main.c revision 325611
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#define LINUXKPI_PARAM_PREFIX mlx4_ 35 36#include <linux/module.h> 37 38#ifdef __linux__ 39#include <linux/proc_fs.h> 40#endif 41 42#include <linux/slab.h> 43#include <linux/errno.h> 44#include <linux/netdevice.h> 45#include <linux/inetdevice.h> 46#include <linux/if_vlan.h> 47#include <linux/bitops.h> 48#include <linux/if_ether.h> 49#include <linux/fs.h> 50 51#include <rdma/ib_smi.h> 52#include <rdma/ib_user_verbs.h> 53#include <rdma/ib_addr.h> 54 55#include <linux/mlx4/driver.h> 56#include <linux/mlx4/cmd.h> 57#include <linux/sched.h> 58#include "mlx4_ib.h" 59#include "user.h" 60#include "wc.h" 61 62#define DRV_NAME MLX4_IB_DRV_NAME 63#define DRV_VERSION "1.0" 64#define DRV_RELDATE "April 4, 2008" 65 66#define MLX4_IB_DRIVER_PROC_DIR_NAME "driver/mlx4_ib" 67#define MLX4_IB_MRS_PROC_DIR_NAME "mrs" 68 69MODULE_AUTHOR("Roland Dreier"); 70MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver"); 71MODULE_LICENSE("Dual BSD/GPL"); 72MODULE_VERSION(DRV_VERSION); 73 74int mlx4_ib_sm_guid_assign = 1; 75 76#ifdef __linux__ 77struct proc_dir_entry *mlx4_mrs_dir_entry; 78static struct proc_dir_entry *mlx4_ib_driver_dir_entry; 79#endif 80 81module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444); 82MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)"); 83 84static char dev_assign_str[512]; 85//module_param_string(dev_assign_str, dev_assign_str, sizeof(dev_assign_str), 0644); 86MODULE_PARM_DESC(dev_assign_str, "Map all device function numbers to " 87 "IB device numbers following the pattern: " 88 "bb:dd.f-0,bb:dd.f-1,... (all numbers are hexadecimals)." 89 " Max supported devices - 32"); 90 91static const char mlx4_ib_version[] = 92 DRV_NAME ": Mellanox ConnectX InfiniBand driver v" 93 DRV_VERSION " (" DRV_RELDATE ")\n"; 94 95struct update_gid_work { 96 struct work_struct work; 97 union ib_gid gids[128]; 98 struct mlx4_ib_dev *dev; 99 int port; 100}; 101 102struct dev_rec { 103 int bus; 104 int dev; 105 int func; 106 int nr; 107}; 108 109#define MAX_DR 32 110static struct dev_rec dr[MAX_DR]; 111 112static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init); 113 114static struct workqueue_struct *wq; 115 116static void init_query_mad(struct ib_smp *mad) 117{ 118 mad->base_version = 1; 119 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED; 120 mad->class_version = 1; 121 mad->method = IB_MGMT_METHOD_GET; 122} 123 124static union ib_gid zgid; 125 126static int mlx4_ib_query_device(struct ib_device *ibdev, 127 struct ib_device_attr *props) 128{ 129 struct mlx4_ib_dev *dev = to_mdev(ibdev); 130 struct ib_smp *in_mad = NULL; 131 struct ib_smp *out_mad = NULL; 132 int err = -ENOMEM; 133 134 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 135 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 136 if (!in_mad || !out_mad) 137 goto out; 138 139 init_query_mad(in_mad); 140 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO; 141 142 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS, 143 1, NULL, NULL, in_mad, out_mad); 144 if (err) 145 goto out; 146 147 memset(props, 0, sizeof *props); 148 149 props->fw_ver = dev->dev->caps.fw_ver; 150 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT | 151 IB_DEVICE_PORT_ACTIVE_EVENT | 152 IB_DEVICE_SYS_IMAGE_GUID | 153 IB_DEVICE_RC_RNR_NAK_GEN | 154 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK | 155 IB_DEVICE_SHARED_MR; 156 157 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR) 158 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR; 159 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR) 160 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR; 161 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM) 162 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG; 163 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT) 164 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE; 165 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM) 166 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM; 167 if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH) 168 props->device_cap_flags |= IB_DEVICE_UD_TSO; 169 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY) 170 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY; 171 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) && 172 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) && 173 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR)) 174 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS; 175 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) 176 props->device_cap_flags |= IB_DEVICE_XRC; 177 178 props->device_cap_flags |= IB_DEVICE_QPG; 179 if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) { 180 props->device_cap_flags |= IB_DEVICE_UD_RSS; 181 props->max_rss_tbl_sz = dev->dev->caps.max_rss_tbl_sz; 182 } 183 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) & 184 0xffffff; 185 props->vendor_part_id = dev->dev->pdev->device; 186 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32)); 187 memcpy(&props->sys_image_guid, out_mad->data + 4, 8); 188 189 props->max_mr_size = ~0ull; 190 props->page_size_cap = dev->dev->caps.page_size_cap; 191 props->max_qp = dev->dev->quotas.qp; 192 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE; 193 props->max_sge = min(dev->dev->caps.max_sq_sg, 194 dev->dev->caps.max_rq_sg); 195 props->max_cq = dev->dev->quotas.cq; 196 props->max_cqe = dev->dev->caps.max_cqes; 197 props->max_mr = dev->dev->quotas.mpt; 198 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds; 199 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma; 200 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma; 201 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp; 202 props->max_srq = dev->dev->quotas.srq; 203 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1; 204 props->max_srq_sge = dev->dev->caps.max_srq_sge; 205 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES; 206 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay; 207 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ? 208 IB_ATOMIC_HCA : IB_ATOMIC_NONE; 209 props->masked_atomic_cap = props->atomic_cap; 210 props->max_pkeys = dev->dev->caps.pkey_table_len[1]; 211 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms; 212 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm; 213 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach * 214 props->max_mcast_grp; 215 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps; 216 217out: 218 kfree(in_mad); 219 kfree(out_mad); 220 221 return err; 222} 223 224static enum rdma_link_layer 225mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num) 226{ 227 struct mlx4_dev *dev = to_mdev(device)->dev; 228 229 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ? 230 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET; 231} 232 233static int ib_link_query_port(struct ib_device *ibdev, u8 port, 234 struct ib_port_attr *props, int netw_view) 235{ 236 struct ib_smp *in_mad = NULL; 237 struct ib_smp *out_mad = NULL; 238 int ext_active_speed; 239 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; 240 int err = -ENOMEM; 241 242 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 243 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 244 if (!in_mad || !out_mad) 245 goto out; 246 247 init_query_mad(in_mad); 248 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO; 249 in_mad->attr_mod = cpu_to_be32(port); 250 251 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view) 252 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; 253 254 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL, 255 in_mad, out_mad); 256 if (err) 257 goto out; 258 259 260 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16)); 261 props->lmc = out_mad->data[34] & 0x7; 262 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18)); 263 props->sm_sl = out_mad->data[36] & 0xf; 264 props->state = out_mad->data[32] & 0xf; 265 props->phys_state = out_mad->data[33] >> 4; 266 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20)); 267 if (netw_view) 268 props->gid_tbl_len = out_mad->data[50]; 269 else 270 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port]; 271 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz; 272 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port]; 273 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46)); 274 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48)); 275 props->active_width = out_mad->data[31] & 0xf; 276 props->active_speed = out_mad->data[35] >> 4; 277 props->max_mtu = out_mad->data[41] & 0xf; 278 props->active_mtu = out_mad->data[36] >> 4; 279 props->subnet_timeout = out_mad->data[51] & 0x1f; 280 props->max_vl_num = out_mad->data[37] >> 4; 281 props->init_type_reply = out_mad->data[41] >> 4; 282 283 /* Check if extended speeds (EDR/FDR/...) are supported */ 284 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) { 285 ext_active_speed = out_mad->data[62] >> 4; 286 287 switch (ext_active_speed) { 288 case 1: 289 props->active_speed = IB_SPEED_FDR; 290 break; 291 case 2: 292 props->active_speed = IB_SPEED_EDR; 293 break; 294 } 295 } 296 297 /* If reported active speed is QDR, check if is FDR-10 */ 298 if (props->active_speed == IB_SPEED_QDR) { 299 init_query_mad(in_mad); 300 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO; 301 in_mad->attr_mod = cpu_to_be32(port); 302 303 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, 304 NULL, NULL, in_mad, out_mad); 305 if (err) 306 goto out; 307 308 /* Checking LinkSpeedActive for FDR-10 */ 309 if (out_mad->data[15] & 0x1) 310 props->active_speed = IB_SPEED_FDR10; 311 } 312 313 /* Avoid wrong speed value returned by FW if the IB link is down. */ 314 if (props->state == IB_PORT_DOWN) 315 props->active_speed = IB_SPEED_SDR; 316 317out: 318 kfree(in_mad); 319 kfree(out_mad); 320 return err; 321} 322 323static u8 state_to_phys_state(enum ib_port_state state) 324{ 325 return state == IB_PORT_ACTIVE ? 