2 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
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>
43 #include <net/addrconf.h>
45 #include <rdma/ib_smi.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_addr.h>
48 #include <rdma/ib_cache.h>
50 #include <net/bonding.h>
52 #include <linux/mlx4/driver.h>
53 #include <linux/mlx4/cmd.h>
54 #include <linux/mlx4/qp.h>
59 #define DRV_NAME MLX4_IB_DRV_NAME
60 #define DRV_VERSION "2.2-1"
61 #define DRV_RELDATE "Feb 2014"
63 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
64 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
65 #define MLX4_IB_CARD_REV_A0 0xA0
67 MODULE_AUTHOR("Roland Dreier");
68 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
69 MODULE_LICENSE("Dual BSD/GPL");
70 MODULE_VERSION(DRV_VERSION);
72 int mlx4_ib_sm_guid_assign = 0;
73 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
74 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
76 static const char mlx4_ib_version[] =
77 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
78 DRV_VERSION " (" DRV_RELDATE ")\n";
80 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
82 static struct workqueue_struct *wq;
84 static void init_query_mad(struct ib_smp *mad)
86 mad->base_version = 1;
87 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
88 mad->class_version = 1;
89 mad->method = IB_MGMT_METHOD_GET;
92 static int check_flow_steering_support(struct mlx4_dev *dev)
94 int eth_num_ports = 0;
97 int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
101 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
103 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
105 dmfs &= (!ib_num_ports ||
106 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
108 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
109 if (ib_num_ports && mlx4_is_mfunc(dev)) {
110 pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
117 static int num_ib_ports(struct mlx4_dev *dev)
122 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
128 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
130 struct mlx4_ib_dev *ibdev = to_mdev(device);
131 struct net_device *dev;
134 dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
137 if (mlx4_is_bonded(ibdev->dev)) {
138 struct net_device *upper = NULL;
140 upper = netdev_master_upper_dev_get_rcu(dev);
142 struct net_device *active;
144 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
157 static int mlx4_ib_update_gids(struct gid_entry *gids,
158 struct mlx4_ib_dev *ibdev,
161 struct mlx4_cmd_mailbox *mailbox;
163 struct mlx4_dev *dev = ibdev->dev;
165 union ib_gid *gid_tbl;
167 mailbox = mlx4_alloc_cmd_mailbox(dev);
171 gid_tbl = mailbox->buf;
173 for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
174 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
176 err = mlx4_cmd(dev, mailbox->dma,
177 MLX4_SET_PORT_GID_TABLE << 8 | port_num,
178 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
180 if (mlx4_is_bonded(dev))
181 err += mlx4_cmd(dev, mailbox->dma,
182 MLX4_SET_PORT_GID_TABLE << 8 | 2,
183 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
186 mlx4_free_cmd_mailbox(dev, mailbox);
190 static int mlx4_ib_add_gid(struct ib_device *device,
193 const union ib_gid *gid,
194 const struct ib_gid_attr *attr,
197 struct mlx4_ib_dev *ibdev = to_mdev(device);
198 struct mlx4_ib_iboe *iboe = &ibdev->iboe;
199 struct mlx4_port_gid_table *port_gid_table;
200 int free = -1, found = -1;
204 struct gid_entry *gids = NULL;
206 if (!rdma_cap_roce_gid_table(device, port_num))
209 if (port_num > MLX4_MAX_PORTS)
215 port_gid_table = &iboe->gids[port_num - 1];
216 spin_lock_bh(&iboe->lock);
217 for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
218 if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid))) {
222 if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
223 free = i; /* HW has space */
230 port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
231 if (!port_gid_table->gids[free].ctx) {
234 *context = port_gid_table->gids[free].ctx;
235 memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
236 port_gid_table->gids[free].ctx->real_index = free;
237 port_gid_table->gids[free].ctx->refcount = 1;
242 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
246 if (!ret && hw_update) {
247 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
251 for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
252 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
255 spin_unlock_bh(&iboe->lock);
257 if (!ret && hw_update) {
258 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
265 static int mlx4_ib_del_gid(struct ib_device *device,
270 struct gid_cache_context *ctx = *context;
271 struct mlx4_ib_dev *ibdev = to_mdev(device);
272 struct mlx4_ib_iboe *iboe = &ibdev->iboe;
273 struct mlx4_port_gid_table *port_gid_table;
276 struct gid_entry *gids = NULL;
278 if (!rdma_cap_roce_gid_table(device, port_num))
281 if (port_num > MLX4_MAX_PORTS)
284 port_gid_table = &iboe->gids[port_num - 1];
285 spin_lock_bh(&iboe->lock);
288 if (!ctx->refcount) {
289 unsigned int real_index = ctx->real_index;
291 memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
292 kfree(port_gid_table->gids[real_index].ctx);
293 port_gid_table->gids[real_index].ctx = NULL;
297 if (!ret && hw_update) {
300 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
304 for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
305 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
308 spin_unlock_bh(&iboe->lock);
310 if (!ret && hw_update) {
311 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
317 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
318 u8 port_num, int index)
320 struct mlx4_ib_iboe *iboe = &ibdev->iboe;
321 struct gid_cache_context *ctx = NULL;
323 struct mlx4_port_gid_table *port_gid_table;
324 int real_index = -EINVAL;
329 if (port_num > MLX4_MAX_PORTS)
332 if (mlx4_is_bonded(ibdev->dev))
335 if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
338 ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, NULL);
342 if (!memcmp(&gid, &zgid, sizeof(gid)))
345 spin_lock_irqsave(&iboe->lock, flags);
346 port_gid_table = &iboe->gids[port_num - 1];
348 for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
349 if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid))) {
350 ctx = port_gid_table->gids[i].ctx;
354 real_index = ctx->real_index;
355 spin_unlock_irqrestore(&iboe->lock, flags);
359 static int mlx4_ib_query_device(struct ib_device *ibdev,
360 struct ib_device_attr *props,
361 struct ib_udata *uhw)
363 struct mlx4_ib_dev *dev = to_mdev(ibdev);
364 struct ib_smp *in_mad = NULL;
365 struct ib_smp *out_mad = NULL;
368 struct mlx4_uverbs_ex_query_device cmd;
369 struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
370 struct mlx4_clock_params clock_params;
373 if (uhw->inlen < sizeof(cmd))
376 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
387 resp.response_length = offsetof(typeof(resp), response_length) +
388 sizeof(resp.response_length);
389 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
390 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
391 if (!in_mad || !out_mad)
394 init_query_mad(in_mad);
395 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
397 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
398 1, NULL, NULL, in_mad, out_mad);
402 memset(props, 0, sizeof *props);
404 have_ib_ports = num_ib_ports(dev->dev);
406 props->fw_ver = dev->dev->caps.fw_ver;
407 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
408 IB_DEVICE_PORT_ACTIVE_EVENT |
409 IB_DEVICE_SYS_IMAGE_GUID |
410 IB_DEVICE_RC_RNR_NAK_GEN |
411 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
412 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
413 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
414 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
415 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
416 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
417 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
418 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
419 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
420 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
421 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
422 if (dev->dev->caps.max_gso_sz &&
423 (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
424 (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
