1 /* 2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved. 3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved. 4 * 5 * This software is available to you under a choice of one of two 6 * licenses. You may choose to be licensed under the terms of the GNU 7 * General Public License (GPL) Version 2, available from the file 8 * COPYING in the main directory of this source tree, or the 9 * OpenIB.org BSD license below: 10 * 11 * Redistribution and use in source and binary forms, with or 12 * without modification, are permitted provided that the following 13 * conditions are met: 14 * 15 * - Redistributions of source code must retain the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer. 18 * 19 * - Redistributions in binary form must reproduce the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer in the documentation and/or other materials 22 * provided with the distribution. 23 * 24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 31 * SOFTWARE. 32 */ 33 34 #include <linux/module.h> 35 #include <linux/init.h> 36 #include <linux/slab.h> 37 #include <linux/errno.h> 38 #include <linux/netdevice.h> 39 #include <linux/inetdevice.h> 40 #include <linux/rtnetlink.h> 41 #include <linux/if_vlan.h> 42 #include <net/ipv6.h> 43 #include <net/addrconf.h> 44 45 #include <rdma/ib_smi.h> 46 #include <rdma/ib_user_verbs.h> 47 #include <rdma/ib_addr.h> 48 49 #include <linux/mlx4/driver.h> 50 #include <linux/mlx4/cmd.h> 51 #include <linux/mlx4/qp.h> 52 53 #include "mlx4_ib.h" 54 #include "user.h" 55 56 #define DRV_NAME MLX4_IB_DRV_NAME 57 #define DRV_VERSION "2.2-1" 58 #define DRV_RELDATE "Feb 2014" 59 60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF 61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF 62 63 MODULE_AUTHOR("Roland Dreier"); 64 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver"); 65 MODULE_LICENSE("Dual BSD/GPL"); 66 MODULE_VERSION(DRV_VERSION); 67 68 int mlx4_ib_sm_guid_assign = 1; 69 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444); 70 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)"); 71 72 static const char mlx4_ib_version[] = 73 DRV_NAME ": Mellanox ConnectX InfiniBand driver v" 74 DRV_VERSION " (" DRV_RELDATE ")\n"; 75 76 struct update_gid_work { 77 struct work_struct work; 78 union ib_gid gids[128]; 79 struct mlx4_ib_dev *dev; 80 int port; 81 }; 82 83 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init); 84 85 static struct workqueue_struct *wq; 86 87 static void init_query_mad(struct ib_smp *mad) 88 { 89 mad->base_version = 1; 90 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED; 91 mad->class_version = 1; 92 mad->method = IB_MGMT_METHOD_GET; 93 } 94 95 static union ib_gid zgid; 96 97 static int check_flow_steering_support(struct mlx4_dev *dev) 98 { 99 int eth_num_ports = 0; 100 int ib_num_ports = 0; 101 102 int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED; 103 104 if (dmfs) { 105 int i; 106 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) 107 eth_num_ports++; 108 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) 109 ib_num_ports++; 110 dmfs &= (!ib_num_ports || 111 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) && 112 (!eth_num_ports || 113 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN)); 114 if (ib_num_ports && mlx4_is_mfunc(dev)) { 115 pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n"); 116 dmfs = 0; 117 } 118 } 119 return dmfs; 120 } 121 122 static int mlx4_ib_query_device(struct ib_device *ibdev, 123 struct ib_device_attr *props) 124 { 125 struct mlx4_ib_dev *dev = to_mdev(ibdev); 126 struct ib_smp *in_mad = NULL; 127 struct ib_smp *out_mad = NULL; 128 int err = -ENOMEM; 129 130 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 131 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 132 if (!in_mad || !out_mad) 133 goto out; 134 135 init_query_mad(in_mad); 136 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO; 137 138 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, 139 1, NULL, NULL, in_mad, out_mad); 140 if (err) 141 goto out; 142 143 memset(props, 0, sizeof *props); 144 145 props->fw_ver = dev->dev->caps.fw_ver; 146 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT | 147 IB_DEVICE_PORT_ACTIVE_EVENT | 148 IB_DEVICE_SYS_IMAGE_GUID | 149 IB_DEVICE_RC_RNR_NAK_GEN | 150 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK; 151 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR) 152 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR; 153 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR) 154 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR; 155 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM) 156 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG; 157 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT) 158 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE; 159 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM) 160 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM; 161 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH) 162 props->device_cap_flags |= IB_DEVICE_UD_TSO; 163 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY) 164 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY; 165 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) && 166 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) && 167 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR)) 168 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS; 169 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) 170 props->device_cap_flags |= IB_DEVICE_XRC; 171 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW) 172 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW; 173 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) { 174 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B) 175 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B; 176 else 177 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A; 178 if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) 179 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING; 180 } 181 182 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) & 183 0xffffff; 184 props->vendor_part_id = dev->dev->pdev->device; 185 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32)); 186 memcpy(&props->sys_image_guid, out_mad->data + 4, 8); 187 188 props->max_mr_size = ~0ull; 189 props->page_size_cap = dev->dev->caps.page_size_cap; 190 props->max_qp = dev->dev->quotas.qp; 191 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE; 192 props->max_sge = min(dev->dev->caps.max_sq_sg, 193 dev->dev->caps.max_rq_sg); 194 props->max_cq = dev->dev->quotas.cq; 195 props->max_cqe = dev->dev->caps.max_cqes; 196 props->max_mr = dev->dev->quotas.mpt; 197 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds; 198 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma; 199 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma; 200 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp; 201 props->max_srq = dev->dev->quotas.srq; 202 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1; 203 props->max_srq_sge = dev->dev->caps.max_srq_sge; 204 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES; 205 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay; 206 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ? 207 IB_ATOMIC_HCA : IB_ATOMIC_NONE; 208 props->masked_atomic_cap = props->atomic_cap; 209 props->max_pkeys = dev->dev->caps.pkey_table_len[1]; 210 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms; 211 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm; 212 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach * 213 props->max_mcast_grp; 214 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps; 215 216 out: 217 kfree(in_mad); 218 kfree(out_mad); 219 220 return err; 221 } 222 223 static enum rdma_link_layer 224 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num) 225 { 226 struct mlx4_dev *dev = to_mdev(device)->dev; 227 228 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ? 229 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET; 230 } 231 232 static int ib_link_query_port(struct ib_device *ibdev, u8 port, 233 struct ib_port_attr *props, int netw_view) 234 { 235 struct ib_smp *in_mad = NULL; 236 struct ib_smp *out_mad = NULL; 237 int ext_active_speed; 238 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; 239 int err = -ENOMEM; 240 241 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 242 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 243 if (!in_mad || !out_mad) 244 goto out; 245 246 init_query_mad(in_mad); 247 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO; 248 in_mad->attr_mod = cpu_to_be32(port); 249 250 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view) 251 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; 252 253 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL, 254 in_mad, out_mad); 255 if (err) 256 goto out; 257 258 259 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16)); 260 props->lmc = out_mad->data[34] & 0x7; 261 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18)); 262 props->sm_sl = out_mad->data[36] & 0xf; 263 props->state = out_mad->data[32] & 0xf; 264 props->phys_state = out_mad->data[33] >> 4; 265 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20)); 266 if (netw_view) 267 props->gid_tbl_len = out_mad->data[50]; 268 else 269 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port]; 270 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz; 271 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port]; 272 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46)); 273 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48)); 274 props->active_width = out_mad->data[31] & 0xf; 275 props->active_speed = out_mad->data[35] >> 4; 276 props->max_mtu = out_mad->data[41] & 0xf; 277 props->active_mtu = out_mad->data[36] >> 4; 278 props->subnet_timeout = out_mad->data[51] & 0x1f; 279 props->max_vl_num = out_mad->data[37] >> 4; 280 props->init_type_reply = out_mad->data[41] >> 4; 281 282 /* Check if extended speeds (EDR/FDR/...) are supported */ 283 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) { 284 ext_active_speed = out_mad->data[62] >> 4; 285 286 switch (ext_active_speed) { 287 case 1: 288 props->active_speed = IB_SPEED_FDR; 289 break; 290 case 2: 291 props->active_speed = IB_SPEED_EDR; 292 break; 293 } 294 } 295 296 /* If reported active speed is QDR, check if is FDR-10 */ 297 if (props->active_speed == IB_SPEED_QDR) { 298 init_query_mad(in_mad); 299 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO; 300 in_mad->attr_mod = cpu_to_be32(port); 301 302 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, 303 NULL, NULL, in_mad, out_mad); 304 if (err) 305 goto out; 306 307 /* Checking LinkSpeedActive for FDR-10 */ 308 if (out_mad->data[15] & 0x1) 309 props->active_speed = IB_SPEED_FDR10; 310 } 311 312 /* Avoid wrong speed value returned by FW if the IB link is down. */ 313 if (props->state == IB_PORT_DOWN) 314 props->active_speed = IB_SPEED_SDR; 315 316 out: 317 kfree(in_mad); 318 kfree(out_mad); 319 return err; 320 } 321 322 static u8 state_to_phys_state(enum ib_port_state state) 323 { 324 return state == IB_PORT_ACTIVE ? 5 : 3; 325 } 326 327 static int eth_link_query_port(struct ib_device *ibdev, u8 port, 328 struct ib_port_attr *props, int netw_view) 329 { 330 331 struct mlx4_ib_dev *mdev = to_mdev(ibdev); 332 struct mlx4_ib_iboe *iboe = &mdev->iboe; 333 struct net_device *ndev; 334 enum ib_mtu tmp; 335 struct mlx4_cmd_mailbox *mailbox; 336 int err = 0; 337 338 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev); 339 if (IS_ERR(mailbox)) 340 return PTR_ERR(mailbox); 341 342 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0, 343 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B, 344 MLX4_CMD_WRAPPED); 345 if (err) 346 goto out; 347 348 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ? 349 IB_WIDTH_4X : IB_WIDTH_1X; 350 props->active_speed = IB_SPEED_QDR; 351 props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS; 352 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port]; 353 props->max_msg_sz = mdev->dev->caps.max_msg_sz; 354 props->pkey_tbl_len = 1; 355 props->max_mtu = IB_MTU_4096; 356 props->max_vl_num = 2; 357 props->state = IB_PORT_DOWN; 358 props->phys_state = state_to_phys_state(props->state); 359 props->active_mtu = IB_MTU_256; 360 spin_lock(&iboe->lock); 361 ndev = iboe->netdevs[port - 1]; 362 if (!ndev) 363 goto out_unlock; 364 365 tmp = iboe_get_mtu(ndev->mtu); 366 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256; 367 368 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ? 369 IB_PORT_ACTIVE : IB_PORT_DOWN; 370 props->phys_state = state_to_phys_state(props->state); 371 out_unlock: 372 spin_unlock(&iboe->lock); 373 out: 374 mlx4_free_cmd_mailbox(mdev->dev, mailbox); 375 return err; 376 } 377 378 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port, 379 struct ib_port_attr *props, int netw_view) 380 { 381 int err; 382 383 memset(props, 0, sizeof *props); 384 385 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ? 386 ib_link_query_port(ibdev, port, props, netw_view) : 387 eth_link_query_port(ibdev, port, props, netw_view); 388 389 return err; 390 } 391 392 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port, 393 struct ib_port_attr *props) 394 { 395 /* returns host view */ 396 return __mlx4_ib_query_port(ibdev, port, props, 0); 397 } 398 399 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, 400 union ib_gid *gid, int netw_view) 401 { 402 struct ib_smp *in_mad = NULL; 403 struct ib_smp *out_mad = NULL; 404 int err = -ENOMEM; 405 struct mlx4_ib_dev *dev = to_mdev(ibdev); 406 int clear = 0; 407 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; 408 409 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 410 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 411 if (!in_mad || !out_mad) 412 goto out; 413 414 init_query_mad(in_mad); 415 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO; 416 in_mad->attr_mod = cpu_to_be32(port); 417 418 if (mlx4_is_mfunc(dev->dev) && netw_view) 419 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; 420 421 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad); 422 if (err) 423 goto out; 424 425 memcpy(gid->raw, out_mad->data + 8, 8); 426 427 if (mlx4_is_mfunc(dev->dev) && !netw_view) { 428 if (index) { 429 /* For any index > 0, return the null guid */ 430 err = 0; 431 clear = 1; 432 goto out; 433 } 434 } 435 436 init_query_mad(in_mad); 437 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO; 438 in_mad->attr_mod = cpu_to_be32(index / 8); 439 440 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, 441 NULL, NULL, in_mad, out_mad); 442 if (err) 443 goto out; 444 445 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8); 446 447 out: 448 if (clear) 449 memset(gid->raw + 8, 0, 8); 450 kfree(in_mad); 451 kfree(out_mad); 452 return err; 453 } 454 455 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index, 456 union ib_gid *gid) 457 { 458 struct mlx4_ib_dev *dev = to_mdev(ibdev); 459 460 *gid = dev->iboe.gid_table[port - 1][index]; 461 462 return 0; 463 } 464 465 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, 466 union ib_gid *gid) 467 { 468 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND) 469 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0); 470 else 471 return iboe_query_gid(ibdev, port, index, gid); 472 } 473 474 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, 475 u16 *pkey, int netw_view) 476 { 477 struct ib_smp *in_mad = NULL; 478 struct ib_smp *out_mad = NULL; 479 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; 480 int err = -ENOMEM; 481 482 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 483 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 484 if (!in_mad || !out_mad) 485 goto out; 486 487 init_query_mad(in_mad); 488 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE; 489 in_mad->attr_mod = cpu_to_be32(index / 32); 490 491 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view) 492 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; 493 494 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL, 495 in_mad, out_mad); 496 if (err) 497 goto out; 498 499 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]); 500 501 out: 502 kfree(in_mad); 503 kfree(out_mad); 504 return err; 505 } 506 507 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey) 508 { 509 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0); 510 } 511 512 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask, 513 struct ib_device_modify *props) 514 { 515 struct mlx4_cmd_mailbox *mailbox; 516 unsigned long flags; 517 518 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC) 519 return -EOPNOTSUPP; 520 521 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC)) 522 return 0; 523 524 if (mlx4_is_slave(to_mdev(ibdev)->dev)) 525 return -EOPNOTSUPP; 526 527 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags); 528 memcpy(ibdev->node_desc, props->node_desc, 64); 529 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags); 530 531 /* 532 * If possible, pass node desc to FW, so it can generate 533 * a 144 trap. If cmd fails, just ignore. 534 */ 535 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev); 536 if (IS_ERR(mailbox)) 537 return 0; 538 539 memcpy(mailbox->buf, props->node_desc, 64); 540 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0, 541 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE); 542 543 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox); 544 545 return 0; 546 } 547 548 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols, 549 u32 cap_mask) 550 { 551 struct mlx4_cmd_mailbox *mailbox; 552 int err; 553 554 mailbox = mlx4_alloc_cmd_mailbox(dev->dev); 555 if (IS_ERR(mailbox)) 556 return PTR_ERR(mailbox); 557 558 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) { 559 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6; 560 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask); 561 } else { 562 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols; 563 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask); 564 } 565 566 err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT, 567 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED); 568 569 mlx4_free_cmd_mailbox(dev->dev, mailbox); 570 return err; 571 } 572 573 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask, 574 struct ib_port_modify *props) 575 { 576 struct mlx4_ib_dev *mdev = to_mdev(ibdev); 577 u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH; 578 struct ib_port_attr attr; 579 u32 cap_mask; 580 int err; 581 582 /* return OK if this is RoCE. CM calls ib_modify_port() regardless 583 * of whether port link layer is ETH or IB. For ETH ports, qkey 584 * violations and port capabilities are not meaningful. 585 */ 586 if (is_eth) 587 return 0; 588 589 mutex_lock(&mdev->cap_mask_mutex); 590 591 err = mlx4_ib_query_port(ibdev, port, &attr); 592 if (err) 593 goto out; 594 595 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) & 596 ~props->clr_port_cap_mask; 597 598 err = mlx4_ib_SET_PORT(mdev, port, 599 !!