5 : 3; 326} 327 328static int eth_link_query_port(struct ib_device *ibdev, u8 port, 329 struct ib_port_attr *props, int netw_view) 330{ 331 332 struct mlx4_ib_dev *mdev = to_mdev(ibdev); 333 struct mlx4_ib_iboe *iboe = &mdev->iboe; 334 struct net_device *ndev; 335 enum ib_mtu tmp; 336 struct mlx4_cmd_mailbox *mailbox; 337 int err = 0; 338 339 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev); 340 if (IS_ERR(mailbox)) 341 return PTR_ERR(mailbox); 342 343 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0, 344 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B, 345 MLX4_CMD_WRAPPED); 346 if (err) 347 goto out; 348 349 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ? 350 IB_WIDTH_4X : IB_WIDTH_1X; 351 props->active_speed = IB_SPEED_QDR; 352 props->port_cap_flags = IB_PORT_CM_SUP; 353 if (netw_view) 354 props->gid_tbl_len = MLX4_ROCE_MAX_GIDS; 355 else 356 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port]; 357 358 props->max_msg_sz = mdev->dev->caps.max_msg_sz; 359 props->pkey_tbl_len = 1; 360 props->max_mtu = IB_MTU_4096; 361 props->max_vl_num = 2; 362 props->state = IB_PORT_DOWN; 363 props->phys_state = state_to_phys_state(props->state); 364 props->active_mtu = IB_MTU_256; 365 spin_lock(&iboe->lock); 366 ndev = iboe->netdevs[port - 1]; 367 if (!ndev) 368 goto out_unlock; 369 370 tmp = iboe_get_mtu(ndev->if_mtu); 371 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256; 372 373 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ? 374 IB_PORT_ACTIVE : IB_PORT_DOWN; 375 props->phys_state = state_to_phys_state(props->state); 376out_unlock: 377 spin_unlock(&iboe->lock); 378out: 379 mlx4_free_cmd_mailbox(mdev->dev, mailbox); 380 return err; 381} 382 383int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port, 384 struct ib_port_attr *props, int netw_view) 385{ 386 int err; 387 388 memset(props, 0, sizeof *props); 389 390 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ? 391 ib_link_query_port(ibdev, port, props, netw_view) : 392 eth_link_query_port(ibdev, port, props, netw_view); 393 394 return err; 395} 396 397static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port, 398 struct ib_port_attr *props) 399{ 400 /* returns host view */ 401 return __mlx4_ib_query_port(ibdev, port, props, 0); 402} 403 404int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, 405 union ib_gid *gid, int netw_view) 406{ 407 struct ib_smp *in_mad = NULL; 408 struct ib_smp *out_mad = NULL; 409 int err = -ENOMEM; 410 struct mlx4_ib_dev *dev = to_mdev(ibdev); 411 int clear = 0; 412 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; 413 414 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 415 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 416 if (!in_mad || !out_mad) 417 goto out; 418 419 init_query_mad(in_mad); 420 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO; 421 in_mad->attr_mod = cpu_to_be32(port); 422 423 if (mlx4_is_mfunc(dev->dev) && netw_view) 424 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; 425 426 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad); 427 if (err) 428 goto out; 429 430 memcpy(gid->raw, out_mad->data + 8, 8); 431 432 if (mlx4_is_mfunc(dev->dev) && !netw_view) { 433 if (index) { 434 /* For any index > 0, return the null guid */ 435 err = 0; 436 clear = 1; 437 goto out; 438 } 439 } 440 441 init_query_mad(in_mad); 442 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO; 443 in_mad->attr_mod = cpu_to_be32(index / 8); 444 445 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, 446 NULL, NULL, in_mad, out_mad); 447 if (err) 448 goto out; 449 450 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8); 451 452out: 453 if (clear) 454 memset(gid->raw + 8, 0, 8); 455 kfree(in_mad); 456 kfree(out_mad); 457 return err; 458} 459 460static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index, 461 union ib_gid *gid) 462{ 463 struct mlx4_ib_dev *dev = to_mdev(ibdev); 464 465 *gid = dev->iboe.gid_table[port - 1][index]; 466 467 return 0; 468} 469 470static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index, 471 union ib_gid *gid) 472{ 473 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND) 474 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0); 475 else 476 return iboe_query_gid(ibdev, port, index, gid); 477} 478 479int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, 480 u16 *pkey, int netw_view) 481{ 482 struct ib_smp *in_mad = NULL; 483 struct ib_smp *out_mad = NULL; 484 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; 485 int err = -ENOMEM; 486 487 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 488 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 489 if (!in_mad || !out_mad) 490 goto out; 491 492 init_query_mad(in_mad); 493 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE; 494 in_mad->attr_mod = cpu_to_be32(index / 32); 495 496 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view) 497 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; 498 499 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL, 500 in_mad, out_mad); 501 if (err) 502 goto out; 503 504 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]); 505 506out: 507 kfree(in_mad); 508 kfree(out_mad); 509 return err; 510} 511 512static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey) 513{ 514 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0); 515} 516 517static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask, 518 struct ib_device_modify *props) 519{ 520 struct mlx4_cmd_mailbox *mailbox; 521 unsigned long flags; 522 523 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC) 524 return -EOPNOTSUPP; 525 526 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC)) 527 return 0; 528 529 if (mlx4_is_slave(to_mdev(ibdev)->dev)) 530 return -EOPNOTSUPP; 531 532 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags); 533 memcpy(ibdev->node_desc, props->node_desc, 64); 534 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags); 535 536 /* 537 * If possible, pass node desc to FW, so it can generate 538 * a 144 trap. If cmd fails, just ignore. 539 */ 540 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev); 541 if (IS_ERR(mailbox)) 542 return 0; 543 544 memset(mailbox->buf, 0, 256); 545 memcpy(mailbox->buf, props->node_desc, 64); 546 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0, 547 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE); 548 549 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox); 550 551 return 0; 552} 553 554static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols, 555 u32 cap_mask) 556{ 557 struct mlx4_cmd_mailbox *mailbox; 558 int err; 559 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH; 560 561 mailbox = mlx4_alloc_cmd_mailbox(dev->dev); 562 if (IS_ERR(mailbox)) 563 return PTR_ERR(mailbox); 564 565 memset(mailbox->buf, 0, 256); 566 567 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) { 568 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6; 569 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask); 570 } else { 571 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols; 572 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask); 573 } 574 575 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT, 576 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE); 577 578 mlx4_free_cmd_mailbox(dev->dev, mailbox); 579 return err; 580} 581 582static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask, 583 struct ib_port_modify *props) 584{ 585 struct ib_port_attr attr; 586 u32 cap_mask; 587 int err; 588 589 mutex_lock(&to_mdev(ibdev)->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_SET_PORT(to_mdev(ibdev), port, 599 !!