| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * Copyright (c) 2022, Microsoft Corporation. All rights reserved. |
| 4 | */ |
| 5 | |
| 6 | #include "mana_ib.h" |
| 7 | #include <net/mana/mana_auxiliary.h> |
| 8 | #include <net/addrconf.h> |
| 9 | |
| 10 | MODULE_DESCRIPTION("Microsoft Azure Network Adapter IB driver" ); |
| 11 | MODULE_LICENSE("GPL" ); |
| 12 | MODULE_IMPORT_NS("NET_MANA" ); |
| 13 | |
| 14 | static const struct ib_device_ops mana_ib_dev_ops = { |
| 15 | .owner = THIS_MODULE, |
| 16 | .driver_id = RDMA_DRIVER_MANA, |
| 17 | .uverbs_abi_ver = MANA_IB_UVERBS_ABI_VERSION, |
| 18 | |
| 19 | .add_gid = mana_ib_gd_add_gid, |
| 20 | .alloc_pd = mana_ib_alloc_pd, |
| 21 | .alloc_ucontext = mana_ib_alloc_ucontext, |
| 22 | .create_ah = mana_ib_create_ah, |
| 23 | .create_cq = mana_ib_create_cq, |
| 24 | .create_qp = mana_ib_create_qp, |
| 25 | .create_rwq_ind_table = mana_ib_create_rwq_ind_table, |
| 26 | .create_wq = mana_ib_create_wq, |
| 27 | .dealloc_pd = mana_ib_dealloc_pd, |
| 28 | .dealloc_ucontext = mana_ib_dealloc_ucontext, |
| 29 | .del_gid = mana_ib_gd_del_gid, |
| 30 | .dereg_mr = mana_ib_dereg_mr, |
| 31 | .destroy_ah = mana_ib_destroy_ah, |
| 32 | .destroy_cq = mana_ib_destroy_cq, |
| 33 | .destroy_qp = mana_ib_destroy_qp, |
| 34 | .destroy_rwq_ind_table = mana_ib_destroy_rwq_ind_table, |
| 35 | .destroy_wq = mana_ib_destroy_wq, |
| 36 | .disassociate_ucontext = mana_ib_disassociate_ucontext, |
| 37 | .get_dma_mr = mana_ib_get_dma_mr, |
| 38 | .get_link_layer = mana_ib_get_link_layer, |
| 39 | .get_port_immutable = mana_ib_get_port_immutable, |
| 40 | .mmap = mana_ib_mmap, |
| 41 | .modify_qp = mana_ib_modify_qp, |
| 42 | .modify_wq = mana_ib_modify_wq, |
| 43 | .poll_cq = mana_ib_poll_cq, |
| 44 | .post_recv = mana_ib_post_recv, |
| 45 | .post_send = mana_ib_post_send, |
| 46 | .query_device = mana_ib_query_device, |
| 47 | .query_gid = mana_ib_query_gid, |
| 48 | .query_pkey = mana_ib_query_pkey, |
| 49 | .query_port = mana_ib_query_port, |
| 50 | .reg_user_mr = mana_ib_reg_user_mr, |
| 51 | .reg_user_mr_dmabuf = mana_ib_reg_user_mr_dmabuf, |
| 52 | .req_notify_cq = mana_ib_arm_cq, |
| 53 | |
| 54 | INIT_RDMA_OBJ_SIZE(ib_ah, mana_ib_ah, ibah), |
| 55 | INIT_RDMA_OBJ_SIZE(ib_cq, mana_ib_cq, ibcq), |
| 56 | INIT_RDMA_OBJ_SIZE(ib_pd, mana_ib_pd, ibpd), |
| 57 | INIT_RDMA_OBJ_SIZE(ib_qp, mana_ib_qp, ibqp), |
| 58 | INIT_RDMA_OBJ_SIZE(ib_ucontext, mana_ib_ucontext, ibucontext), |
| 59 | INIT_RDMA_OBJ_SIZE(ib_rwq_ind_table, mana_ib_rwq_ind_table, |
| 60 | ib_ind_table), |
| 61 | }; |
| 62 | |
| 63 | static const struct ib_device_ops mana_ib_stats_ops = { |