(mask & IB_PORT_RESET_QKEY_CNTR), 600 cap_mask); 601 602 out: 603 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex); 604 return err; 605 } 606 607 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev, 608 struct ib_udata *udata) 609 { 610 struct mlx4_ib_dev *dev = to_mdev(ibdev); 611 struct mlx4_ib_ucontext *context; 612 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3; 613 struct mlx4_ib_alloc_ucontext_resp resp; 614 int err; 615 616 if (!dev->ib_active) 617 return ERR_PTR(-EAGAIN); 618 619 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) { 620 resp_v3.qp_tab_size = dev->dev->caps.num_qps; 621 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size; 622 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page; 623 } else { 624 resp.dev_caps = dev->dev->caps.userspace_caps; 625 resp.qp_tab_size = dev->dev->caps.num_qps; 626 resp.bf_reg_size = dev->dev->caps.bf_reg_size; 627 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page; 628 resp.cqe_size = dev->dev->caps.cqe_size; 629 } 630 631 context = kmalloc(sizeof *context, GFP_KERNEL); 632 if (!context) 633 return ERR_PTR(-ENOMEM); 634 635 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar); 636 if (err) { 637 kfree(context); 638 return ERR_PTR(err); 639 } 640 641 INIT_LIST_HEAD(&context->db_page_list); 642 mutex_init(&context->db_page_mutex); 643 644 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) 645 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3)); 646 else 647 err = ib_copy_to_udata(udata, &resp, sizeof(resp)); 648 649 if (err) { 650 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar); 651 kfree(context); 652 return ERR_PTR(-EFAULT); 653 } 654 655 return &context->ibucontext; 656 } 657 658 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) 659 { 660 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext); 661 662 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar); 663 kfree(context); 664 665 return 0; 666 } 667 668 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) 669 { 670 struct mlx4_ib_dev *dev = to_mdev(context->device); 671 672 if (vma->vm_end - vma->vm_start != PAGE_SIZE) 673 return -EINVAL; 674 675 if (vma->vm_pgoff == 0) { 676 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 677 678 if (io_remap_pfn_range(vma, vma->vm_start, 679 to_mucontext(context)->uar.pfn, 680 PAGE_SIZE, vma->vm_page_prot)) 681 return -EAGAIN; 682 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) { 683 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot); 684 685 if (io_remap_pfn_range(vma, vma->vm_start, 686 to_mucontext(context)->uar.pfn + 687 dev->dev->caps.num_uars, 688 PAGE_SIZE, vma->vm_page_prot)) 689 return -EAGAIN; 690 } else 691 return -EINVAL; 692 693 return 0; 694 } 695 696 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev, 697 struct ib_ucontext *context, 698 struct ib_udata *udata) 699 { 700 struct mlx4_ib_pd *pd; 701 int err; 702 703 pd = kmalloc(sizeof *pd, GFP_KERNEL); 704 if (!pd) 705 return ERR_PTR(-ENOMEM); 706 707 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn); 708 if (err) { 709 kfree(pd); 710 return ERR_PTR(err); 711 } 712 713 if (context) 714 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) { 715 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn); 716 kfree(pd); 717 return ERR_PTR(-EFAULT); 718 } 719 720 return &pd->ibpd; 721 } 722 723 static int mlx4_ib_dealloc_pd(struct ib_pd *pd) 724 { 725 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn); 726 kfree(pd); 727 728 return 0; 729 } 730 731 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev, 732 struct ib_ucontext *context, 733 struct ib_udata *udata) 734 { 735 struct mlx4_ib_xrcd *xrcd; 736 int err; 737 738 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)) 739 return ERR_PTR(-ENOSYS); 740 741 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL); 742 if (!xrcd) 743 return ERR_PTR(-ENOMEM); 744 745 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn); 746 if (err) 747 goto err1; 748 749 xrcd->pd = ib_alloc_pd(ibdev); 750 if (IS_ERR(xrcd->pd)) { 751 err = PTR_ERR(xrcd->pd); 752 goto err2; 753 } 754 755 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0); 756 if (IS_ERR(xrcd->cq)) { 757 err = PTR_ERR(xrcd->cq); 758 goto err3; 759 } 760 761 return &xrcd->ibxrcd; 762 763 err3: 764 ib_dealloc_pd(xrcd->pd); 765 err2: 766 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn); 767 err1: 768 kfree(xrcd); 769 return ERR_PTR(err); 770 } 771 772 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd) 773 { 774 ib_destroy_cq(to_mxrcd(xrcd)->cq); 775 ib_dealloc_pd(to_mxrcd(xrcd)->pd); 776 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn); 777 kfree(xrcd); 778 779 return 0; 780 } 781 782 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid) 783 { 784 struct mlx4_ib_qp *mqp = to_mqp(ibqp); 785 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device); 786 struct mlx4_ib_gid_entry *ge; 787 788 ge = kzalloc(sizeof *ge, GFP_KERNEL); 789 if (!ge) 790 return -ENOMEM; 791 792 ge->gid = *gid; 793 if (mlx4_ib_add_mc(mdev, mqp, gid)) { 794 ge->port = mqp->port; 795 ge->added = 1; 796 } 797 798 mutex_lock(&mqp->mutex); 799 list_add_tail(&ge->list, &mqp->gid_list); 800 mutex_unlock(&mqp->mutex); 801 802 return 0; 803 } 804 805 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp, 806 union ib_gid *gid) 807 { 808 struct net_device *ndev; 809 int ret = 0; 810 811 if (!mqp->port) 812 return 0; 813 814 spin_lock(&mdev->iboe.lock); 815 ndev = mdev->iboe.netdevs[mqp->port - 1]; 816 if (ndev) 817 dev_hold(ndev); 818 spin_unlock(&mdev->iboe.lock); 819 820 if (ndev) { 821 ret = 1; 822 dev_put(ndev); 823 } 824 825 return ret; 826 } 827 828 struct mlx4_ib_steering { 829 struct list_head list; 830 u64 reg_id; 831 union ib_gid gid; 832 }; 833 834 static int parse_flow_attr(struct mlx4_dev *dev, 835 u32 qp_num, 836 union ib_flow_spec *ib_spec, 837 struct _rule_hw *mlx4_spec) 838 { 839 enum mlx4_net_trans_rule_id type; 840 841 switch (ib_spec->type) { 842 case IB_FLOW_SPEC_ETH: 843 type = MLX4_NET_TRANS_RULE_ID_ETH; 844 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac, 845 ETH_ALEN); 846 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac, 847 ETH_ALEN); 848 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag; 849 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag; 850 break; 851 case IB_FLOW_SPEC_IB: 852 type = MLX4_NET_TRANS_RULE_ID_IB; 853 mlx4_spec->ib.l3_qpn = 854 cpu_to_be32(qp_num); 855 mlx4_spec->ib.qpn_mask = 856 cpu_to_be32(MLX4_IB_FLOW_QPN_MASK); 857 break; 858 859 860 case IB_FLOW_SPEC_IPV4: 861 type = MLX4_NET_TRANS_RULE_ID_IPV4; 862 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip; 863 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip; 864 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip; 865 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip; 866 break; 867 868 case IB_FLOW_SPEC_TCP: 869 case IB_FLOW_SPEC_UDP: 870 type = ib_spec->type == IB_FLOW_SPEC_TCP ? 871 MLX4_NET_TRANS_RULE_ID_TCP : 872 MLX4_NET_TRANS_RULE_ID_UDP; 873 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port; 874 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port; 875 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port; 876 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port; 877 break; 878 879 default: 880 return -EINVAL; 881 } 882 if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 || 883 mlx4_hw_rule_sz(dev, type) < 0) 884 return -EINVAL; 885 mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type)); 886 mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2; 887 return mlx4_hw_rule_sz(dev, type); 888 } 889 890 struct default_rules { 891 __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS]; 892 __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS]; 893 __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS]; 894 __u8 link_layer; 895 }; 896 static const struct default_rules default_table[] = { 897 { 898 .mandatory_fields = {IB_FLOW_SPEC_IPV4}, 899 .mandatory_not_fields = {IB_FLOW_SPEC_ETH}, 900 .rules_create_list = {IB_FLOW_SPEC_IB}, 901 .link_layer = IB_LINK_LAYER_INFINIBAND 902 } 903 }; 904 905 static int __mlx4_ib_default_rules_match(struct ib_qp *qp, 906 struct ib_flow_attr *flow_attr) 907 { 908 int i, j, k; 909 void *ib_flow; 910 const struct default_rules *pdefault_rules = default_table; 911 u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port); 912 913 for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) { 914 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS]; 915 memset(&field_types, 0, sizeof(field_types)); 916 917 if (link_layer != pdefault_rules->link_layer) 918 continue; 919 920 ib_flow = flow_attr + 1; 921 /* we assume the specs are sorted */ 922 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS && 923 j < flow_attr->num_of_specs; k++) { 924 union ib_flow_spec *current_flow = 925 (union ib_flow_spec *)ib_flow; 926 927 /* same layer but different type */ 928 if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) == 929 (pdefault_rules->mandatory_fields[k] & 930 IB_FLOW_SPEC_LAYER_MASK)) && 931 (current_flow->type != 932 pdefault_rules->mandatory_fields[k])) 933 goto out; 934 935 /* same layer, try match next one */ 936 if (current_flow->type == 937 pdefault_rules->mandatory_fields[k]) { 938 j++; 939 ib_flow += 940 ((union ib_flow_spec *)ib_flow)->size; 941 } 942 } 943 944 ib_flow = flow_attr + 1; 945 for (j = 0; j < flow_attr->num_of_specs; 946 j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size) 947 for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++) 948 /* same layer and same type */ 949 if (((union ib_flow_spec *)ib_flow)->type == 950 pdefault_rules->mandatory_not_fields[k]) 951 goto out; 952 953 return i; 954 } 955 out: 956 return -1; 957 } 958 959 static int __mlx4_ib_create_default_rules( 960 struct mlx4_ib_dev *mdev, 961 struct ib_qp *qp, 