(mask & IB_PORT_RESET_QKEY_CNTR), 600 cap_mask); 601 602out: 603 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex); 604 return err; 605} 606 607static 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 if (mlx4_wc_enabled()) { 622 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size; 623 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page; 624 } else { 625 resp_v3.bf_reg_size = 0; 626 resp_v3.bf_regs_per_page = 0; 627 } 628 } else { 629 resp.dev_caps = dev->dev->caps.userspace_caps; 630 resp.qp_tab_size = dev->dev->caps.num_qps; 631 if (mlx4_wc_enabled()) { 632 resp.bf_reg_size = dev->dev->caps.bf_reg_size; 633 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page; 634 } else { 635 resp.bf_reg_size = 0; 636 resp.bf_regs_per_page = 0; 637 } 638 resp.cqe_size = dev->dev->caps.cqe_size; 639 } 640 641 context = kmalloc(sizeof *context, GFP_KERNEL); 642 if (!context) 643 return ERR_PTR(-ENOMEM); 644 645 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar); 646 if (err) { 647 kfree(context); 648 return ERR_PTR(err); 649 } 650 651 INIT_LIST_HEAD(&context->db_page_list); 652 mutex_init(&context->db_page_mutex); 653 654 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) 655 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3)); 656 else 657 err = ib_copy_to_udata(udata, &resp, sizeof(resp)); 658 659 if (err) { 660 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar); 661 kfree(context); 662 return ERR_PTR(-EFAULT); 663 } 664 665 return &context->ibucontext; 666} 667 668static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext) 669{ 670 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext); 671 672 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar); 673 kfree(context); 674 675 return 0; 676} 677#ifdef __linux__ 678static unsigned long mlx4_ib_get_unmapped_area(struct file *file, 679 unsigned long addr, 680 unsigned long len, unsigned long pgoff, 681 unsigned long flags) 682{ 683 struct mm_struct *mm; 684 struct vm_area_struct *vma; 685 unsigned long start_addr; 686 unsigned long page_size_order; 687 unsigned long command; 688 689 mm = current->mm; 690 if (addr) 691 return current->mm->get_unmapped_area(file, addr, len, 692 pgoff, flags); 693 694 /* Last 8 bits hold the command others are data per that command */ 695 command = pgoff & MLX4_IB_MMAP_CMD_MASK; 696 if (command != MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES) 697 return current->mm->get_unmapped_area(file, addr, len, 698 pgoff, flags); 699 700 page_size_order = pgoff >> MLX4_IB_MMAP_CMD_BITS; 701 /* code is based on the huge-pages get_unmapped_area code */ 702 start_addr = mm->free_area_cache; 703 704 if (len <= mm->cached_hole_size) 705 start_addr = TASK_UNMAPPED_BASE; 706 707 708full_search: 709 addr = ALIGN(start_addr, 1 << page_size_order); 710 711 for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { 712 /* At this point: (!vma || addr < vma->vm_end). */ 713 if (TASK_SIZE - len < addr) { 714 /* 715 * Start a new search - just in case we missed 716 * some holes. 717 */ 718 if (start_addr != TASK_UNMAPPED_BASE) { 719 start_addr = TASK_UNMAPPED_BASE; 720 goto full_search; 721 } 722 return -ENOMEM; 723 } 724 725 if (!vma || addr + len <= vma->vm_start) 726 return addr; 727 addr = ALIGN(vma->vm_end, 1 << page_size_order); 728 } 729} 730#endif 731 732static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma) 733{ 734 struct mlx4_ib_dev *dev = to_mdev(context->device); 735 int err; 736 737 /* Last 8 bits hold the command others are data per that command */ 738 unsigned long command = vma->vm_pgoff & MLX4_IB_MMAP_CMD_MASK; 739 740 if (command < MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES) { 741 /* compatability handling for commands 0 & 1*/ 742 if (vma->vm_end - vma->vm_start != PAGE_SIZE) 743 return -EINVAL; 744 } 745 if (command == MLX4_IB_MMAP_UAR_PAGE) { 746 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 747 748 if (io_remap_pfn_range(vma, vma->vm_start, 749 to_mucontext(context)->uar.pfn, 750 PAGE_SIZE, vma->vm_page_prot)) 751 return -EAGAIN; 752 } else if (command == MLX4_IB_MMAP_BLUE_FLAME_PAGE && 753 dev->dev->caps.bf_reg_size != 0) { 754 vma->vm_page_prot = pgprot_wc(vma->vm_page_prot); 755 756 if (io_remap_pfn_range(vma, vma->vm_start, 757 to_mucontext(context)->uar.pfn + 758 dev->dev->caps.num_uars, 759 PAGE_SIZE, vma->vm_page_prot)) 760 return -EAGAIN; 761 } else if (command == MLX4_IB_MMAP_GET_CONTIGUOUS_PAGES) { 762 /* Getting contiguous physical pages */ 763 unsigned long total_size = vma->vm_end - vma->vm_start; 764 unsigned long page_size_order = (vma->vm_pgoff) >> 765 MLX4_IB_MMAP_CMD_BITS; 766 struct ib_cmem *ib_cmem; 767 ib_cmem = ib_cmem_alloc_contiguous_pages(context, total_size, 768 page_size_order); 769 if (IS_ERR(ib_cmem)) { 770 err = PTR_ERR(ib_cmem); 771 return err; 772 } 773 774 err = ib_cmem_map_contiguous_pages_to_vma(ib_cmem, vma); 775 if (err) { 776 ib_cmem_release_contiguous_pages(ib_cmem); 777 return err; 778 } 779 return 0; 780 } else 781 return -EINVAL; 782 783 return 0; 784} 785 786static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev, 787 struct ib_ucontext *context, 788 struct ib_udata *udata) 789{ 790 struct mlx4_ib_pd *pd; 791 int err; 792 793 pd = kmalloc(sizeof *pd, GFP_KERNEL); 794 if (!pd) 795 return ERR_PTR(-ENOMEM); 796 797 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn); 798 if (err) { 799 kfree(pd); 800 return ERR_PTR(err); 801 } 802 803 if (context) 804 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) { 805 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn); 806 kfree(pd); 807 return ERR_PTR(-EFAULT); 808 } 809 810 return &pd->ibpd; 811} 812 813static int mlx4_ib_dealloc_pd(struct ib_pd *pd) 814{ 815 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn); 816 kfree(pd); 817 818 return 0; 819} 820 821static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev, 822 struct ib_ucontext *context, 823 struct ib_udata *udata) 824{ 825 struct mlx4_ib_xrcd *xrcd; 826 int err; 827 828 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)) 829 return ERR_PTR(-ENOSYS); 830 831 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL); 832 if (!xrcd) 833 return ERR_PTR(-ENOMEM); 834 835 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn); 836 if (err) 837 goto err1; 838 839 xrcd->pd = ib_alloc_pd(ibdev); 840 if (IS_ERR(xrcd->pd)) { 841 err = PTR_ERR(xrcd->pd); 842 goto err2; 843 } 844 845 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0); 846 if (IS_ERR(xrcd->cq)) { 847 err = PTR_ERR(xrcd->cq); 848 goto err3; 849 } 850 851 return &xrcd->ibxrcd; 852 853err3: 854 ib_dealloc_pd(xrcd->pd); 855err2: 856 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn); 857err1: 858 kfree(xrcd); 859 return ERR_PTR(err); 860} 861 862static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd) 863{ 864 ib_destroy_cq(to_mxrcd(xrcd)->cq); 865 ib_dealloc_pd(to_mxrcd(xrcd)->pd); 866 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn); 867 kfree(xrcd); 868 869 return 0; 870} 871 872static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid) 873{ 874 struct mlx4_ib_qp *mqp = to_mqp(ibqp); 875 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device); 876 struct mlx4_ib_gid_entry *ge; 877 878 ge = kzalloc(sizeof *ge, GFP_KERNEL); 879 if (!ge) 880 return -ENOMEM; 881 882 ge->gid = *gid; 883 if (mlx4_ib_add_mc(mdev, mqp, gid)) { 884 ge->port = mqp->port; 885 ge->added = 1; 886 } 887 888 mutex_lock(&mqp->mutex); 889 list_add_tail(&ge->list, &mqp->gid_list); 890 mutex_unlock(&mqp->mutex); 891 892 return 0; 893} 894 895int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp, 896 union ib_gid *gid) 897{ 898 u8 mac[6]; 899 struct net_device *ndev; 900 int ret = 0; 901 902 if (!mqp->port) 903 return 0; 904 905 spin_lock(&mdev->iboe.lock); 906 ndev = mdev->iboe.netdevs[mqp->port - 1]; 907 if (ndev) 908 dev_hold(ndev); 909 spin_unlock(&mdev->iboe.lock); 910 911 if (ndev) { 912 rdma_get_mcast_mac((struct in6_addr *)gid, mac); 913 rtnl_lock(); 914 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac, 6, 0); 915 ret = 1; 916 rtnl_unlock(); 917 dev_put(ndev); 918 } 919 920 return ret; 921} 922 923struct mlx4_ib_steering { 924 struct list_head list; 925 u64 reg_id; 926 union ib_gid gid; 927}; 928 929static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid) 930{ 931 int err; 932 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device); 933 struct mlx4_ib_qp *mqp = to_mqp(ibqp); 934 u64 reg_id; 935 struct mlx4_ib_steering *ib_steering = NULL; 936 937 if (mdev->dev->caps.steering_mode == 938 MLX4_STEERING_MODE_DEVICE_MANAGED) { 939 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL); 940 if (!ib_steering) 941 return -ENOMEM; 942 } 943 944 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port, 945 !!