425 props->device_cap_flags |= IB_DEVICE_UD_TSO;
426 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
427 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
428 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
429 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
430 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
431 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
432 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
433 props->device_cap_flags |= IB_DEVICE_XRC;
434 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
435 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
436 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
437 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
438 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
440 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
441 if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
442 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
445 props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
447 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
449 props->vendor_part_id = dev->dev->persist->pdev->device;
450 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
451 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
453 props->max_mr_size = ~0ull;
454 props->page_size_cap = dev->dev->caps.page_size_cap;
455 props->max_qp = dev->dev->quotas.qp;
456 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
457 props->max_sge = min(dev->dev->caps.max_sq_sg,
458 dev->dev->caps.max_rq_sg);
459 props->max_sge_rd = MLX4_MAX_SGE_RD;
460 props->max_cq = dev->dev->quotas.cq;
461 props->max_cqe = dev->dev->caps.max_cqes;
462 props->max_mr = dev->dev->quotas.mpt;
463 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
464 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
465 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
466 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
467 props->max_srq = dev->dev->quotas.srq;
468 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
469 props->max_srq_sge = dev->dev->caps.max_srq_sge;
470 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
471 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
472 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
473 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
474 props->masked_atomic_cap = props->atomic_cap;
475 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
476 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
477 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
478 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
479 props->max_mcast_grp;
480 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
481 props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
482 props->timestamp_mask = 0xFFFFFFFFFFFFULL;
484 if (!mlx4_is_slave(dev->dev))
485 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
487 if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
488 resp.response_length += sizeof(resp.hca_core_clock_offset);
489 if (!err && !mlx4_is_slave(dev->dev)) {
490 resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
491 resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
496 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
507 static enum rdma_link_layer
508 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
510 struct mlx4_dev *dev = to_mdev(device)->dev;
512 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
513 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
516 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
517 struct ib_port_attr *props, int netw_view)
519 struct ib_smp *in_mad = NULL;
520 struct ib_smp *out_mad = NULL;
521 int ext_active_speed;
522 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
525 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
526 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
527 if (!in_mad || !out_mad)
530 init_query_mad(in_mad);
531 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
532 in_mad->attr_mod = cpu_to_be32(port);
534 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
535 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
537 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
543 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
544 props->lmc = out_mad->data[34] & 0x7;
545 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
546 props->sm_sl = out_mad->data[36] & 0xf;
547 props->state = out_mad->data[32] & 0xf;
548 props->phys_state = out_mad->data[33] >> 4;
549 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
551 props->gid_tbl_len = out_mad->data[50];
553 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
554 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
555 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
556 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
557 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
558 props->active_width = out_mad->data[31] & 0xf;
559 props->active_speed = out_mad->data[35] >> 4;
560 props->max_mtu = out_mad->data[41] & 0xf;
561 props->active_mtu = out_mad->data[36] >> 4;
562 props->subnet_timeout = out_mad->data[51] & 0x1f;
563 props->max_vl_num = out_mad->data[37] >> 4;
564 props->init_type_reply = out_mad->data[41] >> 4;
566 /* Check if extended speeds (EDR/FDR/...) are supported */
567 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
568 ext_active_speed = out_mad->data[62] >> 4;
570 switch (ext_active_speed) {
572 props->active_speed = IB_SPEED_FDR;
575 props->active_speed = IB_SPEED_EDR;
580 /* If reported active speed is QDR, check if is FDR-10 */
581 if (props->active_speed == IB_SPEED_QDR) {
582 init_query_mad(in_mad);
583 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
584 in_mad->attr_mod = cpu_to_be32(port);
586 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
587 NULL, NULL, in_mad, out_mad);
591 /* Checking LinkSpeedActive for FDR-10 */
592 if (out_mad->data[15] & 0x1)
593 props->active_speed = IB_SPEED_FDR10;
596 /* Avoid wrong speed value returned by FW if the IB link is down. */
597 if (props->state == IB_PORT_DOWN)
598 props->active_speed = IB_SPEED_SDR;
606 static u8 state_to_phys_state(enum ib_port_state state)
608 return state == IB_PORT_ACTIVE ? 5 : 3;
611 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
612 struct ib_port_attr *props, int netw_view)
615 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
616 struct mlx4_ib_iboe *iboe = &mdev->iboe;
617 struct net_device *ndev;
619 struct mlx4_cmd_mailbox *mailbox;
621 int is_bonded = mlx4_is_bonded(mdev->dev);
623 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
625 return PTR_ERR(mailbox);
627 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
628 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
633 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ||
634 (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
635 IB_WIDTH_4X : IB_WIDTH_1X;
636 props->active_speed = (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
637 IB_SPEED_FDR : IB_SPEED_QDR;
638 props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
639 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
640 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
641 props->pkey_tbl_len = 1;
642 props->max_mtu = IB_MTU_4096;
643 props->max_vl_num = 2;
644 props->state = IB_PORT_DOWN;
645 props->phys_state = state_to_phys_state(props->state);
646 props->active_mtu = IB_MTU_256;
647 spin_lock_bh(&iboe->lock);
648 ndev = iboe->netdevs[port - 1];
649 if (ndev && is_bonded) {
650 rcu_read_lock(); /* required to get upper dev */
651 ndev = netdev_master_upper_dev_get_rcu(ndev);
657 tmp = iboe_get_mtu(ndev->mtu);
658 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
660 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
661 IB_PORT_ACTIVE : IB_PORT_DOWN;
662 props->phys_state = state_to_phys_state(props->state);
664 spin_unlock_bh(&iboe->lock);
666 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
670 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
671 struct ib_port_attr *props, int netw_view)
675 memset(props, 0, sizeof *props);
677 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
678 ib_link_query_port(ibdev, port, props, netw_view) :
679 eth_link_query_port(ibdev, port, props, netw_view);
684 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
685 struct ib_port_attr *props)
