + } >> + >> + return 0; >> + >> +err_register: >> + return err; >> +} >> + >> +static void __exit efa_exit(void) >> +{ >> + pci_unregister_driver(&efa_pci_driver); >> +} >> + >> +module_init(efa_init); >> +module_exit(efa_exit); > > > Reviewed-by: Steve Wise <swise@opengridcomputing.com> Thanks Steve!
>Subject: [PATCH rdma-next v2 10/11] RDMA/efa: Add the efa module > >Add the main EFA module file which takes care of device >probe/initialization/registration/etc. > >Signed-off-by: Gal Pressman <galpress@amazon.com> >--- > drivers/infiniband/hw/efa/efa_main.c | 577 >+++++++++++++++++++++++++++++++++++ [...] >+static int __init efa_init(void) >+{ >+ int err; >+ >+ err = pci_register_driver(&efa_pci_driver); >+ if (err) { >+ pr_err("Couldn't register efa driver\n"); >+ goto err_register; Nit: just return err here. >+ } >+ >+ return 0; >+ >+err_register: >+ return err; >+} >+ Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com>
On 28-Feb-19 00:23, Saleem, Shiraz wrote: >> Subject: [PATCH rdma-next v2 10/11] RDMA/efa: Add the efa module >> >> Add the main EFA module file which takes care of device >> probe/initialization/registration/etc. >> >> Signed-off-by: Gal Pressman <galpress@amazon.com> >> --- >> drivers/infiniband/hw/efa/efa_main.c | 577 >> +++++++++++++++++++++++++++++++++++ > > [...] >> +static int __init efa_init(void) >> +{ >> + int err; >> + >> + err = pci_register_driver(&efa_pci_driver); >> + if (err) { >> + pr_err("Couldn't register efa driver\n"); >> + goto err_register; > Nit: just return err here. >> + } >> + >> + return 0; >> + >> +err_register: >> + return err; >> +} >> + > > Reviewed-by: Shiraz Saleem <shiraz.saleem@intel.com> Thanks Shiraz!
On Tue, Feb 26, 2019 at 09:28:52PM +0200, Leon Romanovsky wrote: > On Thu, Feb 21, 2019 at 05:33:12PM +0200, Gal Pressman wrote: > > Add the main EFA module file which takes care of device > > probe/initialization/registration/etc. > > > > Signed-off-by: Gal Pressman <galpress@amazon.com> > > drivers/infiniband/hw/efa/efa_main.c | 577 +++++++++++++++++++++++++++++++++++ > > 1 file changed, 577 insertions(+) > > create mode 100644 drivers/infiniband/hw/efa/efa_main.c > > > > diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c > > new file mode 100644 > > index 000000000000..a4ccebc9665d > > +++ b/drivers/infiniband/hw/efa/efa_main.c > > @@ -0,0 +1,577 @@ > > +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause > > Are you sure that this is the license and Linux-OpenIB? If you copied any code from any of the other drivers then the license should match or you are re-licensing the code you copied. Jason
On 04-Mar-19 23:13, Jason Gunthorpe wrote: > On Tue, Feb 26, 2019 at 09:28:52PM +0200, Leon Romanovsky wrote: >> On Thu, Feb 21, 2019 at 05:33:12PM +0200, Gal Pressman wrote: >>> Add the main EFA module file which takes care of device >>> probe/initialization/registration/etc. >>> >>> Signed-off-by: Gal Pressman <galpress@amazon.com> >>> drivers/infiniband/hw/efa/efa_main.c | 577 +++++++++++++++++++++++++++++++++++ >>> 1 file changed, 577 insertions(+) >>> create mode 100644 drivers/infiniband/hw/efa/efa_main.c >>> >>> diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c >>> new file mode 100644 >>> index 000000000000..a4ccebc9665d >>> +++ b/drivers/infiniband/hw/efa/efa_main.c >>> @@ -0,0 +1,577 @@ >>> +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause >> >> Are you sure that this is the license and Linux-OpenIB? > > If you copied any code from any of the other drivers then the license > should match or you are re-licensing the code you copied. Thanks, Changing efa_verbs.c to GPL-2.0 OR Linux-OpenIB.
diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c new file mode 100644 index 000000000000..a4ccebc9665d --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_main.c @@ -0,0 +1,577 @@ +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#include <linux/module.h> +#include <linux/pci.h> + +#include <rdma/ib_user_verbs.h> + +#include "efa.h" + +#define PCI_DEV_ID_EFA_VF 0xefa0 + +static const struct pci_device_id efa_pci_tbl[] = { + { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA_VF) }, + { } +}; + +MODULE_AUTHOR("Amazon.com, Inc. or its affiliates"); +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_DESCRIPTION(DEVICE_NAME); +MODULE_DEVICE_TABLE(pci, efa_pci_tbl); + +#define EFA_REG_BAR 0 +#define EFA_MEM_BAR 2 +#define EFA_BASE_BAR_MASK (BIT(EFA_REG_BAR) | BIT(EFA_MEM_BAR)) + +#define EFA_AENQ_ENABLED_GROUPS \ + (BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \ + BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE)) + +static void efa_update_network_attr(struct efa_dev *dev, + struct efa_com_get_network_attr_result *network_attr) +{ + memcpy(dev->addr, network_attr->addr, sizeof(network_attr->addr)); + dev->mtu = network_attr->mtu; + + efa_dbg(&dev->pdev->dev, "Full address %pI6\n", dev->addr); +} + +/* This handler will called for unknown event group or unimplemented handlers */ +static void unimplemented_aenq_handler(void *data, + struct efa_admin_aenq_entry *aenq_e) +{ + struct efa_dev *dev = (struct efa_dev *)data; + + efa_err_rl(&dev->ibdev.dev, + "Unknown event was received or event with unimplemented handler\n"); +} + +static void efa_keep_alive(void *data, struct efa_admin_aenq_entry *aenq_e) +{ + struct efa_dev *dev = (struct efa_dev *)data; + + efa_stat_inc(dev, dev->stats.keep_alive_rcvd); +} + +static struct efa_aenq_handlers aenq_handlers = { + .handlers = { + [EFA_ADMIN_KEEP_ALIVE] = efa_keep_alive, + }, + .unimplemented_handler = unimplemented_aenq_handler +}; + +static void efa_release_bars(struct efa_dev *dev, int bars_mask) +{ + struct pci_dev *pdev = dev->pdev; + int release_bars; + + release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & bars_mask; + pci_release_selected_regions(pdev, release_bars); +} + +static irqreturn_t efa_intr_msix_mgmnt(int irq, void *data) +{ + struct efa_dev *dev = (struct efa_dev *)data; + + efa_com_admin_q_comp_intr_handler(dev->edev); + + /* Don't call the aenq handler before probe is done */ + if (likely(test_bit(EFA_DEVICE_RUNNING_BIT, &dev->state))) + efa_com_aenq_intr_handler(dev->edev, data); + + return IRQ_HANDLED; +} + +static int efa_request_mgmnt_irq(struct efa_dev *dev) +{ + struct efa_irq *irq; + int err; + + irq = &dev->admin_irq; + err = request_irq(irq->vector, irq->handler, 0, irq->name, + irq->data); + if (err) { + efa_err(&dev->pdev->dev, "Failed to request admin irq (%d)\n", + err); + return err; + } + + efa_dbg(&dev->pdev->dev, "Set affinity hint of mgmnt irq to %*pbl (irq vector: %d)\n", + nr_cpumask_bits, &irq->affinity_hint_mask, irq->vector); + irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask); + + return err; +} + +static void efa_setup_mgmnt_irq(struct efa_dev *dev) +{ + u32 cpu; + + snprintf(dev->admin_irq.name, EFA_IRQNAME_SIZE, + "efa-mgmnt@pci:%s", pci_name(dev->pdev)); + dev->admin_irq.handler = efa_intr_msix_mgmnt; + dev->admin_irq.data = dev; + dev->admin_irq.vector = + pci_irq_vector(dev->pdev, dev->admin_msix_vector_idx); + cpu = cpumask_first(cpu_online_mask); + dev->admin_irq.cpu = cpu; + cpumask_set_cpu(cpu, + &dev->admin_irq.affinity_hint_mask); + efa_info(&dev->pdev->dev, "Setup irq:%p vector:%d name:%s\n", + &dev->admin_irq, + dev->admin_irq.vector, + dev->admin_irq.name); +} + +static void efa_free_mgmnt_irq(struct efa_dev *dev) +{ + struct efa_irq *irq; + + irq = &dev->admin_irq; + irq_set_affinity_hint(irq->vector, NULL); + free_irq(irq->vector, irq->data); +} + +static int efa_set_mgmnt_irq(struct efa_dev *dev) +{ + efa_setup_mgmnt_irq(dev); + + return efa_request_mgmnt_irq(dev); +} + +static int efa_request_doorbell_bar(struct efa_dev *dev) +{ + u8 db_bar_idx = dev->dev_attr.db_bar; + struct pci_dev *pdev = dev->pdev; + int bars; + int err; + + if (!