(mqp->flags & 946 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK), 947 MLX4_PROT_IB_IPV6, ®_id); 948 if (err) 949 goto err_malloc; 950 951 err = add_gid_entry(ibqp, gid); 952 if (err) 953 goto err_add; 954 955 if (ib_steering) { 956 memcpy(ib_steering->gid.raw, gid->raw, 16); 957 ib_steering->reg_id = reg_id; 958 mutex_lock(&mqp->mutex); 959 list_add(&ib_steering->list, &mqp->steering_rules); 960 mutex_unlock(&mqp->mutex); 961 } 962 return 0; 963 964err_add: 965 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw, 966 MLX4_PROT_IB_IPV6, reg_id); 967err_malloc: 968 kfree(ib_steering); 969 970 return err; 971} 972 973enum { 974 IBV_FLOW_L4_NONE = 0, 975 IBV_FLOW_L4_OTHER = 3, 976 IBV_FLOW_L4_UDP = 5, 977 IBV_FLOW_L4_TCP = 6 978}; 979 980struct mlx4_cm_steering { 981 struct list_head list; 982 u64 reg_id; 983 struct ib_flow_spec spec; 984}; 985 986static int flow_spec_to_net_rule(struct ib_device *dev, struct ib_flow_spec *flow_spec, 987 struct list_head *rule_list_h) 988{ 989 struct mlx4_spec_list *spec_l2, *spec_l3, *spec_l4; 990 u64 mac_msk = cpu_to_be64(MLX4_MAC_MASK << 16); 991 992 spec_l2 = kzalloc(sizeof *spec_l2, GFP_KERNEL); 993 if (!spec_l2) 994 return -ENOMEM; 995 996 switch (flow_spec->type) { 997 case IB_FLOW_ETH: 998 spec_l2->id = MLX4_NET_TRANS_RULE_ID_ETH; 999 memcpy(spec_l2->eth.dst_mac, flow_spec->l2_id.eth.mac, ETH_ALEN); 1000 memcpy(spec_l2->eth.dst_mac_msk, &mac_msk, ETH_ALEN); 1001 spec_l2->eth.ether_type = flow_spec->l2_id.eth.ethertype; 1002 if (flow_spec->l2_id.eth.vlan_present) { 1003 spec_l2->eth.vlan_id = flow_spec->l2_id.eth.vlan; 1004 spec_l2->eth.vlan_id_msk = cpu_to_be16(0x0fff); 1005 } 1006 break; 1007 case IB_FLOW_IB_UC: 1008 spec_l2->id = MLX4_NET_TRANS_RULE_ID_IB; 1009 if(flow_spec->l2_id.ib_uc.qpn) { 1010 spec_l2->ib.l3_qpn = cpu_to_be32(flow_spec->l2_id.ib_uc.qpn); 1011 spec_l2->ib.qpn_msk = cpu_to_be32(0xffffff); 1012 } 1013 break; 1014 case IB_FLOW_IB_MC_IPV4: 1015 case IB_FLOW_IB_MC_IPV6: 1016 spec_l2->id = MLX4_NET_TRANS_RULE_ID_IB; 1017 memcpy(spec_l2->ib.dst_gid, flow_spec->l2_id.ib_mc.mgid, 16); 1018 memset(spec_l2->ib.dst_gid_msk, 0xff, 16); 1019 break; 1020 } 1021 1022 1023 list_add_tail(&spec_l2->list, rule_list_h); 1024 1025 if (flow_spec->l2_id.eth.ethertype == cpu_to_be16(ETH_P_IP) || 1026 flow_spec->type != IB_FLOW_ETH) { 1027 spec_l3 = kzalloc(sizeof *spec_l3, GFP_KERNEL); 1028 if (!spec_l3) 1029 return -ENOMEM; 1030 1031 spec_l3->id = MLX4_NET_TRANS_RULE_ID_IPV4; 1032 spec_l3->ipv4.src_ip = flow_spec->src_ip; 1033 if (flow_spec->type != IB_FLOW_IB_MC_IPV4 && 1034 flow_spec->type != IB_FLOW_IB_MC_IPV6) 1035 spec_l3->ipv4.dst_ip = flow_spec->dst_ip; 1036 1037 if (spec_l3->ipv4.src_ip) 1038 spec_l3->ipv4.src_ip_msk = MLX4_BE_WORD_MASK; 1039 if (spec_l3->ipv4.dst_ip) 1040 spec_l3->ipv4.dst_ip_msk = MLX4_BE_WORD_MASK; 1041 1042 list_add_tail(&spec_l3->list, rule_list_h); 1043 } 1044 1045 if (flow_spec->l4_protocol) { 1046 spec_l4 = kzalloc(sizeof(*spec_l4), GFP_KERNEL); 1047 if (!spec_l4) 1048 return -ENOMEM; 1049 1050 spec_l4->tcp_udp.src_port = flow_spec->src_port; 1051 spec_l4->tcp_udp.dst_port = flow_spec->dst_port; 1052 if (spec_l4->tcp_udp.src_port) 1053 spec_l4->tcp_udp.src_port_msk = 1054 MLX4_BE_SHORT_MASK; 1055 if (spec_l4->tcp_udp.dst_port) 1056 spec_l4->tcp_udp.dst_port_msk = 1057 MLX4_BE_SHORT_MASK; 1058 1059 switch (flow_spec->l4_protocol) { 1060 case IBV_FLOW_L4_UDP: 1061 spec_l4->id = MLX4_NET_TRANS_RULE_ID_UDP; 1062 break; 1063 case IBV_FLOW_L4_TCP: 1064 spec_l4->id = MLX4_NET_TRANS_RULE_ID_TCP; 1065 break; 1066 default: 1067 dev_err(dev->dma_device, 1068 "Unsupported l4 protocol.\n"); 1069 kfree(spec_l4); 1070 return -EPROTONOSUPPORT; 1071 } 1072 list_add_tail(&spec_l4->list, rule_list_h); 1073 } 1074 return 0; 1075} 1076 1077static int __mlx4_ib_flow_attach(struct mlx4_ib_dev *mdev, 1078 struct mlx4_ib_qp *mqp, 1079 struct ib_flow_spec *flow_spec, 1080 int priority, int lock_qp) 1081{ 1082 u64 reg_id = 0; 1083 int err = 0; 1084 struct mlx4_cm_steering *cm_flow; 1085 struct mlx4_spec_list *spec, *tmp_spec; 1086 1087 struct mlx4_net_trans_rule rule = 1088 { .queue_mode = MLX4_NET_TRANS_Q_FIFO, 1089 .exclusive = 0, 1090 }; 1091 1092 rule.promisc_mode = flow_spec->rule_type; 1093 rule.port = mqp->port; 1094 rule.qpn = mqp->mqp.qpn; 1095 INIT_LIST_HEAD(&rule.list); 1096 1097 cm_flow = kmalloc(sizeof(*cm_flow), GFP_KERNEL); 1098 if (!cm_flow) 1099 return -ENOMEM; 1100 1101 if (rule.promisc_mode == MLX4_FS_REGULAR) { 1102 rule.allow_loopback = !flow_spec->block_mc_loopback; 1103 rule.priority = MLX4_DOMAIN_UVERBS | priority; 1104 err = flow_spec_to_net_rule(&mdev->ib_dev, flow_spec, 1105 &rule.list); 1106 if (err) 1107 goto free_list; 1108 } 1109 1110 err = mlx4_flow_attach(mdev->dev, &rule, ®_id); 1111 if (err) 1112 goto free_list; 1113 1114 memcpy(&cm_flow->spec, flow_spec, sizeof(*flow_spec)); 1115 cm_flow->reg_id = reg_id; 1116 1117 if (lock_qp) 1118 mutex_lock(&mqp->mutex); 1119 list_add(&cm_flow->list, &mqp->rules_list); 1120 if (lock_qp) 1121 mutex_unlock(&mqp->mutex); 1122 1123free_list: 1124 list_for_each_entry_safe(spec, tmp_spec, &rule.list, list) { 1125 list_del(&spec->list); 1126 kfree(spec); 1127 } 1128 if (err) { 1129 kfree(cm_flow); 1130 dev_err(mdev->ib_dev.dma_device, 1131 "Fail to attach flow steering rule\n"); 1132 } 1133 return err; 1134} 1135 1136static int __mlx4_ib_flow_detach(struct mlx4_ib_dev *mdev, 1137 struct mlx4_ib_qp *mqp, 1138 struct ib_flow_spec *spec, int priority, 1139 int lock_qp) 1140{ 1141 struct mlx4_cm_steering *cm_flow; 1142 int ret; 1143 1144 if (lock_qp) 1145 mutex_lock(&mqp->mutex); 1146 list_for_each_entry(cm_flow, &mqp->rules_list, list) { 1147 if (!memcmp(&cm_flow->spec, spec, sizeof(*spec))) { 1148 list_del(&cm_flow->list); 1149 break; 1150 } 1151 } 1152 if (lock_qp) 1153 mutex_unlock(&mqp->mutex); 1154 1155 if (&cm_flow->list == &mqp->rules_list) { 1156 dev_err(mdev->ib_dev.dma_device, "Couldn't find reg_id for flow spec. " 1157 "Steering rule is left attached\n"); 1158 return -EINVAL; 1159 } 1160 1161 ret = mlx4_flow_detach(mdev->dev, cm_flow->reg_id); 1162 1163 kfree(cm_flow); 1164 return ret; 1165} 1166 1167static int mlx4_ib_flow_attach(struct ib_qp *qp, struct ib_flow_spec *flow_spec, 1168 int priority) 1169{ 1170 return __mlx4_ib_flow_attach(to_mdev(qp->device), to_mqp(qp), 1171 flow_spec, priority, 1); 1172} 1173 1174static int mlx4_ib_flow_detach(struct ib_qp *qp, struct ib_flow_spec *spec, 1175 int priority) 1176{ 1177 return __mlx4_ib_flow_detach(to_mdev(qp->device), to_mqp(qp), 1178 spec, priority, 1); 1179} 1180 1181static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw) 1182{ 1183 struct mlx4_ib_gid_entry *ge; 1184 struct mlx4_ib_gid_entry *tmp; 1185 struct mlx4_ib_gid_entry *ret = NULL; 1186 1187 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) { 1188 if (!memcmp(raw, ge->gid.raw, 16)) { 1189 ret = ge; 1190 break; 1191 } 1192 } 1193 1194 return ret; 1195} 1196 1197static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid) 1198{ 1199 int err; 1200 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device); 1201 struct mlx4_ib_qp *mqp = to_mqp(ibqp); 1202 u8 mac[6]; 1203 struct net_device *ndev; 1204 struct mlx4_ib_gid_entry *ge; 1205 u64 reg_id = 0; 1206 1207 if (mdev->dev->caps.steering_mode == 1208 MLX4_STEERING_MODE_DEVICE_MANAGED) { 1209 struct mlx4_ib_steering *ib_steering; 1210 1211 mutex_lock(&mqp->mutex); 1212 list_for_each_entry(ib_steering, &mqp->steering_rules, list) { 1213 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) { 1214 list_del(&ib_steering->list); 1215 break; 1216 } 1217 } 1218 mutex_unlock(&mqp->mutex); 1219 if (&ib_steering->list == &mqp->steering_rules) { 1220 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n"); 1221 return -EINVAL; 1222 } 1223 reg_id = ib_steering->reg_id; 1224 kfree(ib_steering); 1225 } 1226 1227 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw, 1228 MLX4_PROT_IB_IPV6, reg_id); 1229 if (err) 1230 return err; 1231 1232 mutex_lock(&mqp->mutex); 1233 ge = find_gid_entry(mqp, gid->raw); 1234 if (ge) { 1235 spin_lock(&mdev->iboe.lock); 1236 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL; 1237 if (ndev) 1238 dev_hold(ndev); 1239 spin_unlock(&mdev->iboe.lock); 1240 rdma_get_mcast_mac((struct in6_addr *)gid, mac); 1241 if (ndev) { 1242 rtnl_lock(); 1243 dev_mc_delete(mdev->iboe.netdevs[ge->port - 1], mac, 6, 0); 1244 rtnl_unlock(); 1245 dev_put(ndev); 1246 } 1247 list_del(&ge->list); 1248 kfree(ge); 1249 } else 1250 pr_warn("could not find mgid entry\n"); 1251 1252 mutex_unlock(&mqp->mutex); 1253 1254 return 0; 1255} 1256 1257static int init_node_data(struct mlx4_ib_dev *dev) 1258{ 1259 struct ib_smp *in_mad = NULL; 1260 struct ib_smp *out_mad = NULL; 1261 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS; 1262 int err = -ENOMEM; 1263 1264 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL); 1265 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL); 1266 if (!in_mad || !out_mad) 1267 goto out; 1268 1269 init_query_mad(in_mad); 1270 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC; 1271 if (mlx4_is_master(dev->dev)) 1272 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW; 1273 1274 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad); 1275 if (err) 1276 goto out; 1277 1278 memcpy(dev->ib_dev.node_desc, out_mad->data, 64); 1279 1280 