687 /* returns host view */
688 return __mlx4_ib_query_port(ibdev, port, props, 0);
691 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
692 union ib_gid *gid, int netw_view)
694 struct ib_smp *in_mad = NULL;
695 struct ib_smp *out_mad = NULL;
697 struct mlx4_ib_dev *dev = to_mdev(ibdev);
699 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
701 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
702 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
703 if (!in_mad || !out_mad)
706 init_query_mad(in_mad);
707 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
708 in_mad->attr_mod = cpu_to_be32(port);
710 if (mlx4_is_mfunc(dev->dev) && netw_view)
711 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
713 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
717 memcpy(gid->raw, out_mad->data + 8, 8);
719 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
721 /* For any index > 0, return the null guid */
728 init_query_mad(in_mad);
729 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
730 in_mad->attr_mod = cpu_to_be32(index / 8);
732 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
733 NULL, NULL, in_mad, out_mad);
737 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
741 memset(gid->raw + 8, 0, 8);
747 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
752 if (rdma_protocol_ib(ibdev, port))
753 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
755 if (!rdma_protocol_roce(ibdev, port))
758 if (!rdma_cap_roce_gid_table(ibdev, port))
761 ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
762 if (ret == -EAGAIN) {
763 memcpy(gid, &zgid, sizeof(*gid));
770 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
771 u16 *pkey, int netw_view)
773 struct ib_smp *in_mad = NULL;
774 struct ib_smp *out_mad = NULL;
775 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
778 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
779 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
780 if (!in_mad || !out_mad)
783 init_query_mad(in_mad);
784 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
785 in_mad->attr_mod = cpu_to_be32(index / 32);
787 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
788 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
790 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
795 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
803 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
805 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
808 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
809 struct ib_device_modify *props)
811 struct mlx4_cmd_mailbox *mailbox;
814 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
817 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
820 if (mlx4_is_slave(to_mdev(ibdev)->dev))
823 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
824 memcpy(ibdev->node_desc, props->node_desc, 64);
825 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
828 * If possible, pass node desc to FW, so it can generate
829 * a 144 trap. If cmd fails, just ignore.
831 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
835 memcpy(mailbox->buf, props->node_desc, 64);
836 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
837 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
839 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
844 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
847 struct mlx4_cmd_mailbox *mailbox;
850 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
852 return PTR_ERR(mailbox);
854 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
855 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
856 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
858 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
859 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
862 err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
863 MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
866 mlx4_free_cmd_mailbox(dev->dev, mailbox);
870 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
871 struct ib_port_modify *props)
873 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
874 u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
875 struct ib_port_attr attr;
879 /* return OK if this is RoCE. CM calls ib_modify_port() regardless
880 * of whether port link layer is ETH or IB. For ETH ports, qkey
881 * violations and port capabilities are not meaningful.
886 mutex_lock(&mdev->cap_mask_mutex);
888 err = mlx4_ib_query_port(ibdev, port, &attr);
892 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
893 ~props->clr_port_cap_mask;
895 err = mlx4_ib_SET_PORT(mdev, port,
896 !!(mask & IB_PORT_RESET_QKEY_CNTR),
900 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
904 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
905 struct ib_udata *udata)
907 struct mlx4_ib_dev *dev = to_mdev(ibdev);
908 struct mlx4_ib_ucontext *context;
909 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
910 struct mlx4_ib_alloc_ucontext_resp resp;
914 return ERR_PTR(-EAGAIN);
916 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
917 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
918 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
919 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
921 resp.dev_caps = dev->dev->caps.userspace_caps;
922 resp.qp_tab_size = dev->dev->caps.num_qps;
923 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
924 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
925 resp.cqe_size = dev->dev->caps.cqe_size;
928 context = kzalloc(sizeof(*context), GFP_KERNEL);
930 return ERR_PTR(-ENOMEM);
932 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
938 INIT_LIST_HEAD(&context->db_page_list);
939 mutex_init(&context->db_page_mutex);
941 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
942 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
944 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
947 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
949 return ERR_PTR(-EFAULT);
952 return &context->ibucontext;
955 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
957 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
959 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
965 static void mlx4_ib_vma_open(struct vm_area_struct *area)
967 /* vma_open is called when a new VMA is created on top of our VMA.
968 * This is done through either mremap flow or split_vma (usually due
969 * to mlock, madvise, munmap, etc.). We do not support a clone of the
970 * vma, as this VMA is strongly hardware related. Therefore we set the
971 * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
972 * calling us again and trying to do incorrect actions. We assume that
973 * the original vma size is exactly a single page that there will be no
974 * "splitting" operations on.
979 static void mlx4_ib_vma_close(struct vm_area_struct *area)
981 struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
983 /* It's guaranteed that all VMAs opened on a FD are closed before the
984 * file itself is closed, therefore no sync is needed with the regular
985 * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
986 * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
987 * The close operation is usually called under mm->mmap_sem except when
988 * process is exiting. The exiting case is handled explicitly as part
989 * of mlx4_ib_disassociate_ucontext.
991 mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
992 area->vm_private_data;
994 /* set the vma context pointer to null in the mlx4_ib driver's private
995 * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
997 mlx4_ib_vma_priv_data->vma = NULL;
1000 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1001 .open = mlx4_ib_vma_open,
1002 .close = mlx4_ib_vma_close
1005 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1009 struct vm_area_struct *vma;
1010 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1011 struct task_struct *owning_process = NULL;
1012 struct mm_struct *owning_mm = NULL;
1014 owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1015 if (!owning_process)
1018 owning_mm = get_task_mm(owning_process);
1020 pr_info("no mm, disassociate ucontext is pending task termination\n");
1022 /* make sure that task is dead before returning, it may
1023 * prevent a rare case of module down in parallel to a
1024 * call to mlx4_ib_vma_close.