| 64 | .alloc_hw_port_stats = mana_ib_alloc_hw_port_stats, |
| 65 | .get_hw_stats = mana_ib_get_hw_stats, |
| 66 | }; |
| 67 | |
| 68 | static const struct ib_device_ops mana_ib_device_stats_ops = { |
| 69 | .alloc_hw_device_stats = mana_ib_alloc_hw_device_stats, |
| 70 | }; |
| 71 | |
| 72 | static int mana_ib_netdev_event(struct notifier_block *this, |
| 73 | unsigned long event, void *ptr) |
| 74 | { |
| 75 | struct mana_ib_dev *dev = container_of(this, struct mana_ib_dev, nb); |
| 76 | struct net_device *event_dev = netdev_notifier_info_to_dev(info: ptr); |
| 77 | struct gdma_context *gc = dev->gdma_dev->gdma_context; |
| 78 | struct mana_context *mc = gc->mana.driver_data; |
| 79 | struct net_device *ndev; |
| 80 | int i; |
| 81 | |
| 82 | /* Only process events from our parent device */ |
| 83 | for (i = 0; i < dev->ib_dev.phys_port_cnt; i++) |
| 84 | if (event_dev == mc->ports[i]) { |
| 85 | switch (event) { |
| 86 | case NETDEV_CHANGEUPPER: |
| 87 | ndev = mana_get_primary_netdev(ac: mc, port_index: i, tracker: &dev->dev_tracker); |
| 88 | /* |
| 89 | * RDMA core will setup GID based on updated netdev. |
| 90 | * It's not possible to race with the core as rtnl lock is being |
| 91 | * held. |
| 92 | */ |
| 93 | ib_device_set_netdev(ib_dev: &dev->ib_dev, ndev, port: i + 1); |
| 94 | |
| 95 | /* mana_get_primary_netdev() returns ndev with refcount held */ |
| 96 | if (ndev) |
| 97 | netdev_put(dev: ndev, tracker: &dev->dev_tracker); |
| 98 | |
| 99 | return NOTIFY_OK; |
| 100 | default: |
| 101 | return NOTIFY_DONE; |
| 102 | } |
| 103 | } |
| 104 | return NOTIFY_DONE; |
| 105 | } |
| 106 | |
| 107 | static int mana_ib_probe(struct auxiliary_device *adev, |
| 108 | const struct auxiliary_device_id *id) |
| 109 | { |
| 110 | struct mana_adev *madev = container_of(adev, struct mana_adev, adev); |
| 111 | struct gdma_context *gc = madev->mdev->gdma_context; |
| 112 | struct mana_context *mc = gc->mana.driver_data; |
| 113 | struct gdma_dev *mdev = madev->mdev; |
| 114 | struct net_device *ndev; |
| 115 | struct mana_ib_dev *dev; |
| 116 | u8 mac_addr[ETH_ALEN]; |
| 117 | int ret, i; |
| 118 | |
| 119 | dev = ib_alloc_device(mana_ib_dev, ib_dev); |
| 120 | if (!dev) |
| 121 | return -ENOMEM; |
| 122 | |
| 123 | ib_set_device_ops(device: &dev->ib_dev, ops: &mana_ib_dev_ops); |
| 124 | dev->ib_dev.node_type = RDMA_NODE_IB_CA; |
| 125 | dev->ib_dev.num_comp_vectors = gc->max_num_queues; |
| 126 | dev->ib_dev.dev.parent = gc->dev; |
| 127 | dev->gdma_dev = mdev; |
| 128 | xa_init_flags(xa: &dev->qp_table_wq, XA_FLAGS_LOCK_IRQ); |
| 129 | |