(BIT(db_bar_idx) & EFA_BASE_BAR_MASK)) { + bars = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(db_bar_idx); + + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); + if (err) { + efa_err(&dev->pdev->dev, + "pci_request_selected_regions for bar %d failed %d\n", + db_bar_idx, err); + return err; + } + } + + dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx); + dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx); + + return 0; +} + +static void efa_release_doorbell_bar(struct efa_dev *dev) +{ + if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK)) + efa_release_bars(dev, BIT(dev->dev_attr.db_bar)); +} + +static void efa_update_hw_hints(struct efa_dev *dev, + struct efa_com_get_hw_hints_result *hw_hints) +{ + struct efa_com_dev *edev = dev->edev; + + if (hw_hints->mmio_read_timeout) + edev->mmio_read.mmio_read_timeout = + hw_hints->mmio_read_timeout * 1000; + + if (hw_hints->poll_interval) + edev->aq.poll_interval = hw_hints->poll_interval; + + if (hw_hints->admin_completion_timeout) + edev->aq.completion_timeout = + hw_hints->admin_completion_timeout; +} + +static const struct ib_device_ops efa_dev_ops = { + .alloc_pd = efa_alloc_pd, + .alloc_ucontext = efa_alloc_ucontext, + .create_ah = efa_create_ah, + .create_cq = efa_create_cq, + .create_qp = efa_create_qp, + .dealloc_pd = efa_dealloc_pd, + .dealloc_ucontext = efa_dealloc_ucontext, + .dereg_mr = efa_dereg_mr, + .destroy_ah = efa_destroy_ah, + .destroy_cq = efa_destroy_cq, + .destroy_qp = efa_destroy_qp, + .get_link_layer = efa_port_link_layer, + .get_port_immutable = efa_get_port_immutable, + .mmap = efa_mmap, + .modify_qp = efa_modify_qp, + .query_device = efa_query_device, + .query_gid = efa_query_gid, + .query_pkey = efa_query_pkey, + .query_port = efa_query_port, + .query_qp = efa_query_qp, + .reg_user_mr = efa_reg_mr, + INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd), +}; + +static int efa_ib_device_add(struct efa_dev *dev) +{ + struct efa_com_get_network_attr_result network_attr; + struct efa_com_get_hw_hints_result hw_hints; + struct pci_dev *pdev = dev->pdev; + int err; + + spin_lock_init(&dev->stats_lock); + + err = efa_com_get_device_attr(dev->edev, &dev->dev_attr); + if (err) + return err; + + efa_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n", dev->dev_attr.db_bar); + err = efa_request_doorbell_bar(dev); + if (err) + return err; + + err = efa_com_get_network_attr(dev->edev, &network_attr); + if (err) + goto err_release_doorbell_bar; + + efa_update_network_attr(dev, &network_attr); + + err = efa_com_get_hw_hints(dev->edev, &hw_hints); + if (err) + goto err_release_doorbell_bar; + + efa_update_hw_hints(dev, &hw_hints); + + /* Try to enable all the available aenq groups */ + err = efa_com_set_aenq_config(dev->edev, EFA_AENQ_ENABLED_GROUPS); + if (err) + goto err_release_doorbell_bar; + + dev->ibdev.owner = THIS_MODULE; + dev->ibdev.node_type = RDMA_NODE_EFA; + dev->ibdev.phys_port_cnt = 1; + dev->ibdev.num_comp_vectors = 1; + dev->ibdev.dev.parent = &pdev->dev; + dev->ibdev.uverbs_abi_ver = EFA_UVERBS_ABI_VERSION; + + dev->ibdev.uverbs_cmd_mask = + (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | + (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) | + (1ull << IB_USER_VERBS_CMD_QUERY_PORT) | + (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | + (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | + (1ull << IB_USER_VERBS_CMD_REG_MR) | + (1ull << IB_USER_VERBS_CMD_DEREG_MR) | + (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | + (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | + (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) | + (1ull << IB_USER_VERBS_CMD_CREATE_QP) | + (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | + (1ull << IB_USER_VERBS_CMD_QUERY_QP) | + (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | + (1ull << IB_USER_VERBS_CMD_CREATE_AH) | + (1ull << IB_USER_VERBS_CMD_DESTROY_AH); + + dev->ibdev.uverbs_ex_cmd_mask = + (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE); + + dev->ibdev.driver_id = RDMA_DRIVER_EFA; + ib_set_device_ops(&dev->ibdev, &efa_dev_ops); + + err = ib_register_device(&dev->ibdev, "efa_%d"); + if (err) + goto err_release_doorbell_bar; + + efa_info(&dev->ibdev.dev, "IB device registered\n"); + + set_bit(EFA_DEVICE_RUNNING_BIT, &dev->state); + + return 0; + +err_release_doorbell_bar: + efa_release_doorbell_bar(dev); + return err; +} + +static void efa_ib_device_remove(struct efa_dev *dev) +{ + + /* Reset the device only if the device is running. */ + if (test_bit(EFA_DEVICE_RUNNING_BIT, &dev->state)) + efa_com_dev_reset(dev->edev, EFA_REGS_RESET_NORMAL); + + efa_info(&dev->ibdev.dev, "Unregister ib device\n"); + ib_unregister_device(&dev->ibdev); + efa_release_doorbell_bar(dev); +} + +static void efa_disable_msix(struct efa_dev *dev) +{ + if (test_and_clear_bit(EFA_MSIX_ENABLED_BIT, &dev->state)) + pci_free_irq_vectors(dev->pdev); +} + +static int efa_enable_msix(struct efa_dev *dev) +{ + int msix_vecs, irq_num; + + if (test_bit(EFA_MSIX_ENABLED_BIT, &dev->state)) { + efa_err(&dev->pdev->dev, "Error, MSI-X is already enabled\n"); + return -EPERM; + } + + /* Reserve the max msix vectors we might need */ + msix_vecs = EFA_NUM_MSIX_VEC; + efa_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n", + msix_vecs); + + dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX; + irq_num = pci_alloc_irq_vectors(dev->pdev, msix_vecs, + msix_vecs, PCI_IRQ_MSIX); + + if (irq_num < 0) { + efa_err(&dev->pdev->dev, "Failed to enable MSI-X. irq_num %d\n", + irq_num); + return -ENOSPC; + } + + if (irq_num != msix_vecs) { + efa_warn(&dev->pdev->dev, + "Allocated %d MSI-X (out of %d requested)\n", + irq_num, msix_vecs); + return -ENOSPC; + } + + set_bit(EFA_MSIX_ENABLED_BIT, &dev->state); + + return 0; +} + +static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev) +{ + int dma_width; + int err; + + err = efa_com_dev_reset(edev, EFA_REGS_RESET_NORMAL); + if (err) + return err; + + err = efa_com_validate_version(edev); + if (err) + return err; + + dma_width = efa_com_get_dma_width(edev); + if (dma_width < 0) { + err = dma_width; + return err; + } + + err = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width)); + if (err) { + efa_err(&pdev->dev, "pci_set_dma_mask failed 0x%x\n", err); + return err; + } + + err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width)); + if (err) { + efa_err(&pdev->dev, + "err_pci_set_consistent_dma_mask failed 0x%x\n", + err); + return err; + } + + return 0; +} + +static int efa_probe_device(struct pci_dev *pdev) +{ + struct efa_com_dev *edev; + struct efa_dev *dev; + int bars; + int err; + + err = pci_enable_device_mem(pdev); + if (err) { + efa_err(&pdev->dev, "pci_enable_device_mem() failed!