1026 put_task_struct(owning_process);
1028 owning_process = get_pid_task(ibcontext->tgid,
1030 if (!owning_process ||
1031 owning_process->state == TASK_DEAD) {
1032 pr_info("disassociate ucontext done, task was terminated\n");
1033 /* in case task was dead need to release the task struct */
1035 put_task_struct(owning_process);
1041 /* need to protect from a race on closing the vma as part of
1042 * mlx4_ib_vma_close().
1044 down_write(&owning_mm->mmap_sem);
1045 if (!mmget_still_valid(owning_mm))
1047 for (i = 0; i < HW_BAR_COUNT; i++) {
1048 vma = context->hw_bar_info[i].vma;
1052 ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1053 context->hw_bar_info[i].vma->vm_start,
1056 pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1060 context->hw_bar_info[i].vma->vm_flags &=
1061 ~(VM_SHARED | VM_MAYSHARE);
1062 /* context going to be destroyed, should not access ops any more */
1063 context->hw_bar_info[i].vma->vm_ops = NULL;
1067 up_write(&owning_mm->mmap_sem);
1069 put_task_struct(owning_process);
1072 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1073 struct mlx4_ib_vma_private_data *vma_private_data)
1075 vma_private_data->vma = vma;
1076 vma->vm_private_data = vma_private_data;
1077 vma->vm_ops = &mlx4_ib_vm_ops;
1080 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1082 struct mlx4_ib_dev *dev = to_mdev(context->device);
1083 struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1085 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1088 if (vma->vm_pgoff == 0) {
1089 /* We prevent double mmaping on same context */
1090 if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1093 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1095 if (io_remap_pfn_range(vma, vma->vm_start,
1096 to_mucontext(context)->uar.pfn,
1097 PAGE_SIZE, vma->vm_page_prot))
1100 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1102 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1103 /* We prevent double mmaping on same context */
1104 if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1107 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1109 if (io_remap_pfn_range(vma, vma->vm_start,
1110 to_mucontext(context)->uar.pfn +
1111 dev->dev->caps.num_uars,
1112 PAGE_SIZE, vma->vm_page_prot))
1115 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1117 } else if (vma->vm_pgoff == 3) {
1118 struct mlx4_clock_params params;
1121 /* We prevent double mmaping on same context */
1122 if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1125 ret = mlx4_get_internal_clock_params(dev->dev, ¶ms);
1130 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1131 if (io_remap_pfn_range(vma, vma->vm_start,
1132 (pci_resource_start(dev->dev->persist->pdev,
1136 PAGE_SIZE, vma->vm_page_prot))
1139 mlx4_ib_set_vma_data(vma,
1140 &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1148 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1149 struct ib_ucontext *context,
1150 struct ib_udata *udata)
1152 struct mlx4_ib_pd *pd;
1155 pd = kmalloc(sizeof *pd, GFP_KERNEL);
1157 return ERR_PTR(-ENOMEM);
1159 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1162 return ERR_PTR(err);
1166 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1167 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1169 return ERR_PTR(-EFAULT);
1175 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1177 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1183 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1184 struct ib_ucontext *context,
1185 struct ib_udata *udata)
1187 struct mlx4_ib_xrcd *xrcd;
1188 struct ib_cq_init_attr cq_attr = {};
1191 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1192 return ERR_PTR(-ENOSYS);
1194 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1196 return ERR_PTR(-ENOMEM);
1198 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1202 xrcd->pd = ib_alloc_pd(ibdev);
1203 if (IS_ERR(xrcd->pd)) {
1204 err = PTR_ERR(xrcd->pd);
1209 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1210 if (IS_ERR(xrcd->cq)) {
1211 err = PTR_ERR(xrcd->cq);
1215 return &xrcd->ibxrcd;
1218 ib_dealloc_pd(xrcd->pd);
1220 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1223 return ERR_PTR(err);
1226 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1228 ib_destroy_cq(to_mxrcd(xrcd)->cq);
1229 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1230 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1236 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1238 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1239 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1240 struct mlx4_ib_gid_entry *ge;
1242 ge = kzalloc(sizeof *ge, GFP_KERNEL);
1247 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1248 ge->port = mqp->port;
1252 mutex_lock(&mqp->mutex);
1253 list_add_tail(&ge->list, &mqp->gid_list);
1254 mutex_unlock(&mqp->mutex);
1259 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1260 struct mlx4_ib_counters *ctr_table)
1262 struct counter_index *counter, *tmp_count;
1264 mutex_lock(&ctr_table->mutex);
1265 list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1267 if (counter->allocated)
1268 mlx4_counter_free(ibdev->dev, counter->index);
1269 list_del(&counter->list);
1272 mutex_unlock(&ctr_table->mutex);
1275 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1278 struct net_device *ndev;
1284 spin_lock_bh(&mdev->iboe.lock);
1285 ndev = mdev->iboe.netdevs[mqp->port - 1];
1288 spin_unlock_bh(&mdev->iboe.lock);
1298 struct mlx4_ib_steering {
1299 struct list_head list;
1300 struct mlx4_flow_reg_id reg_id;
1304 static int parse_flow_attr(struct mlx4_dev *dev,
1306 union ib_flow_spec *ib_spec,
1307 struct _rule_hw *mlx4_spec)
1309 enum mlx4_net_trans_rule_id type;
1311 switch (ib_spec->type) {
1312 case IB_FLOW_SPEC_ETH:
1313 type = MLX4_NET_TRANS_RULE_ID_ETH;
1314 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1316 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1318 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1319 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1321 case IB_FLOW_SPEC_IB:
1322 type = MLX4_NET_TRANS_RULE_ID_IB;
1323 mlx4_spec->ib.l3_qpn =
1324 cpu_to_be32(qp_num);
1325 mlx4_spec->ib.qpn_mask =
1326 cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1330 case IB_FLOW_SPEC_IPV4:
1331 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1332 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1333 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1334 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1335 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1338 case IB_FLOW_SPEC_TCP:
1339 case IB_FLOW_SPEC_UDP:
1340 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1341 MLX4_NET_TRANS_RULE_ID_TCP :
1342 MLX4_NET_TRANS_RULE_ID_UDP;
1343 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1344 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1345 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1346 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1352 if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1353 mlx4_hw_rule_sz(dev, type) < 0)
1355 mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1356 mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1357 return mlx4_hw_rule_sz(dev, type);
1360 struct default_rules {
1361 __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1362 __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1363 __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1366 static const struct default_rules default_table[] = {
1368 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1369 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1370 .rules_create_list = {IB_FLOW_SPEC_IB},