| 130 | if (mana_ib_is_rnic(mdev: dev)) { |
| 131 | dev->ib_dev.phys_port_cnt = 1; |
| 132 | addrconf_addr_eui48(eui: (u8 *)&dev->ib_dev.node_guid, addr: mc->ports[0]->dev_addr); |
| 133 | ret = mana_ib_gd_query_adapter_caps(mdev: dev); |
| 134 | if (ret) { |
| 135 | ibdev_err(ibdev: &dev->ib_dev, format: "Failed to query device caps, ret %d" , ret); |
| 136 | goto free_ib_device; |
| 137 | } |
| 138 | |
| 139 | ib_set_device_ops(device: &dev->ib_dev, ops: &mana_ib_stats_ops); |
| 140 | if (dev->adapter_caps.feature_flags & MANA_IB_FEATURE_DEV_COUNTERS_SUPPORT) |
| 141 | ib_set_device_ops(device: &dev->ib_dev, ops: &mana_ib_device_stats_ops); |
| 142 | |
| 143 | ret = mana_ib_create_eqs(mdev: dev); |
| 144 | if (ret) { |
| 145 | ibdev_err(ibdev: &dev->ib_dev, format: "Failed to create EQs, ret %d" , ret); |
| 146 | goto free_ib_device; |
| 147 | } |
| 148 | |
| 149 | ret = mana_ib_gd_create_rnic_adapter(mdev: dev); |
| 150 | if (ret) |
| 151 | goto destroy_eqs; |
| 152 | |
| 153 | if (dev->adapter_caps.feature_flags & MANA_IB_FEATURE_MULTI_PORTS_SUPPORT) |
| 154 | dev->ib_dev.phys_port_cnt = mc->num_ports; |
| 155 | |
| 156 | for (i = 0; i < dev->ib_dev.phys_port_cnt; i++) { |
| 157 | ndev = mana_get_primary_netdev(ac: mc, port_index: i, tracker: &dev->dev_tracker); |
| 158 | if (!ndev) { |
| 159 | ret = -ENODEV; |
| 160 | ibdev_err(ibdev: &dev->ib_dev, |
| 161 | format: "Failed to get netdev for IB port %d" , i + 1); |
| 162 | goto destroy_rnic; |
| 163 | } |
| 164 | ether_addr_copy(dst: mac_addr, src: ndev->dev_addr); |
| 165 | ret = ib_device_set_netdev(ib_dev: &dev->ib_dev, ndev, port: i + 1); |
| 166 | /* mana_get_primary_netdev() returns ndev with refcount held */ |
| 167 | netdev_put(dev: ndev, tracker: &dev->dev_tracker); |
| 168 | if (ret) { |
| 169 | ibdev_err(ibdev: &dev->ib_dev, format: "Failed to set ib netdev, ret %d" , ret); |
| 170 | goto destroy_rnic; |
| 171 | } |
| 172 | ret = mana_ib_gd_config_mac(mdev: dev, op: ADDR_OP_ADD, mac: mac_addr); |
| 173 | if (ret) { |
| 174 | ibdev_err(ibdev: &dev->ib_dev, format: "Failed to add Mac address, ret %d" , ret); |
| 175 | goto destroy_rnic; |
| 176 | } |
| 177 | } |
| 178 | dev->nb.notifier_call = mana_ib_netdev_event; |
| 179 | ret = register_netdevice_notifier(nb: &dev->nb); |
| 180 | if (ret) { |
| 181 | ibdev_err(ibdev: &dev->ib_dev, format: "Failed to register net notifier, %d" , ret); |
| 182 | goto destroy_rnic; |
| 183 | } |
| 184 | } else { |
| 185 | dev->ib_dev.phys_port_cnt = mc->num_ports; |
| 186 | ret = mana_eth_query_adapter_caps(mdev: dev); |
| 187 | if (ret) { |