962 const struct default_rules *pdefault_rules, 963 struct _rule_hw *mlx4_spec) { 964 int size = 0; 965 int i; 966 967 for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) { 968 int ret; 969 union ib_flow_spec ib_spec; 970 switch (pdefault_rules->rules_create_list[i]) { 971 case 0: 972 /* no rule */ 973 continue; 974 case IB_FLOW_SPEC_IB: 975 ib_spec.type = IB_FLOW_SPEC_IB; 976 ib_spec.size = sizeof(struct ib_flow_spec_ib); 977 978 break; 979 default: 980 /* invalid rule */ 981 return -EINVAL; 982 } 983 /* We must put empty rule, qpn is being ignored */ 984 ret = parse_flow_attr(mdev->dev, 0, &ib_spec, 985 mlx4_spec); 986 if (ret < 0) { 987 pr_info("invalid parsing\n"); 988 return -EINVAL; 989 } 990 991 mlx4_spec = (void *)mlx4_spec + ret; 992 size += ret; 993 } 994 return size; 995 } 996 997 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr, 998 int domain, 999 enum mlx4_net_trans_promisc_mode flow_type, 1000 u64 *reg_id) 1001 { 1002 int ret, i; 1003 int size = 0; 1004 void *ib_flow; 1005 struct mlx4_ib_dev *mdev = to_mdev(qp->device); 1006 struct mlx4_cmd_mailbox *mailbox; 1007 struct mlx4_net_trans_rule_hw_ctrl *ctrl; 1008 int default_flow; 1009 1010 static const u16 __mlx4_domain[] = { 1011 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS, 1012 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL, 1013 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS, 1014 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC, 1015 }; 1016 1017 if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) { 1018 pr_err("Invalid priority value %d\n", flow_attr->priority); 1019 return -EINVAL; 1020 } 1021 1022 if (domain >= IB_FLOW_DOMAIN_NUM) { 1023 pr_err("Invalid domain value %d\n", domain); 1024 return -EINVAL; 1025 } 1026 1027 if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0) 1028 return -EINVAL; 1029 1030 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev); 1031 if (IS_ERR(mailbox)) 1032 return PTR_ERR(mailbox); 1033 ctrl = mailbox->buf; 1034 1035 ctrl->prio = cpu_to_be16(__mlx4_domain[domain] | 1036 flow_attr->priority); 1037 ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type); 1038 ctrl->port = flow_attr->port; 1039 ctrl->qpn = cpu_to_be32(qp->qp_num); 1040 1041 ib_flow = flow_attr + 1; 1042 size += sizeof(struct mlx4_net_trans_rule_hw_ctrl); 1043 /* Add default flows */ 1044 default_flow = __mlx4_ib_default_rules_match(qp, flow_attr); 1045 if (default_flow >= 0) { 1046 ret = __mlx4_ib_create_default_rules( 1047 mdev, qp, default_table + default_flow, 1048 mailbox->buf + size); 1049 if (ret < 0) { 1050 mlx4_free_cmd_mailbox(mdev->dev, mailbox); 1051 return -EINVAL; 1052 } 1053 size += ret; 1054 } 1055 for (i = 0; i < flow_attr->num_of_specs; i++) { 1056 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow, 1057 mailbox->buf + size); 1058 if (ret < 0) { 1059 mlx4_free_cmd_mailbox(mdev->dev, mailbox); 1060 return -EINVAL; 1061 } 1062 ib_flow += ((union ib_flow_spec *) ib_flow)->size; 1063 size += ret; 1064 } 1065 1066 ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0, 1067 MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A, 1068 MLX4_CMD_NATIVE); 1069 if (ret == -ENOMEM) 1070 pr_err("mcg table is full. Fail to register network rule.\n"); 1071 else if (ret == -ENXIO) 1072 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n"); 1073 else if (ret) 1074 pr_err("Invalid argumant. Fail to register network rule.\n"); 1075 1076 mlx4_free_cmd_mailbox(mdev->dev, mailbox); 1077 return ret; 1078 } 1079 1080 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id) 1081 { 1082 int err; 1083 err = mlx4_cmd(dev, reg_id, 0, 0, 1084 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A, 1085 MLX4_CMD_NATIVE); 1086 if (err) 1087 pr_err("Fail to detach network rule. registration id = 0x%llx\n", 1088 reg_id); 1089 return err; 1090 } 1091 1092 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr, 1093 u64 *reg_id) 1094 { 1095 void *ib_flow; 1096 union ib_flow_spec *ib_spec; 1097 struct mlx4_dev *dev = to_mdev(qp->device)->dev; 1098 int err = 0; 1099 1100 if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) 1101 return 0; /* do nothing */ 1102 1103 ib_flow = flow_attr + 1; 1104 ib_spec = (union ib_flow_spec *)ib_flow; 1105 1106 if (ib_spec->type != IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1) 1107 return 0; /* do nothing */ 1108 1109 err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac, 1110 flow_attr->port, qp->qp_num, 1111 MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff), 1112 reg_id); 1113 return err; 1114 } 1115 1116 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp, 1117 struct ib_flow_attr *flow_attr, 1118 int domain) 1119 { 1120 int err = 0, i = 0; 1121 struct mlx4_ib_flow *mflow; 1122 enum mlx4_net_trans_promisc_mode type[2]; 1123 1124 memset(type, 0, sizeof(type)); 1125 1126 mflow = kzalloc(sizeof(*mflow), GFP_KERNEL); 1127 if (!mflow) { 1128 err = -ENOMEM; 1129 goto err_free; 1130 } 1131 1132 switch (flow_attr->type) { 1133 case IB_FLOW_ATTR_NORMAL: 1134 type[0] = MLX4_FS_REGULAR; 1135 break; 1136 1137 case IB_FLOW_ATTR_ALL_DEFAULT: 1138 type[0] = MLX4_FS_ALL_DEFAULT; 1139 break; 1140 1141 case IB_FLOW_ATTR_MC_DEFAULT: 1142 type[0] = MLX4_FS_MC_DEFAULT; 1143 break; 1144 1145 case IB_FLOW_ATTR_SNIFFER: 1146 type[0] = MLX4_FS_UC_SNIFFER; 1147 type[1] = MLX4_FS_MC_SNIFFER; 1148 break; 1149 1150 default: 1151 err = -EINVAL; 1152 goto err_free; 1153 } 1154 1155 while (i < ARRAY_SIZE(type) && type[i]) { 1156 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i], 1157 &mflow->reg_id[i]); 1158 if (err) 1159 goto err_free; 1160 i++; 1161 } 1162 1163 if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) { 1164 err = mlx4_ib_tunnel_steer_add(qp, flow_attr, &mflow->reg_id[i]); 1165 if (err) 1166 goto err_free; 1167 } 1168 1169 return &mflow->ibflow; 1170 1171 err_free: 1172 kfree(mflow); 1173 return ERR_PTR(err); 1174 } 1175 1176 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id) 1177 { 1178 int err, ret = 0; 1179 int i = 0; 1180 struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device); 1181 struct mlx4_ib_flow *mflow = to_mflow(flow_id); 1182 1183 while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) { 1184 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]); 1185 if (err) 1186 ret = err; 1187 i++; 1188 } 1189 1190 kfree(mflow); 1191 return ret; 1192 } 1193 1194 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid) 1195 { 1196 int err; 1197 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device); 1198 struct mlx4_ib_qp *mqp = to_mqp(ibqp); 1199 u64 reg_id; 1200 struct mlx4_ib_steering *ib_steering = NULL; 1201 enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ? 1202 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6; 1203 1204 if (mdev->dev->caps.steering_mode == 1205 MLX4_STEERING_MODE_DEVICE_MANAGED) { 1206 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL); 1207 if (!ib_steering) 1208 return -ENOMEM; 1209 } 1210 1211 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port, 1212 !!(mqp->flags & 1213 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK), 1214 prot, ®_id); 1215 if (err) 1216 goto err_malloc; 1217 1218 err = add_gid_entry(ibqp, gid); 1219 if (err) 1220 goto err_add; 1221 1222 if (ib_steering) { 1223 memcpy(ib_steering->gid.raw, gid->raw, 16); 1224 ib_steering->reg_id = reg_id; 1225 mutex_lock(&mqp->mutex); 1226 list_add(&ib_steering->list, &mqp->steering_rules); 1227 mutex_unlock(&mqp->mutex); 1228 } 1229 return 0; 1230 1231 err_add: 1232 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw, 1233 prot, reg_id); 1234 err_malloc: 1235 kfree(ib_steering); 1236 1237 return err; 1238 } 1239 1240 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw) 1241 { 1242 struct mlx4_ib_gid_entry *ge; 1243 struct mlx4_ib_gid_entry *tmp; 1244 struct mlx4_ib_gid_entry *ret = NULL; 1245 1246 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) { 1247 if (!memcmp(raw, ge->gid.raw, 16)) { 1248 ret = ge; 1249 break; 1250 } 1251 } 1252 1253 return ret; 1254 } 1255 1256 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid) 1257 { 1258 int err; 1259 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device); 1260 struct mlx4_ib_qp *mqp = to_mqp(ibqp); 1261 struct net_device *ndev; 1262 struct mlx4_ib_gid_entry *ge; 1263 u64 reg_id = 0; 1264 enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ? 