1371 .link_layer = IB_LINK_LAYER_INFINIBAND
1375 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1376 struct ib_flow_attr *flow_attr)
1380 const struct default_rules *pdefault_rules = default_table;
1381 u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1383 for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1384 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1385 memset(&field_types, 0, sizeof(field_types));
1387 if (link_layer != pdefault_rules->link_layer)
1390 ib_flow = flow_attr + 1;
1391 /* we assume the specs are sorted */
1392 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1393 j < flow_attr->num_of_specs; k++) {
1394 union ib_flow_spec *current_flow =
1395 (union ib_flow_spec *)ib_flow;
1397 /* same layer but different type */
1398 if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1399 (pdefault_rules->mandatory_fields[k] &
1400 IB_FLOW_SPEC_LAYER_MASK)) &&
1401 (current_flow->type !=
1402 pdefault_rules->mandatory_fields[k]))
1405 /* same layer, try match next one */
1406 if (current_flow->type ==
1407 pdefault_rules->mandatory_fields[k]) {
1410 ((union ib_flow_spec *)ib_flow)->size;
1414 ib_flow = flow_attr + 1;
1415 for (j = 0; j < flow_attr->num_of_specs;
1416 j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1417 for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1418 /* same layer and same type */
1419 if (((union ib_flow_spec *)ib_flow)->type ==
1420 pdefault_rules->mandatory_not_fields[k])
1429 static int __mlx4_ib_create_default_rules(
1430 struct mlx4_ib_dev *mdev,
1432 const struct default_rules *pdefault_rules,
1433 struct _rule_hw *mlx4_spec) {
1437 for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1438 union ib_flow_spec ib_spec = {};
1441 switch (pdefault_rules->rules_create_list[i]) {
1445 case IB_FLOW_SPEC_IB:
1446 ib_spec.type = IB_FLOW_SPEC_IB;
1447 ib_spec.size = sizeof(struct ib_flow_spec_ib);
1454 /* We must put empty rule, qpn is being ignored */
1455 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1458 pr_info("invalid parsing\n");
1462 mlx4_spec = (void *)mlx4_spec + ret;
1468 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1470 enum mlx4_net_trans_promisc_mode flow_type,
1476 struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1477 struct mlx4_cmd_mailbox *mailbox;
1478 struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1481 static const u16 __mlx4_domain[] = {
1482 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1483 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1484 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1485 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1488 if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1489 pr_err("Invalid priority value %d\n", flow_attr->priority);
1493 if (domain >= IB_FLOW_DOMAIN_NUM) {
1494 pr_err("Invalid domain value %d\n", domain);
1498 if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1501 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1502 if (IS_ERR(mailbox))
1503 return PTR_ERR(mailbox);
1504 ctrl = mailbox->buf;
1506 ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1507 flow_attr->priority);
1508 ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1509 ctrl->port = flow_attr->port;
1510 ctrl->qpn = cpu_to_be32(qp->qp_num);
1512 ib_flow = flow_attr + 1;
1513 size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1514 /* Add default flows */
1515 default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1516 if (default_flow >= 0) {
1517 ret = __mlx4_ib_create_default_rules(
1518 mdev, qp, default_table + default_flow,
1519 mailbox->buf + size);
1521 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1526 for (i = 0; i < flow_attr->num_of_specs; i++) {
1527 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1528 mailbox->buf + size);
1530 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1533 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1537 ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1538 MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1541 pr_err("mcg table is full. Fail to register network rule.\n");
1542 else if (ret == -ENXIO)
1543 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1545 pr_err("Invalid argumant. Fail to register network rule.\n");
1547 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1551 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1554 err = mlx4_cmd(dev, reg_id, 0, 0,
1555 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1558 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1563 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1567 union ib_flow_spec *ib_spec;
1568 struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1571 if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1572 dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1573 return 0; /* do nothing */
1575 ib_flow = flow_attr + 1;
1576 ib_spec = (union ib_flow_spec *)ib_flow;
1578 if (ib_spec->type != IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1579 return 0; /* do nothing */
1581 err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1582 flow_attr->port, qp->qp_num,
1583 MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1588 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1589 struct ib_flow_attr *flow_attr,
1592 int err = 0, i = 0, j = 0;
1593 struct mlx4_ib_flow *mflow;
1594 enum mlx4_net_trans_promisc_mode type[2];
1595 struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1596 int is_bonded = mlx4_is_bonded(dev);
1598 memset(type, 0, sizeof(type));
1600 mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1606 switch (flow_attr->type) {
1607 case IB_FLOW_ATTR_NORMAL:
1608 type[0] = MLX4_FS_REGULAR;
1611 case IB_FLOW_ATTR_ALL_DEFAULT:
1612 type[0] = MLX4_FS_ALL_DEFAULT;
1615 case IB_FLOW_ATTR_MC_DEFAULT:
1616 type[0] = MLX4_FS_MC_DEFAULT;
1619 case IB_FLOW_ATTR_SNIFFER:
1620 type[0] = MLX4_FS_UC_SNIFFER;
1621 type[1] = MLX4_FS_MC_SNIFFER;
1629 while (i < ARRAY_SIZE(type) && type[i]) {
1630 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1631 &mflow->reg_id[i].id);
1633 goto err_create_flow;
1635 /* Application always sees one port so the mirror rule
1636 * must be on port #2
1638 flow_attr->port = 2;
1639 err = __mlx4_ib_create_flow(qp, flow_attr,
1641 &mflow->reg_id[j].mirror);
1642 flow_attr->port = 1;
1644 goto err_create_flow;
1651 if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1652 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1653 &mflow->reg_id[i].id);
1655 goto err_create_flow;
1658 flow_attr->port = 2;
1659 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1660 &mflow->reg_id[j].mirror);
1661 flow_attr->port = 1;
1663 goto err_create_flow;
1666 /* function to create mirror rule */
1670 return &mflow->ibflow;
1674 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1675 mflow->reg_id[i].id);
1680 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1681 mflow->reg_id[j].mirror);
1686 return ERR_PTR(err);
1689 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1693 struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1694 struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1696 while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1697 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1700 if (mflow->reg_id[i].mirror) {
1701 err = __mlx4_ib_destroy_flow(mdev->dev,
1702 mflow->reg_id[i].mirror);
1713 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1716 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1717 struct mlx4_dev *dev = mdev->dev;
1718 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1719 struct mlx4_ib_steering *ib_steering = NULL;