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO; 1281 1282 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad); 1283 if (err) 1284 goto out; 1285 1286 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32)); 1287 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8); 1288 1289out: 1290 kfree(in_mad); 1291 kfree(out_mad); 1292 return err; 1293} 1294 1295static ssize_t show_hca(struct device *device, struct device_attribute *attr, 1296 char *buf) 1297{ 1298 struct mlx4_ib_dev *dev = 1299 container_of(device, struct mlx4_ib_dev, ib_dev.dev); 1300 return sprintf(buf, "MT%d\n", dev->dev->pdev->device); 1301} 1302 1303static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr, 1304 char *buf) 1305{ 1306 struct mlx4_ib_dev *dev = 1307 container_of(device, struct mlx4_ib_dev, ib_dev.dev); 1308 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32), 1309 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff, 1310 (int) dev->dev->caps.fw_ver & 0xffff); 1311} 1312 1313static ssize_t show_rev(struct device *device, struct device_attribute *attr, 1314 char *buf) 1315{ 1316 struct mlx4_ib_dev *dev = 1317 container_of(device, struct mlx4_ib_dev, ib_dev.dev); 1318 return sprintf(buf, "%x\n", dev->dev->rev_id); 1319} 1320 1321static ssize_t show_board(struct device *device, struct device_attribute *attr, 1322 char *buf) 1323{ 1324 struct mlx4_ib_dev *dev = 1325 container_of(device, struct mlx4_ib_dev, ib_dev.dev); 1326 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN, 1327 dev->dev->board_id); 1328} 1329 1330static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL); 1331static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL); 1332static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL); 1333static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL); 1334 1335static struct device_attribute *mlx4_class_attributes[] = { 1336 &dev_attr_hw_rev, 1337 &dev_attr_fw_ver, 1338 &dev_attr_hca_type, 1339 &dev_attr_board_id 1340}; 1341 1342static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev) 1343{ 1344#ifdef __linux__ 1345 memcpy(eui, dev->dev_addr, 3); 1346 memcpy(eui + 5, dev->dev_addr + 3, 3); 1347#else 1348 memcpy(eui, IF_LLADDR(dev), 3); 1349 memcpy(eui + 5, IF_LLADDR(dev) + 3, 3); 1350#endif 1351 if (vlan_id < 0x1000) { 1352 eui[3] = vlan_id >> 8; 1353 eui[4] = vlan_id & 0xff; 1354 } else { 1355 eui[3] = 0xff; 1356 eui[4] = 0xfe; 1357 } 1358 eui[0] ^= 2; 1359} 1360 1361static void update_gids_task(struct work_struct *work) 1362{ 1363 struct update_gid_work *gw = container_of(work, struct update_gid_work, work); 1364 struct mlx4_cmd_mailbox *mailbox; 1365 union ib_gid *gids; 1366 int err; 1367 struct mlx4_dev *dev = gw->dev->dev; 1368 1369 mailbox = mlx4_alloc_cmd_mailbox(dev); 1370 if (IS_ERR(mailbox)) { 1371 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox)); 1372 return; 1373 } 1374 1375 gids = mailbox->buf; 1376 memcpy(gids, gw->gids, sizeof gw->gids); 1377 1378 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port, 1379 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B, 1380 MLX4_CMD_WRAPPED); 1381 if (err) 1382 pr_warn("set port command failed\n"); 1383 else { 1384 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids); 1385 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE); 1386 } 1387 1388 mlx4_free_cmd_mailbox(dev, mailbox); 1389 kfree(gw); 1390} 1391 1392static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear) 1393{ 1394 struct net_device *ndev = dev->iboe.netdevs[port - 1]; 1395 struct update_gid_work *work; 1396 struct net_device *tmp; 1397 int i; 1398 u8 *hits; 1399 union ib_gid gid; 1400 int index_free; 1401 int found; 1402 int need_update = 0; 1403 int max_gids; 1404 u16 vid; 1405 1406 work = kzalloc(sizeof *work, GFP_ATOMIC); 1407 if (!work) 1408 return -ENOMEM; 1409 1410 hits = kzalloc(128, GFP_ATOMIC); 1411 if (!hits) { 1412 kfree(work); 1413 return -ENOMEM; 1414 } 1415 1416 max_gids = dev->dev->caps.gid_table_len[port]; 1417 1418#ifdef __linux__ 1419 rcu_read_lock(); 1420 for_each_netdev_rcu(&init_net, tmp) { 1421#else 1422 IFNET_RLOCK(); 1423 TAILQ_FOREACH(tmp, &V_ifnet, if_link) { 1424#endif 1425 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) { 1426 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL); 1427 vid = rdma_vlan_dev_vlan_id(tmp); 1428 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev); 1429 found = 0; 1430 index_free = -1; 1431 for (i = 0; i < max_gids; ++i) { 1432 if (index_free < 0 && 1433 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid)) 1434 index_free = i; 1435 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) { 1436 hits[i] = 1; 1437 found = 1; 1438 break; 1439 } 1440 } 1441 1442 if (!found) { 1443 if (tmp == ndev && 1444 (memcmp(&dev->iboe.gid_table[port - 1][0], 1445 &gid, sizeof gid) || 1446 !memcmp(&dev->iboe.gid_table[port - 1][0], 1447 &zgid, sizeof gid))) { 1448 dev->iboe.gid_table[port - 1][0] = gid; 1449 ++need_update; 1450 hits[0] = 1; 1451 } else if (index_free >= 0) { 1452 dev->iboe.gid_table[port - 1][index_free] = gid; 1453 hits[index_free] = 1; 1454 ++need_update; 1455 } 1456 } 1457 } 1458#ifdef __linux__ 1459 } 1460 rcu_read_unlock(); 1461#else 1462 } 1463 IFNET_RUNLOCK(); 1464#endif 1465 1466 for (i = 0; i < max_gids; ++i) 1467 if (!hits[i]) { 1468 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid)) 1469 ++need_update; 1470 dev->iboe.gid_table[port - 1][i] = zgid; 1471 } 1472 1473 if (need_update) { 1474 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids); 1475 INIT_WORK(&work->work, update_gids_task); 1476 work->port = port; 1477 work->dev = dev; 1478 queue_work(wq, &work->work); 1479 } else 1480 kfree(work); 1481 1482 kfree(hits); 1483 return 0; 1484} 1485 1486static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event) 1487{ 1488 switch (event) { 1489 case NETDEV_UP: 1490#ifdef __linux__ 1491 case NETDEV_CHANGEADDR: 1492#endif 1493 update_ipv6_gids(dev, port, 0); 1494 break; 1495 1496 case NETDEV_DOWN: 1497 update_ipv6_gids(dev, port, 1); 1498 dev->iboe.netdevs[port - 1] = NULL; 1499 } 1500} 1501 1502static void netdev_added(struct mlx4_ib_dev *dev, int port) 1503{ 1504 update_ipv6_gids(dev, port, 0); 1505} 1506 1507static void netdev_removed(struct mlx4_ib_dev *dev, int port) 1508{ 1509 update_ipv6_gids(dev, port, 1); 1510} 1511 1512static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event, 1513 void *ptr) 1514{ 1515 struct net_device *dev = ptr; 1516 struct mlx4_ib_dev *ibdev; 1517 struct net_device *oldnd; 1518 struct mlx4_ib_iboe *iboe; 1519 int port; 1520 1521#ifdef __linux__ 1522 if (!net_eq(dev_net(dev), &init_net)) 1523 return NOTIFY_DONE; 1524#endif 1525 1526 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb); 1527 iboe = &ibdev->iboe; 1528 1529 spin_lock(&iboe->lock); 1530 mlx4_foreach_ib_transport_port(port, ibdev->dev) { 1531 oldnd = iboe->netdevs[port - 1]; 1532 iboe->netdevs[port - 1] = 1533 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port); 1534 if (oldnd != iboe->netdevs[port - 1]) { 1535 if (iboe->netdevs[port - 1]) 1536 netdev_added(ibdev, port); 1537 else 1538 netdev_removed(ibdev, port); 1539 } 1540 } 1541 1542 if (dev == iboe->netdevs[0] || 1543 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0])) 1544 handle_en_event(ibdev, 1, event); 1545 else if (dev == iboe->netdevs[1] 1546 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1])) 1547 handle_en_event(ibdev, 2, event); 1548 1549 spin_unlock(&iboe->lock); 1550 1551 return NOTIFY_DONE; 1552} 1553 1554static void init_pkeys(struct mlx4_ib_dev *ibdev) 1555{ 1556 int port; 1557 int slave; 1558 int i; 1559 1560 if (mlx4_is_master(ibdev->dev)) { 1561 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) { 1562 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) { 1563 for (i = 0; 1564 i < ibdev->dev->phys_caps.pkey_phys_table_len[port]; 1565 ++i) { 1566 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] = 1567 /* master has the identity virt2phys pkey mapping */ 1568 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i : 1569 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1; 1570 mlx4_sync_pkey_table(ibdev->dev, slave, port, i, 1571 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]); 1572 } 1573 } 1574 } 1575 /* initialize pkey cache */ 1576 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) { 1577 for (i = 0; 1578 i < ibdev->dev->phys_caps.pkey_phys_table_len[port]; 1579 ++i) 1580 ibdev->pkeys.phys_pkey_cache[port-1][i] = 1581 (i) ? 