1265 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6; 1266 1267 if (mdev->dev->caps.steering_mode == 1268 MLX4_STEERING_MODE_DEVICE_MANAGED) { 1269 struct mlx4_ib_steering *ib_steering; 1270 1271 mutex_lock(&mqp->mutex); 1272 list_for_each_entry(ib_steering, &mqp->steering_rules, list) { 1273 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) { 1274 list_del(&ib_steering->list); 1275 break; 1276 } 1277 } 1278 mutex_unlock(&mqp->mutex); 1279 if (&ib_steering->list == &mqp->steering_rules) { 1280 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n"); 1281 return -EINVAL; 1282 } 1283 reg_id = ib_steering->reg_id; 1284 kfree(ib_steering); 1285 } 1286 1287 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw, 1288 prot, reg_id); 1289 if (err) 1290 return err; 1291 1292 mutex_lock(&mqp->mutex); 1293 ge = find_gid_entry(mqp, gid->raw); 1294 if (ge) { 1295 spin_lock(&mdev->iboe.lock); 1296 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL; 1297 if (ndev) 1298 dev_hold(ndev); 1299 spin_unlock(&mdev->iboe.lock); 1300 if (ndev) 1301 dev_put(ndev); 1302 list_del(&ge->list); 1303 kfree(ge); 1304 } else 1305 pr_warn("could not find mgid entry\n"); 1306 1307 mutex_unlock(&mqp->mutex); 1308 1309 return 0; 1310 } 1311 1312 static int init_node_data(struct mlx4_ib_dev *dev) 1313 { 1314 struct ib_smp *in_mad = NULL; 1315 struct ib_smp *out_mad = NULL; 1316 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; 1317 int err = -ENOMEM; 1318 1319 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 1320 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 1321 if (!in_mad || !out_mad) 1322 goto out; 1323 1324 init_query_mad(in_mad); 1325 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC; 1326 if (mlx4_is_master(dev->dev)) 1327 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; 1328 1329 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad); 1330 if (err) 1331 goto out; 1332 1333 memcpy(dev->ib_dev.node_desc, out_mad->data, 64); 1334 1335 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO; 1336 1337 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad); 1338 if (err) 1339 goto out; 1340 1341 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32)); 1342 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8); 1343 1344 out: 1345 kfree(in_mad); 1346 kfree(out_mad); 1347 return err; 1348 } 1349 1350 static ssize_t show_hca(struct device *device, struct device_attribute *attr, 1351 char *buf) 1352 { 1353 struct mlx4_ib_dev *dev = 1354 container_of(device, struct mlx4_ib_dev, ib_dev.dev); 1355 return sprintf(buf, "MT%d\n", dev->dev->pdev->device); 1356 } 1357 1358 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr, 1359 char *buf) 1360 { 1361 struct mlx4_ib_dev *dev = 1362 container_of(device, struct mlx4_ib_dev, ib_dev.dev); 1363 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32), 1364 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff, 1365 (int) dev->dev->caps.fw_ver & 0xffff); 1366 } 1367 1368 static ssize_t show_rev(struct device *device, struct device_attribute *attr, 1369 char *buf) 1370 { 1371 struct mlx4_ib_dev *dev = 1372 container_of(device, struct mlx4_ib_dev, ib_dev.dev); 1373 return sprintf(buf, "%x\n", dev->dev->rev_id); 1374 } 1375 1376 static ssize_t show_board(struct device *device, struct device_attribute *attr, 1377 char *buf) 1378 { 1379 struct mlx4_ib_dev *dev = 1380 container_of(device, struct mlx4_ib_dev, ib_dev.dev); 1381 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN, 1382 dev->dev->board_id); 1383 } 1384 1385 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL); 1386 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL); 1387 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL); 1388 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL); 1389 1390 static struct device_attribute *mlx4_class_attributes[] = { 1391 &dev_attr_hw_rev, 1392 &dev_attr_fw_ver, 1393 &dev_attr_hca_type, 1394 &dev_attr_board_id 1395 }; 1396 1397 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, 1398 struct net_device *dev) 1399 { 1400 memcpy(eui, dev->dev_addr, 3); 1401 memcpy(eui + 5, dev->dev_addr + 3, 3); 1402 if (vlan_id < 0x1000) { 1403 eui[3] = vlan_id >> 8; 1404 eui[4] = vlan_id & 0xff; 1405 } else { 1406 eui[3] = 0xff; 1407 eui[4] = 0xfe; 1408 } 1409 eui[0] ^= 2; 1410 } 1411 1412 static void update_gids_task(struct work_struct *work) 1413 { 1414 struct update_gid_work *gw = container_of(work, struct update_gid_work, work); 1415 struct mlx4_cmd_mailbox *mailbox; 1416 union ib_gid *gids; 1417 int err; 1418 struct mlx4_dev *dev = gw->dev->dev; 1419 1420 mailbox = mlx4_alloc_cmd_mailbox(dev); 1421 if (IS_ERR(mailbox)) { 1422 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox)); 1423 return; 1424 } 1425 1426 gids = mailbox->buf; 1427 memcpy(gids, gw->gids, sizeof gw->gids); 1428 1429 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port, 1430 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B, 1431 MLX4_CMD_WRAPPED); 1432 if (err) 1433 pr_warn("set port command failed\n"); 1434 else 1435 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE); 1436 1437 mlx4_free_cmd_mailbox(dev, mailbox); 1438 kfree(gw); 1439 } 1440 1441 static void reset_gids_task(struct work_struct *work) 1442 { 1443 struct update_gid_work *gw = 1444 container_of(work, struct update_gid_work, work); 1445 struct mlx4_cmd_mailbox *mailbox; 1446 union ib_gid *gids; 1447 int err; 1448 struct mlx4_dev *dev = gw->dev->dev; 1449 1450 mailbox = mlx4_alloc_cmd_mailbox(dev); 1451 if (IS_ERR(mailbox)) { 1452 pr_warn("reset gid table failed\n"); 1453 goto free; 1454 } 1455 1456 gids = mailbox->buf; 1457 memcpy(gids, gw->gids, sizeof(gw->gids)); 1458 1459 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) == 1460 IB_LINK_LAYER_ETHERNET) { 1461 err = mlx4_cmd(dev, mailbox->dma, 1462 MLX4_SET_PORT_GID_TABLE << 8 | gw->port, 1463 1, MLX4_CMD_SET_PORT, 1464 MLX4_CMD_TIME_CLASS_B, 1465 MLX4_CMD_WRAPPED); 1466 if (err) 1467 pr_warn(KERN_WARNING 1468 "set port %d command failed\n", gw->port); 1469 } 1470 1471 mlx4_free_cmd_mailbox(dev, mailbox); 1472 free: 1473 kfree(gw); 1474 } 1475 1476 static int update_gid_table(struct mlx4_ib_dev *dev, int port, 1477 union ib_gid *gid, int clear, 1478 int default_gid) 1479 { 1480 struct update_gid_work *work; 1481 int i; 1482 int need_update = 0; 1483 int free = -1; 1484 int found = -1; 1485 int max_gids; 1486 1487 if (default_gid) { 1488 free = 0; 1489 } else { 1490 max_gids = dev->dev->caps.gid_table_len[port]; 1491 for (i = 1; i < max_gids; ++i) { 1492 if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid, 1493 sizeof(*gid))) 1494 found = i; 1495 1496 if (clear) { 1497 if (found >= 0) { 1498 need_update = 1; 1499 dev->iboe.gid_table[port - 1][found] = 1500 zgid; 1501 break; 1502 } 1503 } else { 1504 if (found >= 0) 1505 break; 1506 1507 if (free < 0 && 1508 !memcmp(&dev->iboe.gid_table[port - 1][i], 1509 &zgid, sizeof(*gid))) 1510 free = i; 1511 } 1512 } 1513 } 1514 1515 if (found == -1 && !clear && free >= 0) { 1516 dev->iboe.gid_table[port - 1][free] = *gid; 1517 need_update = 1; 1518 } 1519 1520 if (!need_update) 1521 return 0; 1522 1523 work = kzalloc(sizeof(*work), GFP_ATOMIC); 1524 if (!work) 1525 return -ENOMEM; 1526 1527 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids)); 1528 INIT_WORK(&work->work, update_gids_task); 1529 work->port = port; 1530 work->dev = dev; 1531 queue_work(wq, &work->work); 1532 1533 return 0; 1534 } 1535 1536 static void mlx4_make_default_gid(struct net_device *dev, union ib_gid *gid) 1537 { 1538 gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL); 1539 mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev); 1540 } 1541 1542 1543 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port) 1544 { 1545 struct update_gid_work *work; 1546 1547 work = kzalloc(sizeof(*work), GFP_ATOMIC); 1548 if (!work) 1549 return -ENOMEM; 1550 1551 memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids)); 1552 memset(work->gids, 0, sizeof(work->gids)); 1553 INIT_WORK(&work->work, reset_gids_task); 1554 work->dev = dev; 1555 work->port = port; 1556 queue_work(wq, &work->work); 1557 return 0; 1558 } 1559 1560 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev, 1561 struct mlx4_ib_dev *ibdev, union ib_gid *gid) 1562 { 1563 struct mlx4_ib_iboe *iboe; 1564 int port = 0; 1565 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ? 1566 rdma_vlan_dev_real_dev(event_netdev) : 1567 event_netdev; 1568 union ib_gid default_gid; 1569 1570 mlx4_make_default_gid(real_dev, &default_gid); 1571 1572 if (!memcmp(gid, &default_gid, sizeof(*gid))) 1573 return 0; 1574 1575 if (event != NETDEV_DOWN && event != NETDEV_UP) 1576 return 0; 1577 1578 if ((real_dev != event_netdev) && 1579 (event == NETDEV_DOWN) && 1580 rdma_link_local_addr((struct in6_addr *)gid)) 1581 return 0; 1582 1583 iboe = &ibdev->iboe; 1584 spin_lock(&iboe->lock); 1585 1586 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) 1587 if ((netif_is_bond_master(real_dev) && 1588 (real_dev == iboe->masters[port - 1])) || 1589 (!netif_is_bond_master(real_dev) && 1590 (real_dev == iboe->netdevs[port - 1]))) 1591 update_gid_table(ibdev, port, gid, 1592 event == NETDEV_DOWN, 0); 1593 1594 spin_unlock(&iboe->lock); 1595 return 0; 1596 1597 } 1598 1599 static u8 mlx4_ib_get_dev_port(struct net_device *dev, 1600 struct mlx4_ib_dev *ibdev) 1601 { 1602 u8 port = 0; 1603 struct mlx4_ib_iboe *iboe; 1604 struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ? 1605 rdma_vlan_dev_real_dev(dev) : dev; 1606 1607 iboe = &ibdev->iboe; 1608 1609 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) 1610 if ((netif_is_bond_master(real_dev) && 1611 (real_dev == iboe->masters[port - 1])) || 1612 (!netif_is_bond_master(real_dev) && 1613 (real_dev == iboe->netdevs[port - 1]))) 1614 break; 1615 1616 if ((port == 0) || (port > ibdev->dev->caps.num_ports)) 1617 return 0; 1618 else 1619 return port; 1620 } 1621 1622 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event, 1623 void *ptr) 1624 { 1625 struct mlx4_ib_dev *ibdev; 1626 struct in_ifaddr *ifa = ptr; 1627 union ib_gid gid; 1628 struct net_device *event_netdev = ifa->ifa_dev->dev; 1629 1630 ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid); 1631 1632 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet); 1633 1634 mlx4_ib_addr_event(event, event_netdev, ibdev, &gid); 1635 return NOTIFY_DONE; 1636 } 1637 1638 #if IS_ENABLED(CONFIG_IPV6) 1639 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event, 1640 void *ptr) 1641 { 1642 struct mlx4_ib_dev *ibdev; 1643 struct inet6_ifaddr *ifa = ptr; 1644 union ib_gid *gid = (union ib_gid *)&ifa->addr; 1645 struct net_device *event_netdev = ifa->idev->dev; 1646 1647 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6); 1648 1649 mlx4_ib_addr_event(event, event_netdev, ibdev, gid); 1650 return NOTIFY_DONE; 1651 } 1652 #endif 1653 1654 #define MLX4_IB_INVALID_MAC ((u64)-1) 1655 