\n"); + return err; + } + + pci_set_master(pdev); + + dev = ib_alloc_device(efa_dev, ibdev); + if (IS_ERR_OR_NULL(dev)) { + efa_err(&pdev->dev, "Device alloc failed\n"); + err = dev ? PTR_ERR(dev) : -ENOMEM; + goto err_disable_device; + } + + edev = kzalloc(sizeof(*edev), GFP_KERNEL); + if (!edev) { + err = -ENOMEM; + goto err_ibdev_destroy; + } + + pci_set_drvdata(pdev, dev); + edev->dmadev = &pdev->dev; + dev->edev = edev; + dev->pdev = pdev; + + bars = pci_select_bars(pdev, IORESOURCE_MEM) & EFA_BASE_BAR_MASK; + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); + if (err) { + efa_err(&pdev->dev, "pci_request_selected_regions failed %d\n", + err); + goto err_free_efa_dev; + } + + dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR); + dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR); + dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR); + dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR); + + edev->reg_bar = devm_ioremap(&pdev->dev, + dev->reg_bar_addr, + dev->reg_bar_len); + if (!edev->reg_bar) { + efa_err(&pdev->dev, "Failed to remap register bar\n"); + err = -EFAULT; + goto err_release_bars; + } + + err = efa_com_mmio_reg_read_init(edev); + if (err) { + efa_err(&pdev->dev, "Failed to init readless MMIO\n"); + goto err_iounmap; + } + + err = efa_device_init(edev, pdev); + if (err) { + efa_err(&pdev->dev, "EFA device init failed\n"); + if (err == -ETIME) + err = -EPROBE_DEFER; + goto err_reg_read_destroy; + } + + err = efa_enable_msix(dev); + if (err) + goto err_reg_read_destroy; + + edev->aq.msix_vector_idx = dev->admin_msix_vector_idx; + edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx; + + err = efa_set_mgmnt_irq(dev); + if (err) + goto err_disable_msix; + + err = efa_com_admin_init(edev, &aenq_handlers); + if (err) + goto err_free_mgmnt_irq; + + return 0; + +err_free_mgmnt_irq: + efa_free_mgmnt_irq(dev); +err_disable_msix: + efa_disable_msix(dev); +err_reg_read_destroy: + efa_com_mmio_reg_read_destroy(edev); +err_iounmap: + devm_iounmap(&pdev->dev, edev->reg_bar); +err_release_bars: + efa_release_bars(dev, EFA_BASE_BAR_MASK); +err_free_efa_dev: + kfree(edev); +err_ibdev_destroy: + ib_dealloc_device(&dev->ibdev); +err_disable_device: + pci_disable_device(pdev); + return err; +} + +static void efa_remove_device(struct pci_dev *pdev) +{ + struct efa_dev *dev = pci_get_drvdata(pdev); + struct efa_com_dev *edev; + + if (!dev) + /* + * This device didn't load properly and its resources + * already released, nothing to do + */ + return; + + edev = dev->edev; + + efa_com_admin_destroy(edev); + efa_free_mgmnt_irq(dev); + efa_disable_msix(dev); + efa_com_mmio_reg_read_destroy(edev); + devm_iounmap(&pdev->dev, edev->reg_bar); + efa_release_bars(dev, EFA_BASE_BAR_MASK); + kfree(edev); + ib_dealloc_device(&dev->ibdev); + pci_disable_device(pdev); +} + +static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent) +{ + struct efa_dev *dev; + int err; + + err = efa_probe_device(pdev); + if (err) + return err; + + dev = pci_get_drvdata(pdev); + err = efa_ib_device_add(dev); + if (err) + goto err_remove_device; + + return 0; + +err_remove_device: + efa_remove_device(pdev); + return err; +} + +static void efa_remove(struct pci_dev *pdev) +{ + struct efa_dev *dev = (struct efa_dev *)pci_get_drvdata(pdev); + + efa_ib_device_remove(dev); + efa_remove_device(pdev); +} + +static struct pci_driver efa_pci_driver = { + .name = DRV_MODULE_NAME, + .id_table = efa_pci_tbl, + .probe = efa_probe, + .remove = efa_remove, +}; + +static int __init efa_init(void) +{ + int err; + + err = pci_register_driver(&efa_pci_driver); + if (err) { + pr_err("Couldn't register efa driver\n"); + goto err_register; + } + + return 0; + +err_register: + return err; +} + +static void __exit efa_exit(void) +{ + pci_unregister_driver(&efa_pci_driver); +} + +module_init(efa_init); +module_exit(efa_exit);
Add the main EFA module file which takes care of device probe/initialization/registration/etc. Signed-off-by: Gal Pressman <galpress@amazon.com> --- drivers/infiniband/hw/efa/efa_main.c | 577 +++++++++++++++++++++++++++++++++++ 1 file changed, 577 insertions(+) create mode 100644 drivers/infiniband/hw/efa/efa_main.c