1720 enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1721 struct mlx4_flow_reg_id reg_id;
1723 if (mdev->dev->caps.steering_mode ==
1724 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1725 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1730 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1732 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1735 pr_err("multicast attach op failed, err %d\n", err);
1740 if (mlx4_is_bonded(dev)) {
1741 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1742 (mqp->port == 1) ? 2 : 1,
1744 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1745 prot, ®_id.mirror);
1750 err = add_gid_entry(ibqp, gid);
1755 memcpy(ib_steering->gid.raw, gid->raw, 16);
1756 ib_steering->reg_id = reg_id;
1757 mutex_lock(&mqp->mutex);
1758 list_add(&ib_steering->list, &mqp->steering_rules);
1759 mutex_unlock(&mqp->mutex);
1764 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1767 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1768 prot, reg_id.mirror);
1775 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1777 struct mlx4_ib_gid_entry *ge;
1778 struct mlx4_ib_gid_entry *tmp;
1779 struct mlx4_ib_gid_entry *ret = NULL;
1781 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1782 if (!memcmp(raw, ge->gid.raw, 16)) {
1791 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1794 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1795 struct mlx4_dev *dev = mdev->dev;
1796 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1797 struct net_device *ndev;
1798 struct mlx4_ib_gid_entry *ge;
1799 struct mlx4_flow_reg_id reg_id = {0, 0};
1800 enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1802 if (mdev->dev->caps.steering_mode ==
1803 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1804 struct mlx4_ib_steering *ib_steering;
1806 mutex_lock(&mqp->mutex);
1807 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1808 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1809 list_del(&ib_steering->list);
1813 mutex_unlock(&mqp->mutex);
1814 if (&ib_steering->list == &mqp->steering_rules) {
1815 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1818 reg_id = ib_steering->reg_id;
1822 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1827 if (mlx4_is_bonded(dev)) {
1828 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1829 prot, reg_id.mirror);
1834 mutex_lock(&mqp->mutex);
1835 ge = find_gid_entry(mqp, gid->raw);
1837 spin_lock_bh(&mdev->iboe.lock);
1838 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1841 spin_unlock_bh(&mdev->iboe.lock);
1844 list_del(&ge->list);
1847 pr_warn("could not find mgid entry\n");
1849 mutex_unlock(&mqp->mutex);
1854 static int init_node_data(struct mlx4_ib_dev *dev)
1856 struct ib_smp *in_mad = NULL;
1857 struct ib_smp *out_mad = NULL;
1858 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1861 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1862 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1863 if (!in_mad || !out_mad)
1866 init_query_mad(in_mad);
1867 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1868 if (mlx4_is_master(dev->dev))
1869 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1871 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1875 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1877 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1879 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1883 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1884 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1892 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1895 struct mlx4_ib_dev *dev =
1896 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1897 return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1900 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1903 struct mlx4_ib_dev *dev =
1904 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1905 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1906 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1907 (int) dev->dev->caps.fw_ver & 0xffff);
1910 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1913 struct mlx4_ib_dev *dev =
1914 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1915 return sprintf(buf, "%x\n", dev->dev->rev_id);
1918 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1921 struct mlx4_ib_dev *dev =
1922 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1923 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1924 dev->dev->board_id);
1927 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1928 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1929 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1930 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1932 static struct device_attribute *mlx4_class_attributes[] = {
1939 #define MLX4_IB_INVALID_MAC ((u64)-1)
1940 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1941 struct net_device *dev,
1945 u64 release_mac = MLX4_IB_INVALID_MAC;
1946 struct mlx4_ib_qp *qp;
1948 read_lock(&dev_base_lock);
1949 new_smac = mlx4_mac_to_u64(dev->dev_addr);
1950 read_unlock(&dev_base_lock);
1952 atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1954 /* no need for update QP1 and mac registration in non-SRIOV */
1955 if (!mlx4_is_mfunc(ibdev->dev))
1958 mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1959 qp = ibdev->qp1_proxy[port - 1];
1963 struct mlx4_update_qp_params update_params;
1965 mutex_lock(&qp->mutex);
1966 old_smac = qp->pri.smac;
1967 if (new_smac == old_smac)
1970 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1972 if (new_smac_index < 0)
1975 update_params.smac_index = new_smac_index;
1976 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1978 release_mac = new_smac;
1981 /* if old port was zero, no mac was yet registered for this QP */
1982 if (qp->pri.smac_port)
1983 release_mac = old_smac;
1984 qp->pri.smac = new_smac;
1985 qp->pri.smac_port = port;
1986 qp->pri.smac_index = new_smac_index;
1990 if (release_mac != MLX4_IB_INVALID_MAC)
1991 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1993 mutex_unlock(&qp->mutex);
1994 mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1997 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1998 struct net_device *dev,
1999 unsigned long event)
2002 struct mlx4_ib_iboe *iboe;
2003 int update_qps_port = -1;
2008 iboe = &ibdev->iboe;
2010 spin_lock_bh(&iboe->lock);
2011 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2013 iboe->netdevs[port - 1] =
2014 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2016 if (dev == iboe->netdevs[port - 1] &&
2017 (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2018 event == NETDEV_UP || event == NETDEV_CHANGE))
2019 update_qps_port = port;
2022 spin_unlock_bh(&iboe->lock);
2024 if (update_qps_port > 0)
2025 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2028 static int mlx4_ib_netdev_event(struct notifier_block *this,
2029 unsigned long event, void *ptr)
2031 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2032 struct mlx4_ib_dev *ibdev;
2034 if (!net_eq(dev_net(dev), &init_net))
2037 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2038 mlx4_ib_scan_netdevs(ibdev, dev, event);
2043 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2049 if (mlx4_is_master(ibdev->dev)) {
2050 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2052 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2054 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2056 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2057 /* master has the identity virt2phys pkey mapping */