0 : 0xFFFF; 1582 } 1583 } 1584} 1585 1586static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev) 1587{ 1588 char name[32]; 1589 int eq_per_port = 0; 1590 int added_eqs = 0; 1591 int total_eqs = 0; 1592 int i, j, eq; 1593 1594 /* Legacy mode or comp_pool is not large enough */ 1595 if (dev->caps.comp_pool == 0 || 1596 dev->caps.num_ports > dev->caps.comp_pool) 1597 return; 1598 1599 eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/ 1600 dev->caps.num_ports); 1601 1602 /* Init eq table */ 1603 added_eqs = 0; 1604 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) 1605 added_eqs += eq_per_port; 1606 1607 total_eqs = dev->caps.num_comp_vectors + added_eqs; 1608 1609 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL); 1610 if (!ibdev->eq_table) 1611 return; 1612 1613 ibdev->eq_added = added_eqs; 1614 1615 eq = 0; 1616 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) { 1617 for (j = 0; j < eq_per_port; j++) { 1618 snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%d:%d:%d:%d", i, j, 1619 pci_get_domain(dev->pdev->dev.bsddev), 1620 pci_get_bus(dev->pdev->dev.bsddev), 1621 PCI_SLOT(dev->pdev->devfn), 1622 PCI_FUNC(dev->pdev->devfn)); 1623 1624 /* Set IRQ for specific name (per ring) */ 1625 if (mlx4_assign_eq(dev, name, 1626 &ibdev->eq_table[eq])) { 1627 /* Use legacy (same as mlx4_en driver) */ 1628 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq); 1629 ibdev->eq_table[eq] = 1630 (eq % dev->caps.num_comp_vectors); 1631 } 1632 eq++; 1633 } 1634 } 1635 1636 /* Fill the reset of the vector with legacy EQ */ 1637 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++) 1638 ibdev->eq_table[eq++] = i; 1639 1640 /* Advertise the new number of EQs to clients */ 1641 ibdev->ib_dev.num_comp_vectors = total_eqs; 1642} 1643 1644static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev) 1645{ 1646 int i; 1647 1648 /* no additional eqs were added */ 1649 if (!ibdev->eq_table) 1650 return; 1651 1652 /* Reset the advertised EQ number */ 1653 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors; 1654 1655 /* Free only the added eqs */ 1656 for (i = 0; i < ibdev->eq_added; i++) { 1657 /* Don't free legacy eqs if used */ 1658 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors) 1659 continue; 1660 mlx4_release_eq(dev, ibdev->eq_table[i]); 1661 } 1662 1663 kfree(ibdev->eq_table); 1664} 1665 1666/* 1667 * create show function and a device_attribute struct pointing to 1668 * the function for _name 1669 */ 1670#define DEVICE_DIAG_RPRT_ATTR(_name, _offset, _op_mod) \ 1671static ssize_t show_rprt_##_name(struct device *dev, \ 1672 struct device_attribute *attr, \ 1673 char *buf){ \ 1674 return show_diag_rprt(dev, buf, _offset, _op_mod); \ 1675} \ 1676static DEVICE_ATTR(_name, S_IRUGO, show_rprt_##_name, NULL); 1677 1678#define MLX4_DIAG_RPRT_CLEAR_DIAGS 3 1679 1680static size_t show_diag_rprt(struct device *device, char *buf, 1681 u32 offset, u8 op_modifier) 1682{ 1683 size_t ret; 1684 u32 counter_offset = offset; 1685 u32 diag_counter = 0; 1686 struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev, 1687 ib_dev.dev); 1688 1689 ret = mlx4_query_diag_counters(dev->dev, 1, op_modifier, 1690 &counter_offset, &diag_counter); 1691 if (ret) 1692 return ret; 1693 1694 return sprintf(buf, "%d\n", diag_counter); 1695} 1696 1697static ssize_t clear_diag_counters(struct device *device, 1698 struct device_attribute *attr, 1699 const char *buf, size_t length) 1700{ 1701 size_t ret; 1702 struct mlx4_ib_dev *dev = container_of(device, struct mlx4_ib_dev, 1703 ib_dev.dev); 1704 1705 ret = mlx4_query_diag_counters(dev->dev, 0, MLX4_DIAG_RPRT_CLEAR_DIAGS, 1706 NULL, NULL); 1707 if (ret) 1708 return ret; 1709 1710 return length; 1711} 1712 1713DEVICE_DIAG_RPRT_ATTR(rq_num_lle , 0x00, 2); 1714DEVICE_DIAG_RPRT_ATTR(sq_num_lle , 0x04, 2); 1715DEVICE_DIAG_RPRT_ATTR(rq_num_lqpoe , 0x08, 2); 1716DEVICE_DIAG_RPRT_ATTR(sq_num_lqpoe , 0x0C, 2); 1717DEVICE_DIAG_RPRT_ATTR(rq_num_lpe , 0x18, 2); 1718DEVICE_DIAG_RPRT_ATTR(sq_num_lpe , 0x1C, 2); 1719DEVICE_DIAG_RPRT_ATTR(rq_num_wrfe , 0x20, 2); 1720DEVICE_DIAG_RPRT_ATTR(sq_num_wrfe , 0x24, 2); 1721DEVICE_DIAG_RPRT_ATTR(sq_num_mwbe , 0x2C, 2); 1722DEVICE_DIAG_RPRT_ATTR(sq_num_bre , 0x34, 2); 1723DEVICE_DIAG_RPRT_ATTR(rq_num_lae , 0x38, 2); 1724DEVICE_DIAG_RPRT_ATTR(sq_num_rire , 0x44, 2); 1725DEVICE_DIAG_RPRT_ATTR(rq_num_rire , 0x48, 2); 1726DEVICE_DIAG_RPRT_ATTR(sq_num_rae , 0x4C, 2); 1727DEVICE_DIAG_RPRT_ATTR(rq_num_rae , 0x50, 2); 1728DEVICE_DIAG_RPRT_ATTR(sq_num_roe , 0x54, 2); 1729DEVICE_DIAG_RPRT_ATTR(sq_num_tree , 0x5C, 2); 1730DEVICE_DIAG_RPRT_ATTR(sq_num_rree , 0x64, 2); 1731DEVICE_DIAG_RPRT_ATTR(rq_num_rnr , 0x68, 2); 1732DEVICE_DIAG_RPRT_ATTR(sq_num_rnr , 0x6C, 2); 1733DEVICE_DIAG_RPRT_ATTR(rq_num_oos , 0x100, 2); 1734DEVICE_DIAG_RPRT_ATTR(sq_num_oos , 0x104, 2); 1735DEVICE_DIAG_RPRT_ATTR(rq_num_mce , 0x108, 2); 1736DEVICE_DIAG_RPRT_ATTR(rq_num_udsdprd , 0x118, 2); 1737DEVICE_DIAG_RPRT_ATTR(rq_num_ucsdprd , 0x120, 2); 1738DEVICE_DIAG_RPRT_ATTR(num_cqovf , 0x1A0, 2); 1739DEVICE_DIAG_RPRT_ATTR(num_eqovf , 0x1A4, 2); 1740DEVICE_DIAG_RPRT_ATTR(num_baddb , 0x1A8, 2); 1741 1742static DEVICE_ATTR(clear_diag, S_IWUSR, NULL, clear_diag_counters); 1743 1744static struct attribute *diag_rprt_attrs[] = { 1745 &dev_attr_rq_num_lle.attr, 1746 &dev_attr_sq_num_lle.attr, 1747 &dev_attr_rq_num_lqpoe.attr, 1748 &dev_attr_sq_num_lqpoe.attr, 1749 &dev_attr_rq_num_lpe.attr, 1750 &dev_attr_sq_num_lpe.attr, 1751 &dev_attr_rq_num_wrfe.attr, 1752 &dev_attr_sq_num_wrfe.attr, 1753 &dev_attr_sq_num_mwbe.attr, 1754 &dev_attr_sq_num_bre.attr, 1755 &dev_attr_rq_num_lae.attr, 1756 &dev_attr_sq_num_rire.attr, 1757 &dev_attr_rq_num_rire.attr, 1758 &dev_attr_sq_num_rae.attr, 1759 &dev_attr_rq_num_rae.attr, 1760 &dev_attr_sq_num_roe.attr, 1761 &dev_attr_sq_num_tree.attr, 1762 &dev_attr_sq_num_rree.attr, 1763 &dev_attr_rq_num_rnr.attr, 1764 &dev_attr_sq_num_rnr.attr, 1765 &dev_attr_rq_num_oos.attr, 1766 &dev_attr_sq_num_oos.attr, 1767 &dev_attr_rq_num_mce.attr, 1768 &dev_attr_rq_num_udsdprd.attr, 1769 &dev_attr_rq_num_ucsdprd.attr, 1770 &dev_attr_num_cqovf.attr, 1771 &dev_attr_num_eqovf.attr, 1772 &dev_attr_num_baddb.attr, 1773 &dev_attr_clear_diag.attr, 1774 NULL 1775}; 1776 1777static struct attribute_group diag_counters_group = { 1778 .name = "diag_counters", 1779 .attrs = diag_rprt_attrs 1780}; 1781 1782#ifdef __linux__ 1783static int mlx4_ib_proc_init(void) 1784{ 1785 /* Creating procfs directories /proc/drivers/mlx4_ib/ && 1786 /proc/drivers/mlx4_ib/mrs for further use by the driver. 1787 */ 1788 int err; 1789 1790 mlx4_ib_driver_dir_entry = proc_mkdir(MLX4_IB_DRIVER_PROC_DIR_NAME, 1791 NULL); 1792 if (!mlx4_ib_driver_dir_entry) { 1793 pr_err("mlx4_ib_proc_init has failed for %s\n", 1794 MLX4_IB_DRIVER_PROC_DIR_NAME); 1795 err = -ENODEV; 1796 goto error; 1797 } 1798 1799 mlx4_mrs_dir_entry = proc_mkdir(MLX4_IB_MRS_PROC_DIR_NAME, 1800 mlx4_ib_driver_dir_entry); 1801 if (!mlx4_mrs_dir_entry) { 1802 pr_err("mlx4_ib_proc_init has failed for %s\n", 1803 MLX4_IB_MRS_PROC_DIR_NAME); 1804 err = -ENODEV; 1805 goto remove_entry; 1806 } 1807 1808 return 0; 1809 1810remove_entry: 1811 remove_proc_entry(MLX4_IB_DRIVER_PROC_DIR_NAME, 1812 NULL); 1813error: 1814 return err; 1815} 1816#endif 1817 1818static void init_dev_assign(void) 1819{ 1820 int bus, slot, fn, ib_idx; 1821 char *p = dev_assign_str, *t; 1822 char curr_val[32] = {0}; 1823 int ret; 1824 int j, i = 0; 1825 1826 memset(dr, 0, sizeof dr); 1827 1828 if (dev_assign_str[0] == 0) 1829 return; 1830 1831 while (strlen(p)) { 1832 ret = sscanf(p, "%02x:%02x.%x-%x", &bus, &slot, &fn, &ib_idx); 1833 if (ret != 4 || ib_idx < 0) 1834 goto err; 1835 1836 for (j = 0; j < i; j++) 1837 if (dr[j].nr == ib_idx) 1838 goto err; 1839 1840 dr[i].bus = bus; 1841 dr[i].dev = slot; 1842 dr[i].func = fn; 1843 dr[i].nr = ib_idx; 1844 1845 t = strchr(p, ','); 1846 sprintf(curr_val, "%02x:%02x.%x-%x", bus, slot, fn, ib_idx); 1847 if ((!t) && strlen(p) == strlen(curr_val)) 1848 return; 1849 1850 if (!t || (t + 1) >= dev_assign_str + sizeof dev_assign_str) 1851 goto err; 1852 1853 ++i; 1854 if (i >= MAX_DR) 1855 goto err; 1856 1857 p = t + 1; 1858 } 1859 1860 return; 1861err: 1862 memset(dr, 0, sizeof dr); 1863 printk(KERN_WARNING "mlx4_ib: The value of 'dev_assign_str' parameter " 1864 "is incorrect. The parameter value is discarded!"); 1865} 1866 1867static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num, 1868 struct ib_port_immutable *immutable) 1869{ 1870 struct ib_port_attr attr; 1871 struct mlx4_ib_dev *mdev = to_mdev(ibdev); 1872 int err; 1873 1874 if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) { 1875 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB; 1876 immutable->max_mad_size = IB_MGMT_MAD_SIZE; 1877 } else { 1878 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) 1879 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE; 1880 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCEV2) 1881 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE | 1882 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP; 1883 immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET; 1884 if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE | 1885 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP)) 1886 immutable->max_mad_size = IB_MGMT_MAD_SIZE; 1887 } 1888 1889 err = ib_query_port(ibdev, port_num, &attr); 1890 if (err) 1891 return err; 1892 1893 immutable->pkey_tbl_len = attr.pkey_tbl_len; 1894 immutable->gid_tbl_len = attr.gid_tbl_len; 1895 1896 return 0; 1897} 1898 1899static void *mlx4_ib_add(struct mlx4_dev *dev) 1900{ 1901 struct mlx4_ib_dev *ibdev; 1902 int num_ports = 0; 1903 int i, j; 1904 int err; 1905 struct mlx4_ib_iboe *iboe; 1906 1907 printk(KERN_INFO "%s", mlx4_ib_version); 1908 1909 mlx4_foreach_ib_transport_port(i, dev) 1910 num_ports++; 1911 1912 /* No point in registering a device with no ports... */ 1913 if (num_ports == 0) 1914 return NULL; 1915 1916 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev); 1917 if (!ibdev) { 1918 dev_err(&dev->pdev->dev, "Device struct alloc failed\n"); 1919 return NULL; 1920 } 1921 1922 iboe = &ibdev->iboe; 1923 1924 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn)) 1925 goto err_dealloc; 1926 1927 if (mlx4_uar_alloc(dev, &ibdev->priv_uar)) 1928 goto err_pd; 1929 1930 ibdev->priv_uar.map = ioremap(ibdev->priv_uar.pfn << PAGE_SHIFT, 1931 PAGE_SIZE); 1932 1933 if (!ibdev->priv_uar.map) 1934 goto err_uar; 1935 1936 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock); 1937 1938 ibdev->dev = dev; 1939 1940 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX); 1941 ibdev->ib_dev.owner = THIS_MODULE; 1942 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA; 1943 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey; 1944 ibdev->num_ports = num_ports; 1945 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports; 1946 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors; 1947 ibdev->ib_dev.dma_device = &dev->pdev->dev; 1948 1949 if (dev->caps.userspace_caps) 1950 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION; 1951 else 1952 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION; 1953 1954 ibdev->ib_dev.uverbs_cmd_mask = 1955 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | 1956 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) | 1957 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) | 1958 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | 1959 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | 1960 (1ull << IB_USER_VERBS_CMD_REG_MR) | 1961 (1ull << IB_USER_VERBS_CMD_DEREG_MR) | 1962 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | 1963 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | 1964 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) | 1965 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) | 1966 (1ull << IB_USER_VERBS_CMD_CREATE_QP) | 1967 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | 1968 (1ull << IB_USER_VERBS_CMD_QUERY_QP) | 1969 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | 1970 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) | 1971 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) | 1972 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) | 1973 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) | 1974 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) | 1975 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) | 1976 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) | 1977 (1ull << IB_USER_VERBS_CMD_OPEN_QP) | 1978 (1ull << IB_USER_VERBS_CMD_ATTACH_FLOW) | 1979 (1ull << IB_USER_VERBS_CMD_DETACH_FLOW) | 1980 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ); 1981 1982 ibdev->ib_dev.query_device = mlx4_ib_query_device; 1983 ibdev->ib_dev.query_port = mlx4_ib_query_port; 1984 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer; 1985 ibdev->ib_dev.query_gid = mlx4_ib_query_gid; 1986 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey; 1987 ibdev->ib_dev.modify_device = mlx4_ib_modify_device; 1988 ibdev->ib_dev.modify_port = mlx4_ib_modify_port; 1989 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext; 1990 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext; 1991 ibdev->ib_dev.mmap = mlx4_ib_mmap; 1992#ifdef __linux__ 1993 ibdev->ib_dev.get_unmapped_area = mlx4_ib_get_unmapped_area; 1994#endif 1995 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd; 1996 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd; 1997 ibdev->ib_dev.create_ah = mlx4_ib_create_ah; 1998 ibdev->ib_dev.query_ah = mlx4_ib_query_ah; 1999 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah; 2000 ibdev->ib_dev.create_srq = mlx4_ib_create_srq; 2001 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq; 2002 ibdev->ib_dev.query_srq = mlx4_ib_query_srq; 2003 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq; 2004 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv; 2005 ibdev->ib_dev.create_qp = mlx4_ib_create_qp; 2006 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp; 2007 ibdev->ib_dev.query_qp = mlx4_ib_query_qp; 2008 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp; 2009 ibdev->ib_dev.post_send = mlx4_ib_post_send; 2010 ibdev->ib_dev.post_recv = mlx4_ib_post_recv; 2011 ibdev->ib_dev.create_cq = mlx4_ib_create_cq; 2012 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq; 2013 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq; 2014 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq; 2015 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq; 2016 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq; 2017 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr; 2018 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr; 2019 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr; 2020 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr; 2021 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list; 2022 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list; 2023 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach; 2024 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach; 2025 ibdev->ib_dev.attach_flow = mlx4_ib_flow_attach; 2026 ibdev->ib_dev.detach_flow = mlx4_ib_flow_detach; 2027 ibdev->ib_dev.process_mad = mlx4_ib_process_mad; 2028 ibdev->ib_dev.get_port_immutable = mlx4_port_immutable; 2029 2030 if (!mlx4_is_slave(ibdev->dev)) { 2031 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc; 2032 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr; 2033 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr; 2034 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc; 2035 } 2036 2037 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) { 2038 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd; 2039 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd; 2040 ibdev->ib_dev.uverbs_cmd_mask |= 2041 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) | 2042 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD); 2043 } 2044 2045 mlx4_ib_alloc_eqs(dev, ibdev); 2046 2047 spin_lock_init(&iboe->lock); 2048 2049 if (init_node_data(ibdev)) 2050 goto err_map; 2051 2052 for (i = 0; i < ibdev->num_ports; ++i) { 2053 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) == 2054 IB_LINK_LAYER_ETHERNET) { 2055 err = mlx4_counter_alloc(ibdev->dev, i + 1, &ibdev->counters[i]); 2056 if (err) 2057 ibdev->counters[i] = -1; 2058 } else 2059 ibdev->counters[i] = -1; 2060 } 2061 2062 spin_lock_init(&ibdev->sm_lock); 2063 mutex_init(&ibdev->cap_mask_mutex); 2064 2065 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED && 2066 !mlx4_is_slave(dev)) { 2067 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS; 2068 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count, 2069 MLX4_IB_UC_STEER_QPN_ALIGN, &ibdev->steer_qpn_base, 0); 2070 if (err) 2071 goto err_counter; 2072 2073 ibdev->ib_uc_qpns_bitmap = 2074 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) * 2075 sizeof(long), 2076 GFP_KERNEL); 2077 if (!ibdev->ib_uc_qpns_bitmap) { 2078 dev_err(&dev->pdev->dev, "bit map alloc failed\n"); 2079 goto err_steer_qp_release; 2080 } 2081 2082 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count); 2083 2084 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(dev, ibdev->steer_qpn_base, 2085 ibdev->steer_qpn_base + ibdev->steer_qpn_count - 1); 2086 if (err) 2087 goto err_steer_free_bitmap; 2088 } 2089 2090 if (ib_register_device(&ibdev->ib_dev, NULL)) 2091 goto err_steer_free_bitmap; 2092 2093 if (mlx4_ib_mad_init(ibdev)) 2094 goto err_reg; 2095 2096 if (mlx4_ib_init_sriov(ibdev)) 2097 goto err_mad; 2098 2099 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) { 2100 iboe->nb.notifier_call = mlx4_ib_netdev_event; 2101 err = register_netdevice_notifier(&iboe->nb); 2102 if (err) 2103 goto err_sriov; 2104 } 2105 2106 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) { 2107 if (device_create_file(&ibdev->ib_dev.dev, 2108 mlx4_class_attributes[j])) 2109 goto err_notif; 2110 } 2111 if (sysfs_create_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group)) 2112 goto err_notif; 2113 2114 ibdev->ib_active = true; 2115 2116 if (mlx4_is_mfunc(ibdev->dev)) 2117 init_pkeys(ibdev); 2118 2119 /* create paravirt contexts for any VFs which are active */ 2120 if (mlx4_is_master(ibdev->dev)) { 2121 for (j = 0; j < MLX4_MFUNC_MAX; j++) { 2122 if (j == mlx4_master_func_num(ibdev->dev)) 2123 