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev, 1656 struct net_device *dev, 1657 int port) 1658 { 1659 u64 new_smac = 0; 1660 u64 release_mac = MLX4_IB_INVALID_MAC; 1661 struct mlx4_ib_qp *qp; 1662 1663 read_lock(&dev_base_lock); 1664 new_smac = mlx4_mac_to_u64(dev->dev_addr); 1665 read_unlock(&dev_base_lock); 1666 1667 mutex_lock(&ibdev->qp1_proxy_lock[port - 1]); 1668 qp = ibdev->qp1_proxy[port - 1]; 1669 if (qp) { 1670 int new_smac_index; 1671 u64 old_smac = qp->pri.smac; 1672 struct mlx4_update_qp_params update_params; 1673 1674 if (new_smac == old_smac) 1675 goto unlock; 1676 1677 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac); 1678 1679 if (new_smac_index < 0) 1680 goto unlock; 1681 1682 update_params.smac_index = new_smac_index; 1683 if (mlx4_update_qp(ibdev->dev, &qp->mqp, MLX4_UPDATE_QP_SMAC, 1684 &update_params)) { 1685 release_mac = new_smac; 1686 goto unlock; 1687 } 1688 1689 qp->pri.smac = new_smac; 1690 qp->pri.smac_index = new_smac_index; 1691 1692 release_mac = old_smac; 1693 } 1694 1695 unlock: 1696 mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]); 1697 if (release_mac != MLX4_IB_INVALID_MAC) 1698 mlx4_unregister_mac(ibdev->dev, port, release_mac); 1699 } 1700 1701 static void mlx4_ib_get_dev_addr(struct net_device *dev, 1702 struct mlx4_ib_dev *ibdev, u8 port) 1703 { 1704 struct in_device *in_dev; 1705 #if IS_ENABLED(CONFIG_IPV6) 1706 struct inet6_dev *in6_dev; 1707 union ib_gid *pgid; 1708 struct inet6_ifaddr *ifp; 1709 #endif 1710 union ib_gid gid; 1711 1712 1713 if ((port == 0) || (port > ibdev->dev->caps.num_ports)) 1714 return; 1715 1716 /* IPv4 gids */ 1717 in_dev = in_dev_get(dev); 1718 if (in_dev) { 1719 for_ifa(in_dev) { 1720 /*ifa->ifa_address;*/ 1721 ipv6_addr_set_v4mapped(ifa->ifa_address, 1722 (struct in6_addr *)&gid); 1723 update_gid_table(ibdev, port, &gid, 0, 0); 1724 } 1725 endfor_ifa(in_dev); 1726 in_dev_put(in_dev); 1727 } 1728 #if IS_ENABLED(CONFIG_IPV6) 1729 /* IPv6 gids */ 1730 in6_dev = in6_dev_get(dev); 1731 if (in6_dev) { 1732 read_lock_bh(&in6_dev->lock); 1733 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) { 1734 pgid = (union ib_gid *)&ifp->addr; 1735 update_gid_table(ibdev, port, pgid, 0, 0); 1736 } 1737 read_unlock_bh(&in6_dev->lock); 1738 in6_dev_put(in6_dev); 1739 } 1740 #endif 1741 } 1742 1743 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev, 1744 struct net_device *dev, u8 port) 1745 { 1746 union ib_gid gid; 1747 mlx4_make_default_gid(dev, &gid); 1748 update_gid_table(ibdev, port, &gid, 0, 1); 1749 } 1750 1751 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev) 1752 { 1753 struct net_device *dev; 1754 struct mlx4_ib_iboe *iboe = &ibdev->iboe; 1755 int i; 1756 1757 for (i = 1; i <= ibdev->num_ports; ++i) 1758 if (reset_gid_table(ibdev, i)) 1759 return -1; 1760 1761 read_lock(&dev_base_lock); 1762 spin_lock(&iboe->lock); 1763 1764 for_each_netdev(&init_net, dev) { 1765 u8 port = mlx4_ib_get_dev_port(dev, ibdev); 1766 if (port) 1767 mlx4_ib_get_dev_addr(dev, ibdev, port); 1768 } 1769 1770 spin_unlock(&iboe->lock); 1771 read_unlock(&dev_base_lock); 1772 1773 return 0; 1774 } 1775 1776 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev, 1777 struct net_device *dev, 1778 unsigned long event) 1779 1780 { 1781 struct mlx4_ib_iboe *iboe; 1782 int update_qps_port = -1; 1783 int port; 1784 1785 iboe = &ibdev->iboe; 1786 1787 spin_lock(&iboe->lock); 1788 mlx4_foreach_ib_transport_port(port, ibdev->dev) { 1789 enum ib_port_state port_state = IB_PORT_NOP; 1790 struct net_device *old_master = iboe->masters[port - 1]; 1791 struct net_device *curr_netdev; 1792 struct net_device *curr_master; 1793 1794 iboe->netdevs[port - 1] = 1795 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port); 1796 if (iboe->netdevs[port - 1]) 1797 mlx4_ib_set_default_gid(ibdev, 1798 iboe->netdevs[port - 1], port); 1799 curr_netdev = iboe->netdevs[port - 1]; 1800 1801 if (iboe->netdevs[port - 1] && 1802 netif_is_bond_slave(iboe->netdevs[port - 1])) { 1803 iboe->masters[port - 1] = netdev_master_upper_dev_get( 1804 iboe->netdevs[port - 1]); 1805 } else { 1806 iboe->masters[port - 1] = NULL; 1807 } 1808 curr_master = iboe->masters[port - 1]; 1809 1810 if (dev == iboe->netdevs[port - 1] && 1811 (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER || 1812 event == NETDEV_UP || event == NETDEV_CHANGE)) 1813 update_qps_port = port; 1814 1815 if (curr_netdev) { 1816 port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ? 1817 IB_PORT_ACTIVE : IB_PORT_DOWN; 1818 mlx4_ib_set_default_gid(ibdev, curr_netdev, port); 1819 } else { 1820 reset_gid_table(ibdev, port); 1821 } 1822 /* if using bonding/team and a slave port is down, we don't the bond IP 1823 * based gids in the table since flows that select port by gid may get 1824 * the down port. 1825 */ 1826 if (curr_master && (port_state == IB_PORT_DOWN)) { 1827 reset_gid_table(ibdev, port); 1828 mlx4_ib_set_default_gid(ibdev, curr_netdev, port); 1829 } 1830 /* if bonding is used it is possible that we add it to masters 1831 * only after IP address is assigned to the net bonding 1832 * interface. 1833 */ 1834 if (curr_master && (old_master != curr_master)) { 1835 reset_gid_table(ibdev, port); 1836 mlx4_ib_set_default_gid(ibdev, curr_netdev, port); 1837 mlx4_ib_get_dev_addr(curr_master, ibdev, port); 1838 } 1839 1840 if (!curr_master && (old_master != curr_master)) { 1841 reset_gid_table(ibdev, port); 1842 mlx4_ib_set_default_gid(ibdev, curr_netdev, port); 1843 mlx4_ib_get_dev_addr(curr_netdev, ibdev, port); 1844 } 1845 } 1846 1847 spin_unlock(&iboe->lock); 1848 1849 if (update_qps_port > 0) 1850 mlx4_ib_update_qps(ibdev, dev, update_qps_port); 1851 } 1852 1853 static int mlx4_ib_netdev_event(struct notifier_block *this, 1854 unsigned long event, void *ptr) 1855 { 1856 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1857 struct mlx4_ib_dev *ibdev; 1858 1859 if (!net_eq(dev_net(dev), &init_net)) 1860 return NOTIFY_DONE; 1861 1862 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb); 1863 mlx4_ib_scan_netdevs(ibdev, dev, event); 1864 1865 return NOTIFY_DONE; 1866 } 1867 1868 static void init_pkeys(struct mlx4_ib_dev *ibdev) 1869 { 1870 int port; 1871 int slave; 1872 int i; 1873 1874 if (mlx4_is_master(ibdev->dev)) { 1875 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) { 1876 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) { 1877 for (i = 0; 1878 i < ibdev->dev->phys_caps.pkey_phys_table_len[port]; 1879 ++i) { 1880 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] = 1881 /* master has the identity virt2phys pkey mapping */ 1882 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i : 1883 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1; 1884 mlx4_sync_pkey_table(ibdev->dev, slave, port, i, 1885 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]); 1886 } 1887 } 1888 } 1889 /* initialize pkey cache */ 1890 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) { 1891 for (i = 0; 1892 i < ibdev->dev->phys_caps.pkey_phys_table_len[port]; 1893 ++i) 1894 ibdev->pkeys.phys_pkey_cache[port-1][i] = 1895 (i) ? 0 : 0xFFFF; 1896 } 1897 } 1898 } 1899 1900 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev) 1901 { 1902 char name[80]; 1903 int eq_per_port = 0; 1904 int added_eqs = 0; 1905 int total_eqs = 0; 1906 int i, j, eq; 1907 1908 /* Legacy mode or comp_pool is not large enough */ 1909 if (dev->caps.comp_pool == 0 || 1910 dev->caps.num_ports > dev->caps.comp_pool) 1911 return; 1912 1913 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/ 1914 dev->caps.num_ports); 1915 1916 /* Init eq table */ 1917 added_eqs = 0; 1918 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) 1919 added_eqs += eq_per_port; 1920 1921 total_eqs = dev->caps.num_comp_vectors + added_eqs; 1922 1923 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL); 1924 if (!ibdev->eq_table) 1925 return; 1926 1927 ibdev->eq_added = added_eqs; 1928 1929 eq = 0; 1930 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) { 1931 for (j = 0; j < eq_per_port; j++) { 1932 snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s", 1933 i, j, dev->pdev->bus->name); 1934 /* Set IRQ for specific name (per ring) */ 1935 if (mlx4_assign_eq(dev, name, NULL, 1936 &ibdev->eq_table[eq])) { 1937 /* Use legacy (same as mlx4_en driver) */ 1938 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq); 1939 ibdev->eq_table[eq] = 1940 (eq % dev->caps.num_comp_vectors); 1941 } 1942 eq++; 1943 } 1944 } 1945 1946 /* Fill the reset of the vector with legacy EQ */ 1947 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++) 1948 ibdev->eq_table[eq++] = i; 1949 1950 /* Advertise the new number of EQs to clients */ 1951 ibdev->ib_dev.num_comp_vectors = total_eqs; 1952 } 1953 1954 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev) 1955 { 1956 int i; 1957 1958 /* no additional eqs were added */ 1959 if (!ibdev->eq_table) 1960 return; 1961 1962 /* Reset the advertised EQ number */ 1963 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors; 1964 1965 /* Free only the added eqs */ 1966 for (i = 0; i < ibdev->eq_added; i++) { 1967 /* Don't free legacy eqs if used */ 1968 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors) 1969 continue; 1970 mlx4_release_eq(dev, ibdev->eq_table[i]); 1971 } 1972 1973 kfree(ibdev->eq_table); 1974 } 1975 1976 static void *mlx4_ib_add(struct mlx4_dev *dev) 1977 { 1978 struct mlx4_ib_dev *ibdev; 1979 int num_ports = 0; 1980 int i, j; 1981 int err; 1982 struct mlx4_ib_iboe *iboe; 1983 int ib_num_ports = 0; 1984 1985 pr_info_once("%s", mlx4_ib_version); 1986 1987 num_ports = 0; 1988 mlx4_foreach_ib_transport_port(i, dev) 1989 num_ports++; 1990 1991 /* No point in registering