| 188 | ibdev_err(ibdev: &dev->ib_dev, format: "Failed to query ETH device caps, ret %d" , ret); |
| 189 | goto free_ib_device; |
| 190 | } |
| 191 | } |
| 192 | |
| 193 | dev->av_pool = dma_pool_create(name: "mana_ib_av" , dev: gc->dev, MANA_AV_BUFFER_SIZE, |
| 194 | MANA_AV_BUFFER_SIZE, boundary: 0); |
| 195 | if (!dev->av_pool) { |
| 196 | ret = -ENOMEM; |
| 197 | goto deregister_net_notifier; |
| 198 | } |
| 199 | |
| 200 | ibdev_dbg(&dev->ib_dev, "mdev=%p id=%d num_ports=%d\n" , mdev, |
| 201 | mdev->dev_id.as_uint32, dev->ib_dev.phys_port_cnt); |
| 202 | |
| 203 | ret = ib_register_device(device: &dev->ib_dev, name: mana_ib_is_rnic(mdev: dev) ? "mana_%d" : "manae_%d" , |
| 204 | dma_device: gc->dev); |
| 205 | if (ret) |
| 206 | goto deallocate_pool; |
| 207 | |
| 208 | dev_set_drvdata(dev: &adev->dev, data: dev); |
| 209 | |
| 210 | return 0; |
| 211 | |
| 212 | deallocate_pool: |
| 213 | dma_pool_destroy(pool: dev->av_pool); |
| 214 | deregister_net_notifier: |
| 215 | if (mana_ib_is_rnic(mdev: dev)) |
| 216 | unregister_netdevice_notifier(nb: &dev->nb); |
| 217 | destroy_rnic: |
| 218 | if (mana_ib_is_rnic(mdev: dev)) |
| 219 | mana_ib_gd_destroy_rnic_adapter(mdev: dev); |
| 220 | destroy_eqs: |
| 221 | if (mana_ib_is_rnic(mdev: dev)) |
| 222 | mana_ib_destroy_eqs(mdev: dev); |
| 223 | free_ib_device: |
| 224 | xa_destroy(&dev->qp_table_wq); |
| 225 | ib_dealloc_device(device: &dev->ib_dev); |
| 226 | return ret; |
| 227 | } |
| 228 | |
| 229 | static void mana_ib_remove(struct auxiliary_device *adev) |
| 230 | { |
| 231 | struct mana_ib_dev *dev = dev_get_drvdata(dev: &adev->dev); |
| 232 | |
| 233 | if (mana_ib_is_rnic(mdev: dev)) |
| 234 | mana_drain_gsi_sqs(mdev: dev); |
| 235 | |
| 236 | ib_unregister_device(device: &dev->ib_dev); |
| 237 | dma_pool_destroy(pool: dev->av_pool); |
| 238 | if (mana_ib_is_rnic(mdev: dev)) { |
| 239 | unregister_netdevice_notifier(nb: &dev->nb); |
| 240 | mana_ib_gd_destroy_rnic_adapter(mdev: dev); |
| 241 | mana_ib_destroy_eqs(mdev: dev); |
| 242 | } |
| 243 | xa_destroy(&dev->qp_table_wq); |
| 244 | ib_dealloc_device(device: &dev->ib_dev); |
| 245 | } |
| 246 | |
| 247 | static const struct auxiliary_device_id mana_id_table[] = { |
| 248 | { .name = "mana.rdma" , }, |
| 249 | { .name = "mana.eth" , }, |
| 250 | {}, |
| 251 | }; |
| 252 | |
| 253 | MODULE_DEVICE_TABLE(auxiliary, mana_id_table); |
| 254 | |
| 255 | static struct auxiliary_driver mana_driver = { |
| 256 | .probe = mana_ib_probe, |
| 257 | .remove = mana_ib_remove, |
| 258 | .id_table = mana_id_table, |
| 259 | }; |
| 260 | |
| 261 | module_auxiliary_driver(mana_driver); |
| 262 | |