continue; 2124 if (mlx4_is_slave_active(ibdev->dev, j)) 2125 do_slave_init(ibdev, j, 1); 2126 } 2127 } 2128 return ibdev; 2129 2130err_notif: 2131 if (unregister_netdevice_notifier(&ibdev->iboe.nb)) 2132 pr_warn("failure unregistering notifier\n"); 2133 flush_workqueue(wq); 2134 2135err_sriov: 2136 mlx4_ib_close_sriov(ibdev); 2137 2138err_mad: 2139 mlx4_ib_mad_cleanup(ibdev); 2140 2141err_reg: 2142 ib_unregister_device(&ibdev->ib_dev); 2143 2144err_steer_free_bitmap: 2145 kfree(ibdev->ib_uc_qpns_bitmap); 2146 2147err_steer_qp_release: 2148 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) 2149 mlx4_qp_release_range(dev, ibdev->steer_qpn_base, 2150 ibdev->steer_qpn_count); 2151err_counter: 2152 for (; i; --i) 2153 if (ibdev->counters[i - 1] != -1) 2154 mlx4_counter_free(ibdev->dev, i, ibdev->counters[i - 1]); 2155 2156err_map: 2157 iounmap(ibdev->priv_uar.map); 2158 mlx4_ib_free_eqs(dev, ibdev); 2159 2160err_uar: 2161 mlx4_uar_free(dev, &ibdev->priv_uar); 2162 2163err_pd: 2164 mlx4_pd_free(dev, ibdev->priv_pdn); 2165 2166err_dealloc: 2167 ib_dealloc_device(&ibdev->ib_dev); 2168 2169 return NULL; 2170} 2171 2172int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn) 2173{ 2174 int offset; 2175 2176 WARN_ON(!dev->ib_uc_qpns_bitmap); 2177 2178 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap, 2179 dev->steer_qpn_count, 2180 get_count_order(count)); 2181 if (offset < 0) 2182 return offset; 2183 2184 *qpn = dev->steer_qpn_base + offset; 2185 return 0; 2186} 2187 2188void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count) 2189{ 2190 if (!qpn || 2191 dev->dev->caps.steering_mode != MLX4_STEERING_MODE_DEVICE_MANAGED) 2192 return; 2193 2194 BUG_ON(qpn < dev->steer_qpn_base); 2195 2196 bitmap_release_region(dev->ib_uc_qpns_bitmap, 2197 qpn - dev->steer_qpn_base, get_count_order(count)); 2198} 2199 2200int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp, 2201 int is_attach) 2202{ 2203 struct ib_flow_spec spec = { 2204 .type = IB_FLOW_IB_UC, 2205 .l2_id.ib_uc.qpn = mqp->ibqp.qp_num, 2206 }; 2207 2208 return is_attach ? 2209 __mlx4_ib_flow_attach(mdev, mqp, &spec, MLX4_DOMAIN_NIC, 0) 2210 : __mlx4_ib_flow_detach(mdev, mqp, &spec, MLX4_DOMAIN_NIC, 0); 2211} 2212 2213static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr) 2214{ 2215 struct mlx4_ib_dev *ibdev = ibdev_ptr; 2216 int p,j; 2217 2218 mlx4_ib_close_sriov(ibdev); 2219 sysfs_remove_group(&ibdev->ib_dev.dev.kobj, &diag_counters_group); 2220 mlx4_ib_mad_cleanup(ibdev); 2221 2222 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) { 2223 device_remove_file(&ibdev->ib_dev.dev, mlx4_class_attributes[j]); 2224 } 2225 2226 ib_unregister_device(&ibdev->ib_dev); 2227 2228 if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) { 2229 mlx4_qp_release_range(dev, ibdev->steer_qpn_base, 2230 ibdev->steer_qpn_count); 2231 kfree(ibdev->ib_uc_qpns_bitmap); 2232 } 2233 2234 if (ibdev->iboe.nb.notifier_call) { 2235 if (unregister_netdevice_notifier(&ibdev->iboe.nb)) 2236 pr_warn("failure unregistering notifier\n"); 2237 ibdev->iboe.nb.notifier_call = NULL; 2238 } 2239 iounmap(ibdev->priv_uar.map); 2240 for (p = 0; p < ibdev->num_ports; ++p) 2241 if (ibdev->counters[p] != -1) 2242 mlx4_counter_free(ibdev->dev, p + 1, ibdev->counters[p]); 2243 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB) 2244 mlx4_CLOSE_PORT(dev, p); 2245 2246 mlx4_ib_free_eqs(dev, ibdev); 2247 2248 mlx4_uar_free(dev, &ibdev->priv_uar); 2249 mlx4_pd_free(dev, ibdev->priv_pdn); 2250 ib_dealloc_device(&ibdev->ib_dev); 2251} 2252 2253static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init) 2254{ 2255 struct mlx4_ib_demux_work **dm = NULL; 2256 struct mlx4_dev *dev = ibdev->dev; 2257 int i; 2258 unsigned long flags; 2259 2260 if (!mlx4_is_master(dev)) 2261 return; 2262 2263 dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC); 2264 if (!dm) { 2265 pr_err("failed to allocate memory for tunneling qp update\n"); 2266 goto out; 2267 } 2268 2269 for (i = 0; i < dev->caps.num_ports; i++) { 2270 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC); 2271 if (!dm[i]) { 2272 pr_err("failed to allocate memory for tunneling qp update work struct\n"); 2273 for (i = 0; i < dev->caps.num_ports; i++) { 2274 if (dm[i]) 2275 kfree(dm[i]); 2276 } 2277 goto out; 2278 } 2279 } 2280 /* initialize or tear down tunnel QPs for the slave */ 2281 for (i = 0; i < dev->caps.num_ports; i++) { 2282 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work); 2283 dm[i]->port = i + 1; 2284 dm[i]->slave = slave; 2285 dm[i]->do_init = do_init; 2286 dm[i]->dev = ibdev; 2287 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags); 2288 if (!ibdev->sriov.is_going_down) 2289 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work); 2290 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags); 2291 } 2292out: 2293 if (dm) 2294 kfree(dm); 2295 return; 2296} 2297 2298static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr, 2299 enum mlx4_dev_event event, unsigned long param) 2300{ 2301 struct ib_event ibev; 2302 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr); 2303 struct mlx4_eqe *eqe = NULL; 2304 struct ib_event_work *ew; 2305 int p = 0; 2306 2307 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE) 2308 eqe = (struct mlx4_eqe *)param; 2309 else 2310 p = (int) param; 2311 2312 switch (event) { 2313 case MLX4_DEV_EVENT_PORT_UP: 2314 if (p > ibdev->num_ports) 2315 return; 2316 if (mlx4_is_master(dev) && 2317 rdma_port_get_link_layer(&ibdev->ib_dev, p) == 2318 IB_LINK_LAYER_INFINIBAND) { 2319 mlx4_ib_invalidate_all_guid_record(ibdev, p); 2320 } 2321 mlx4_ib_info((struct ib_device *) ibdev_ptr, 2322 "Port %d logical link is up\n", p); 2323 ibev.event = IB_EVENT_PORT_ACTIVE; 2324 break; 2325 2326 case MLX4_DEV_EVENT_PORT_DOWN: 2327 if (p > ibdev->num_ports) 2328 return; 2329 mlx4_ib_info((struct ib_device *) ibdev_ptr, 2330 "Port %d logical link is down\n", p); 2331 ibev.event = IB_EVENT_PORT_ERR; 2332 break; 2333 2334 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR: 2335 ibdev->ib_active = false; 2336 ibev.event = IB_EVENT_DEVICE_FATAL; 2337 break; 2338 2339 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE: 2340 ew = kmalloc(sizeof *ew, GFP_ATOMIC); 2341 if (!ew) { 2342 pr_err("failed to allocate memory for events work\n"); 2343 break; 2344 } 2345 2346 INIT_WORK(&ew->work, handle_port_mgmt_change_event); 2347 memcpy(&ew->ib_eqe, eqe, sizeof *eqe); 2348 ew->ib_dev = ibdev; 2349 /* need to queue only for port owner, which uses GEN_EQE */ 2350 if (mlx4_is_master(dev)) 2351 queue_work(wq, &ew->work); 2352 else 2353 handle_port_mgmt_change_event(&ew->work); 2354 return; 2355 2356 case MLX4_DEV_EVENT_SLAVE_INIT: 2357 /* here, p is the slave id */ 2358 do_slave_init(ibdev, p, 1); 2359 return; 2360 2361 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN: 2362 /* here, p is the slave id */ 2363 do_slave_init(ibdev, p, 0); 2364 return; 2365 2366 default: 2367 return; 2368 } 2369 2370 ibev.device = ibdev_ptr; 2371 ibev.element.port_num = (u8) p; 2372 2373 ib_dispatch_event(&ibev); 2374} 2375 2376static struct mlx4_interface mlx4_ib_interface = { 2377 .add = mlx4_ib_add, 2378 .remove = mlx4_ib_remove, 2379 .event = mlx4_ib_event, 2380 .protocol = MLX4_PROT_IB_IPV6 2381}; 2382 2383static int __init mlx4_ib_init(void) 2384{ 2385 int err; 2386 2387 wq = create_singlethread_workqueue("mlx4_ib"); 2388 if (!wq) 2389 return -ENOMEM; 2390 2391#ifdef __linux__ 2392 err = mlx4_ib_proc_init(); 2393 if (err) 2394 goto clean_wq; 2395#endif 2396 2397 err = mlx4_ib_mcg_init(); 2398 if (err) 2399 goto clean_proc; 2400 2401 init_dev_assign(); 2402 2403 err = mlx4_register_interface(&mlx4_ib_interface); 2404 if (err) 2405 goto clean_mcg; 2406 2407 return 0; 2408 2409clean_mcg: 2410 mlx4_ib_mcg_destroy(); 2411 2412clean_proc: 2413#ifdef __linux__ 2414 remove_proc_entry(MLX4_IB_MRS_PROC_DIR_NAME, 2415 mlx4_ib_driver_dir_entry); 2416 remove_proc_entry(MLX4_IB_DRIVER_PROC_DIR_NAME, NULL); 2417 2418clean_wq: 2419#endif 2420 destroy_workqueue(wq); 2421 return err; 2422} 2423 2424static void __exit mlx4_ib_cleanup(void) 2425{ 2426 mlx4_unregister_interface(&mlx4_ib_interface); 2427 mlx4_ib_mcg_destroy(); 2428 destroy_workqueue(wq); 2429 2430 /* Remove proc entries */ 2431#ifdef __linux__ 2432 remove_proc_entry(MLX4_IB_MRS_PROC_DIR_NAME, 2433 mlx4_ib_driver_dir_entry); 2434 remove_proc_entry(MLX4_IB_DRIVER_PROC_DIR_NAME, NULL); 2435#endif 2436 2437} 2438 2439module_init_order(mlx4_ib_init, SI_ORDER_MIDDLE); 2440module_exit(mlx4_ib_cleanup); 2441 2442#undef MODULE_VERSION 2443#include <sys/module.h> 2444static int 2445mlx4ib_evhand(module_t mod, int event, void *arg) 2446{ 2447 return (0); 2448} 2449 2450static moduledata_t mlx4ib_mod = { 2451 .name = "mlx4ib", 2452 .evhand = mlx4ib_evhand, 2453}; 2454 2455DECLARE_MODULE(mlx4ib, mlx4ib_mod, SI_SUB_OFED_PREINIT, SI_ORDER_ANY); 2456MODULE_DEPEND(mlx4ib, mlx4, 1, 1, 1); 2457MODULE_DEPEND(mlx4ib, ibcore, 1, 1, 1); 2458