a device with no ports... */ 1992 if (num_ports == 0) 1993 return NULL; 1994 1995 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev); 1996 if (!ibdev) { 1997 dev_err(&dev->pdev->dev, "Device struct alloc failed\n"); 1998 return NULL; 1999 } 2000 2001 iboe = &ibdev->iboe; 2002 2003 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn)) 2004 goto err_dealloc; 2005 2006 if (mlx4_uar_alloc(dev, &ibdev->priv_uar)) 2007 goto err_pd; 2008 2009 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT, 2010 PAGE_SIZE); 2011 if (!ibdev->uar_map) 2012 goto err_uar; 2013 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock); 2014 2015 ibdev->dev = dev; 2016 2017 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX); 2018 ibdev->ib_dev.owner = THIS_MODULE; 2019 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA; 2020 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey; 2021 ibdev->num_ports = num_ports; 2022 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports; 2023 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors; 2024 ibdev->ib_dev.dma_device = &dev->pdev->dev; 2025 2026 if (dev->caps.userspace_caps) 2027 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION; 2028 else 2029 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION; 2030 2031 ibdev->ib_dev.uverbs_cmd_mask = 2032 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | 2033 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) | 2034 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) | 2035 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | 2036 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | 2037 (1ull << IB_USER_VERBS_CMD_REG_MR) | 2038 (1ull << IB_USER_VERBS_CMD_REREG_MR) | 2039 (1ull << IB_USER_VERBS_CMD_DEREG_MR) | 2040 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | 2041 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | 2042 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) | 2043 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) | 2044 (1ull << IB_USER_VERBS_CMD_CREATE_QP) | 2045 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | 2046 (1ull << IB_USER_VERBS_CMD_QUERY_QP) | 2047 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | 2048 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) | 2049 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) | 2050 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) | 2051 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) | 2052 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) | 2053 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) | 2054 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) | 2055 (1ull << IB_USER_VERBS_CMD_OPEN_QP); 2056 2057 ibdev->ib_dev.query_device = mlx4_ib_query_device; 2058 ibdev->ib_dev.query_port = mlx4_ib_query_port; 2059 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer; 2060 ibdev->ib_dev.query_gid = mlx4_ib_query_gid; 2061 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey; 2062 ibdev->ib_dev.modify_device = mlx4_ib_modify_device; 2063 ibdev->ib_dev.modify_port = mlx4_ib_modify_port; 2064 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext; 2065 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext; 2066 ibdev->ib_dev.mmap = mlx4_ib_mmap; 2067 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd; 2068 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd; 2069 ibdev->ib_dev.create_ah = mlx4_ib_create_ah; 2070 ibdev->ib_dev.query_ah = mlx4_ib_query_ah; 2071 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah; 2072 ibdev->ib_dev.create_srq = mlx4_ib_create_srq; 2073 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq; 2074 ibdev->ib_dev.query_srq = mlx4_ib_query_srq; 2075 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq; 2076 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv; 2077 ibdev->ib_dev.create_qp = mlx4_ib_create_qp; 2078 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp; 2079 ibdev->ib_dev.query_qp = mlx4_ib_query_qp; 2080 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp; 2081 ibdev->ib_dev.post_send = mlx4_ib_post_send; 2082 ibdev->ib_dev.post_recv = mlx4_ib_post_recv; 2083 ibdev->ib_dev.create_cq = mlx4_ib_create_cq; 2084 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq; 2085 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq; 2086 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq; 2087 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq; 2088 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq; 2089 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr; 2090 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr; 2091 ibdev->ib_dev.rereg_user_mr = mlx4_ib_rereg_user_mr; 2092 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr; 2093 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr; 2094 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list; 2095 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list; 2096 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach; 2097 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach; 2098 ibdev->ib_dev.process_mad = mlx4_ib_process_mad; 2099 2100 if (!mlx4_is_slave(ibdev->dev)) { 2101 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc; 2102 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr; 2103 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr; 2104 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc; 2105 } 2106 2107 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW || 2108 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) { 2109 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw; 2110 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw; 2111 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw; 2112 2113 ibdev->ib_dev.uverbs_cmd_mask |= 2114 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) | 2115 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW); 2116 } 2117 2118 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) { 2119 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd; 2120 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd; 2121 ibdev->ib_dev.uverbs_cmd_mask |= 2122 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) | 2123 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD); 2124 } 2125 2126 if (check_flow_steering_support(dev)) { 2127 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED; 2128 ibdev->ib_dev.create_flow = mlx4_ib_create_flow; 2129 ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow; 2130 2131 ibdev->ib_dev.uverbs_ex_cmd_mask |= 2132 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) | 2133 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW); 2134 } 2135 2136 mlx4_ib_alloc_eqs(dev, ibdev); 2137 2138 spin_lock_init(&iboe->lock); 2139 2140 if (init_node_data(ibdev)) 2141 goto err_map; 2142 2143 for (i = 0; i < ibdev->num_ports; ++i) { 2144 mutex_init(&ibdev->qp1_proxy_lock[i]); 2145 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) == 2146 IB_LINK_LAYER_ETHERNET) { 2147 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]); 2148 if (err) 2149 ibdev->counters[i] = -1; 2150 } else { 2151 ibdev->counters[i] = -1; 2152 } 2153 } 2154 2155 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) 2156 ib_num_ports++; 2157 2158 spin_lock_init(&ibdev->sm_lock); 2159 mutex_init(&ibdev->cap_mask_mutex); 2160 2161 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED && 2162 ib_num_ports) { 2163 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS; 2164 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count, 2165 MLX4_IB_UC_STEER_QPN_ALIGN, 2166 &ibdev->steer_qpn_base); 2167 if (err) 2168 goto err_counter; 2169 2170 ibdev->ib_uc_qpns_bitmap = 2171 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) * 2172 sizeof(long), 2173 GFP_KERNEL); 2174 if (!ibdev->ib_uc_qpns_bitmap) { 2175 dev_err(&dev->pdev->dev, "bit map alloc failed\n"); 2176 goto err_steer_qp_release; 2177 } 2178 2179 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count); 2180 2181 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE( 2182 dev, ibdev->steer_qpn_base, 2183 ibdev->steer_qpn_base + 2184 ibdev->steer_qpn_count - 1); 2185 if (err) 2186 goto err_steer_free_bitmap; 2187 } 2188 2189 if (ib_register_device(&ibdev->ib_dev, NULL)) 2190 goto err_steer_free_bitmap; 2191 2192 if (mlx4_ib_mad_init(ibdev)) 2193 goto err_reg; 2194 2195 if (mlx4_ib_init_sriov(ibdev)) 2196 goto err_mad; 2197 2198 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) { 2199 if (!iboe->nb.notifier_call) { 2200 iboe->nb.notifier_call = mlx4_ib_netdev_event; 2201 err = register_netdevice_notifier(&iboe->nb); 2202 if (err) { 2203 iboe->nb.notifier_call = NULL; 2204 goto err_notif; 2205 } 2206 } 2207 if (!iboe->nb_inet.notifier_call) { 2208 iboe->nb_inet.notifier_call = mlx4_ib_inet_event; 2209 err = register_inetaddr_notifier(&iboe->nb_inet); 2210 if (err) { 2211 iboe->nb_inet.notifier_call = NULL; 2212 goto err_notif; 2213 } 2214 } 2215 #if IS_ENABLED(CONFIG_IPV6) 2216 if (!iboe->nb_inet6.notifier_call) { 2217 iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event; 2218 err = register_inet6addr_notifier(&iboe->nb_inet6); 2219 if (err) { 2220 iboe->nb_inet6.notifier_call = NULL; 2221 goto err_notif; 2222 } 2223 } 2224 #endif 2225 for (i = 1 ; i <= ibdev->num_ports ; ++i) 2226 reset_gid_table(ibdev, i); 2227 rtnl_lock(); 2228 mlx4_ib_scan_netdevs(ibdev, NULL, 0); 2229 rtnl_unlock(); 2230 mlx4_ib_init_gid_table(ibdev); 2231 } 2232 2233 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) { 2234 if (device_create_file(&ibdev->ib_dev.dev, 2235 mlx4_class_attributes[j])) 2236 goto err_notif; 2237 } 2238 2239 ibdev->ib_active = true; 2240 2241 