2058 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2059 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2060 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2061 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2065 /* initialize pkey cache */
2066 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2068 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2070 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2076 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2078 int i, j, eq = 0, total_eqs = 0;
2080 ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2081 sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2082 if (!ibdev->eq_table)
2085 for (i = 1; i <= dev->caps.num_ports; i++) {
2086 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2088 if (i > 1 && mlx4_is_eq_shared(dev, total_eqs))
2090 ibdev->eq_table[eq] = total_eqs;
2091 if (!mlx4_assign_eq(dev, i,
2092 &ibdev->eq_table[eq]))
2095 ibdev->eq_table[eq] = -1;
2099 for (i = eq; i < dev->caps.num_comp_vectors;
2100 ibdev->eq_table[i++] = -1)
2103 /* Advertise the new number of EQs to clients */
2104 ibdev->ib_dev.num_comp_vectors = eq;
2107 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2110 int total_eqs = ibdev->ib_dev.num_comp_vectors;
2112 /* no eqs were allocated */
2113 if (!ibdev->eq_table)
2116 /* Reset the advertised EQ number */
2117 ibdev->ib_dev.num_comp_vectors = 0;
2119 for (i = 0; i < total_eqs; i++)
2120 mlx4_release_eq(dev, ibdev->eq_table[i]);
2122 kfree(ibdev->eq_table);
2123 ibdev->eq_table = NULL;
2126 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2127 struct ib_port_immutable *immutable)
2129 struct ib_port_attr attr;
2132 err = mlx4_ib_query_port(ibdev, port_num, &attr);
2136 immutable->pkey_tbl_len = attr.pkey_tbl_len;
2137 immutable->gid_tbl_len = attr.gid_tbl_len;
2139 if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2140 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2142 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2144 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2149 static void *mlx4_ib_add(struct mlx4_dev *dev)
2151 struct mlx4_ib_dev *ibdev;
2155 struct mlx4_ib_iboe *iboe;
2156 int ib_num_ports = 0;
2157 int num_req_counters;
2160 struct counter_index *new_counter_index = NULL;
2162 pr_info_once("%s", mlx4_ib_version);
2165 mlx4_foreach_ib_transport_port(i, dev)
2168 /* No point in registering a device with no ports... */
2172 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2174 dev_err(&dev->persist->pdev->dev,
2175 "Device struct alloc failed\n");
2179 iboe = &ibdev->iboe;
2181 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2184 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2187 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2189 if (!ibdev->uar_map)
2191 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2194 ibdev->bond_next_port = 0;
2196 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2197 ibdev->ib_dev.owner = THIS_MODULE;
2198 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
2199 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
2200 ibdev->num_ports = num_ports;
2201 ibdev->ib_dev.phys_port_cnt = mlx4_is_bonded(dev) ?
2202 1 : ibdev->num_ports;
2203 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2204 ibdev->ib_dev.dma_device = &dev->persist->pdev->dev;
2205 ibdev->ib_dev.get_netdev = mlx4_ib_get_netdev;
2206 ibdev->ib_dev.add_gid = mlx4_ib_add_gid;
2207 ibdev->ib_dev.del_gid = mlx4_ib_del_gid;
2209 if (dev->caps.userspace_caps)
2210 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2212 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2214 ibdev->ib_dev.uverbs_cmd_mask =
2215 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
2216 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
2217 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
2218 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
2219 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
2220 (1ull << IB_USER_VERBS_CMD_REG_MR) |
2221 (1ull << IB_USER_VERBS_CMD_REREG_MR) |
2222 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
2223 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2224 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
2225 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
2226 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
2227 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
2228 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
2229 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
2230 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
2231 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
2232 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
2233 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
2234 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
2235 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
2236 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
2237 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
2238 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2240 ibdev->ib_dev.query_device = mlx4_ib_query_device;
2241 ibdev->ib_dev.query_port = mlx4_ib_query_port;
2242 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
2243 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
2244 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
2245 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
2246 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
2247 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
2248 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
2249 ibdev->ib_dev.mmap = mlx4_ib_mmap;
2250 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
2251 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
2252 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
2253 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
2254 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
2255 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
2256 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
2257 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
2258 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
2259 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
2260 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
2261 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
2262 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
2263 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
2264 ibdev->ib_dev.post_send = mlx4_ib_post_send;
2265 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
2266 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
2267 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
2268 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
2269 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
2270 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
2271 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
2272 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
2273 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
2274 ibdev->ib_dev.rereg_user_mr = mlx4_ib_rereg_user_mr;
2275 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
2276 ibdev->ib_dev.alloc_mr = mlx4_ib_alloc_mr;
2277 ibdev->ib_dev.map_mr_sg = mlx4_ib_map_mr_sg;
2278 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
2279 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
2280 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
2281 ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2282 ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2284 if (!mlx4_is_slave(ibdev->dev)) {
2285 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
2286 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
2287 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