if (mlx4_is_mfunc(ibdev->dev)) 2242 init_pkeys(ibdev); 2243 2244 /* create paravirt contexts for any VFs which are active */ 2245 if (mlx4_is_master(ibdev->dev)) { 2246 for (j = 0; j < MLX4_MFUNC_MAX; j++) { 2247 if (j == mlx4_master_func_num(ibdev->dev)) 2248 continue; 2249 if (mlx4_is_slave_active(ibdev->dev, j)) 2250 do_slave_init(ibdev, j, 1); 2251 } 2252 } 2253 return ibdev; 2254 2255 err_notif: 2256 if (ibdev->iboe.nb.notifier_call) { 2257 if (unregister_netdevice_notifier(&ibdev->iboe.nb)) 2258 pr_warn("failure unregistering notifier\n"); 2259 ibdev->iboe.nb.notifier_call = NULL; 2260 } 2261 if (ibdev->iboe.nb_inet.notifier_call) { 2262 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet)) 2263 pr_warn("failure unregistering notifier\n"); 2264 ibdev->iboe.nb_inet.notifier_call = NULL; 2265 } 2266 #if IS_ENABLED(CONFIG_IPV6) 2267 if (ibdev->iboe.nb_inet6.notifier_call) { 2268 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6)) 2269 pr_warn("failure unregistering notifier\n"); 2270 ibdev->iboe.nb_inet6.notifier_call = NULL; 2271 } 2272 #endif 2273 flush_workqueue(wq); 2274 2275 mlx4_ib_close_sriov(ibdev); 2276 2277 err_mad: 2278 mlx4_ib_mad_cleanup(ibdev); 2279 2280 err_reg: 2281 ib_unregister_device(&ibdev->ib_dev); 2282 2283 err_steer_free_bitmap: 2284 kfree(ibdev->ib_uc_qpns_bitmap); 2285 2286 err_steer_qp_release: 2287 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) 2288 mlx4_qp_release_range(dev, ibdev->steer_qpn_base, 2289 ibdev->steer_qpn_count); 2290 err_counter: 2291 for (; i; --i) 2292 if (ibdev->counters[i - 1] != -1) 2293 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]); 2294 2295 err_map: 2296 iounmap(ibdev->uar_map); 2297 2298 err_uar: 2299 mlx4_uar_free(dev, &ibdev->priv_uar); 2300 2301 err_pd: 2302 mlx4_pd_free(dev, ibdev->priv_pdn); 2303 2304 err_dealloc: 2305 ib_dealloc_device(&ibdev->ib_dev); 2306 2307 return NULL; 2308 } 2309 2310 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn) 2311 { 2312 int offset; 2313 2314 WARN_ON(!dev->ib_uc_qpns_bitmap); 2315 2316 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap, 2317 dev->steer_qpn_count, 2318 get_count_order(count)); 2319 if (offset < 0) 2320 return offset; 2321 2322 *qpn = dev->steer_qpn_base + offset; 2323 return 0; 2324 } 2325 2326 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count) 2327 { 2328 if (!qpn || 2329 dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED) 2330 return; 2331 2332 BUG_ON(qpn < dev->steer_qpn_base); 2333 2334 bitmap_release_region(dev->ib_uc_qpns_bitmap, 2335 qpn - dev->steer_qpn_base, 2336 get_count_order(count)); 2337 } 2338 2339 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp, 2340 int is_attach) 2341 { 2342 int err; 2343 size_t flow_size; 2344 struct ib_flow_attr *flow = NULL; 2345 struct ib_flow_spec_ib *ib_spec; 2346 2347 if (is_attach) { 2348 flow_size = sizeof(struct ib_flow_attr) + 2349 sizeof(struct ib_flow_spec_ib); 2350 flow = kzalloc(flow_size, GFP_KERNEL); 2351 if (!flow) 2352 return -ENOMEM; 2353 flow->port = mqp->port; 2354 flow->num_of_specs = 1; 2355 flow->size = flow_size; 2356 ib_spec = (struct ib_flow_spec_ib *)(flow + 1); 2357 ib_spec->type = IB_FLOW_SPEC_IB; 2358 ib_spec->size = sizeof(struct ib_flow_spec_ib); 2359 /* Add an empty rule for IB L2 */ 2360 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask)); 2361 2362 err = __mlx4_ib_create_flow(&mqp->ibqp, flow, 2363 IB_FLOW_DOMAIN_NIC, 2364 MLX4_FS_REGULAR, 2365 &mqp->reg_id); 2366 } else { 2367 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id); 2368 } 2369 kfree(flow); 2370 return err; 2371 } 2372 2373 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr) 2374 { 2375 struct mlx4_ib_dev *ibdev = ibdev_ptr; 2376 int p; 2377 2378 mlx4_ib_close_sriov(ibdev); 2379 mlx4_ib_mad_cleanup(ibdev); 2380 ib_unregister_device(&ibdev->ib_dev); 2381 if (ibdev->iboe.nb.notifier_call) { 2382 if (unregister_netdevice_notifier(&ibdev->iboe.nb)) 2383 pr_warn("failure unregistering notifier\n"); 2384 ibdev->iboe.nb.notifier_call = NULL; 2385 } 2386 2387 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) { 2388 mlx4_qp_release_range(dev, ibdev->steer_qpn_base, 2389 ibdev->steer_qpn_count); 2390 kfree(ibdev->ib_uc_qpns_bitmap); 2391 } 2392 2393 if (ibdev->iboe.nb_inet.notifier_call) { 2394 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet)) 2395 pr_warn("failure unregistering notifier\n"); 2396 ibdev->iboe.nb_inet.notifier_call = NULL; 2397 } 2398 #if IS_ENABLED(CONFIG_IPV6) 2399 if (ibdev->iboe.nb_inet6.notifier_call) { 2400 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6)) 2401 pr_warn("failure unregistering notifier\n"); 2402 ibdev->iboe.nb_inet6.notifier_call = NULL; 2403 } 2404 #endif 2405 2406 iounmap(ibdev->uar_map); 2407 for (p = 0; p < ibdev->num_ports; ++p) 2408 if (ibdev->counters[p] != -1) 2409 mlx4_counter_free(ibdev->dev, ibdev->counters[p]); 2410 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB) 2411 mlx4_CLOSE_PORT(dev, p); 2412 2413 mlx4_ib_free_eqs(dev, ibdev); 2414 2415 mlx4_uar_free(dev, &ibdev->priv_uar); 2416 mlx4_pd_free(dev, ibdev->priv_pdn); 2417 ib_dealloc_device(&ibdev->ib_dev); 2418 } 2419 2420 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init) 2421 { 2422 struct mlx4_ib_demux_work **dm = NULL; 2423 struct mlx4_dev *dev = ibdev->dev; 2424 int i; 2425 unsigned long flags; 2426 struct mlx4_active_ports actv_ports; 2427 unsigned int ports; 2428 unsigned int first_port; 2429 2430 if (!mlx4_is_master(dev)) 2431 return; 2432 2433 actv_ports = mlx4_get_active_ports(dev, slave); 2434 ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports); 2435 first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports); 2436 2437 dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC); 2438 if (!dm) { 2439 pr_err("failed to allocate memory for tunneling qp update\n"); 2440 goto out; 2441 } 2442 2443 for (i = 0; i < ports; i++) { 2444 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC); 2445 if (!dm[i]) { 2446 pr_err("failed to allocate memory for tunneling qp update work struct\n"); 2447 for (i = 0; i < dev->caps.num_ports; i++) { 2448 if (dm[i]) 2449 kfree(dm[i]); 2450 } 2451 goto out; 2452 } 2453 } 2454 /* initialize or tear down tunnel QPs for the slave */ 2455 for (i = 0; i < ports; i++) { 2456 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work); 2457 dm[i]->port = first_port + i + 1; 2458 dm[i]->slave = slave; 2459 dm[i]->do_init = do_init; 2460 dm[i]->dev = ibdev; 2461 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags); 2462 if (!ibdev->sriov.is_going_down) 2463 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work); 2464 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags); 2465 } 2466 out: 2467 kfree(dm); 2468 return; 2469 } 2470 2471 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr, 2472 enum mlx4_dev_event event, unsigned long param) 2473 { 2474 struct ib_event ibev; 2475 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr); 2476 struct mlx4_eqe *eqe = NULL; 2477 struct ib_event_work *ew; 2478 int p = 0; 2479 2480 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE) 2481 eqe = (struct mlx4_eqe *)param; 2482 else 2483 p = (int) param; 2484 2485 switch (event) { 2486 case MLX4_DEV_EVENT_PORT_UP: 2487 if (p > ibdev->num_ports) 2488 return; 2489 if (mlx4_is_master(dev) && 2490 rdma_port_get_link_layer(&ibdev->ib_dev, p) == 2491 IB_LINK_LAYER_INFINIBAND) { 2492 mlx4_ib_invalidate_all_guid_record(ibdev, p); 2493 } 2494 ibev.event = IB_EVENT_PORT_ACTIVE; 2495 break; 2496 2497 case MLX4_DEV_EVENT_PORT_DOWN: 2498 if (p > ibdev->num_ports) 2499 return; 2500 ibev.event = IB_EVENT_PORT_ERR; 2501 break; 2502 2503 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR: 2504 ibdev->ib_active = false; 2505 ibev.event = IB_EVENT_DEVICE_FATAL; 2506 break; 2507 2508 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE: 2509 ew = kmalloc(sizeof *ew, GFP_ATOMIC); 2510 if (!ew) { 2511 pr_err("failed to allocate memory for events work\n"); 2512 break; 2513 } 2514 2515 INIT_WORK(&ew->work, handle_port_mgmt_change_event); 2516 memcpy(&ew->ib_eqe, eqe, sizeof *eqe); 2517 ew->ib_dev = ibdev; 2518 /* need to queue only for port owner, which uses GEN_EQE */ 2519 if (mlx4_is_master(dev)) 2520 queue_work(wq, &ew->work); 2521 else 2522 handle_port_mgmt_change_event(&ew->work); 2523 return; 2524 2525 case MLX4_DEV_EVENT_SLAVE_INIT: 2526 /* here, p is the slave id */ 2527 do_slave_init(ibdev, p, 1); 2528 return; 2529 2530 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN: 2531 /* here, p is the slave id */ 2532 do_slave_init(ibdev, p, 0); 2533 return; 2534 2535 default: 2536 return; 2537 } 2538 2539 ibev.device = ibdev_ptr; 2540 ibev.element.port_num = (u8) p; 2541 2542 ib_dispatch_event(&ibev); 2543 } 2544 2545 static struct mlx4_interface mlx4_ib_interface = { 2546 .add = mlx4_ib_add, 2547 .remove = mlx4_ib_remove, 2548 .event = mlx4_ib_event, 2549 .protocol = MLX4_PROT_IB_IPV6 2550 }; 2551 2552 static int __init mlx4_ib_init(void) 2553 { 2554 int err; 2555 2556 wq = create_singlethread_workqueue("mlx4_ib"); 2557 if (!wq) 2558 return -ENOMEM; 2559 2560 err = mlx4_ib_mcg_init(); 2561 if (err) 2562 goto clean_wq; 2563 2564 err = mlx4_register_interface(&mlx4_ib_interface); 2565 if (err) 2566 goto clean_mcg; 2567 2568 return 0; 2569 2570 clean_mcg: 2571 mlx4_ib_mcg_destroy(); 2572 2573 clean_wq: 2574 destroy_workqueue(wq); 2575 return err; 2576 } 2577 2578 static void __exit mlx4_ib_cleanup(void) 2579 { 2580 mlx4_unregister_interface(&mlx4_ib_interface); 2581 mlx4_ib_mcg_destroy(); 2582 destroy_workqueue(wq); 2583 } 2584 2585 module_init(mlx4_ib_init); 2586 module_exit(mlx4_ib_cleanup); 2587