2288 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
2291 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2292 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2293 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2294 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2295 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2297 ibdev->ib_dev.uverbs_cmd_mask |=
2298 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2299 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2302 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2303 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2304 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2305 ibdev->ib_dev.uverbs_cmd_mask |=
2306 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2307 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2310 if (check_flow_steering_support(dev)) {
2311 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2312 ibdev->ib_dev.create_flow = mlx4_ib_create_flow;
2313 ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow;
2315 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2316 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2317 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2320 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2321 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2322 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2323 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2325 mlx4_ib_alloc_eqs(dev, ibdev);
2327 spin_lock_init(&iboe->lock);
2329 if (init_node_data(ibdev))
2332 for (i = 0; i < ibdev->num_ports; ++i) {
2333 mutex_init(&ibdev->counters_table[i].mutex);
2334 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2337 num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2338 for (i = 0; i < num_req_counters; ++i) {
2339 mutex_init(&ibdev->qp1_proxy_lock[i]);
2341 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2342 IB_LINK_LAYER_ETHERNET) {
2343 err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2344 /* if failed to allocate a new counter, use default */
2347 mlx4_get_default_counter_index(dev,
2351 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2352 counter_index = mlx4_get_default_counter_index(dev,
2355 new_counter_index = kmalloc(sizeof(*new_counter_index),
2357 if (!new_counter_index) {
2359 mlx4_counter_free(ibdev->dev, counter_index);
2362 new_counter_index->index = counter_index;
2363 new_counter_index->allocated = allocated;
2364 list_add_tail(&new_counter_index->list,
2365 &ibdev->counters_table[i].counters_list);
2366 ibdev->counters_table[i].default_counter = counter_index;
2367 pr_info("counter index %d for port %d allocated %d\n",
2368 counter_index, i + 1, allocated);
2370 if (mlx4_is_bonded(dev))
2371 for (i = 1; i < ibdev->num_ports ; ++i) {
2373 kmalloc(sizeof(struct counter_index),
2375 if (!new_counter_index)
2377 new_counter_index->index = counter_index;
2378 new_counter_index->allocated = 0;
2379 list_add_tail(&new_counter_index->list,
2380 &ibdev->counters_table[i].counters_list);
2381 ibdev->counters_table[i].default_counter =
2385 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2388 spin_lock_init(&ibdev->sm_lock);
2389 mutex_init(&ibdev->cap_mask_mutex);
2390 INIT_LIST_HEAD(&ibdev->qp_list);
2391 spin_lock_init(&ibdev->reset_flow_resource_lock);
2393 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2395 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2396 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2397 MLX4_IB_UC_STEER_QPN_ALIGN,
2398 &ibdev->steer_qpn_base, 0);
2402 ibdev->ib_uc_qpns_bitmap =
2403 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2406 if (!ibdev->ib_uc_qpns_bitmap) {
2407 dev_err(&dev->persist->pdev->dev,
2408 "bit map alloc failed\n");
2409 goto err_steer_qp_release;
2412 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2413 bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2414 ibdev->steer_qpn_count);
2415 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2416 dev, ibdev->steer_qpn_base,
2417 ibdev->steer_qpn_base +
2418 ibdev->steer_qpn_count - 1);
2420 goto err_steer_free_bitmap;
2422 bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2423 ibdev->steer_qpn_count);
2427 for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2428 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2430 if (ib_register_device(&ibdev->ib_dev, NULL))
2431 goto err_steer_free_bitmap;
2433 if (mlx4_ib_mad_init(ibdev))
2436 if (mlx4_ib_init_sriov(ibdev))
2439 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2440 if (!iboe->nb.notifier_call) {
2441 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2442 err = register_netdevice_notifier(&iboe->nb);
2444 iboe->nb.notifier_call = NULL;
2450 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2451 if (device_create_file(&ibdev->ib_dev.dev,
2452 mlx4_class_attributes[j]))
2456 ibdev->ib_active = true;
2458 if (mlx4_is_mfunc(ibdev->dev))
2461 /* create paravirt contexts for any VFs which are active */
2462 if (mlx4_is_master(ibdev->dev)) {
2463 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2464 if (j == mlx4_master_func_num(ibdev->dev))
2466 if (mlx4_is_slave_active(ibdev->dev, j))
2467 do_slave_init(ibdev, j, 1);
2473 if (ibdev->iboe.nb.notifier_call) {
2474 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2475 pr_warn("failure unregistering notifier\n");
2476 ibdev->iboe.nb.notifier_call = NULL;
2478 flush_workqueue(wq);
2480 mlx4_ib_close_sriov(ibdev);
2483 mlx4_ib_mad_cleanup(ibdev);
2486 ib_unregister_device(&ibdev->ib_dev);
2488 err_steer_free_bitmap:
2489 kfree(ibdev->ib_uc_qpns_bitmap);
2491 err_steer_qp_release:
2492 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2493 ibdev->steer_qpn_count);
2495 for (i = 0; i < ibdev->num_ports; ++i)
2496 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2499 mlx4_ib_free_eqs(dev, ibdev);
2500 iounmap(ibdev->uar_map);
2503 mlx4_uar_free(dev, &ibdev->priv_uar);
2506 mlx4_pd_free(dev, ibdev->priv_pdn);
2509 ib_dealloc_device(&ibdev->ib_dev);
2514 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2518 WARN_ON(!dev->ib_uc_qpns_bitmap);
2520 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2521 dev->steer_qpn_count,
2522 get_count_order(count));
2526 *qpn = dev->steer_qpn_base + offset;
2530 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2533 dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2536 BUG_ON(qpn < dev->steer_qpn_base);
2538 bitmap_release_region(dev->ib_uc_qpns_bitmap,
2539 qpn - dev->steer_qpn_base,
2540 get_count_order(count));
2543 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2548 struct ib_flow_attr *flow = NULL;
2549 struct ib_flow_spec_ib *ib_spec;
2552 flow_size = sizeof(struct ib_flow_attr) +
2553 sizeof(struct ib_flow_spec_ib);
2554 flow = kzalloc(flow_size, GFP_KERNEL);
2557 flow->port = mqp->port;
2558 flow->num_of_specs = 1;
2559 flow->size = flow_size;
2560 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2561 ib_spec->type = IB_FLOW_SPEC_IB;
2562 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2563 /* Add an empty rule for IB L2 */
2564 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2566 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2571 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2577 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2579 struct mlx4_ib_dev *ibdev = ibdev_ptr;
2582 ibdev->ib_active = false;
2583 flush_workqueue(wq);
2585 mlx4_ib_close_sriov(ibdev);
2586 mlx4_ib_mad_cleanup(ibdev);
2587 ib_unregister_device(&ibdev->ib_dev);
2588 if (ibdev->iboe.nb.notifier_call) {
2589 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2590 pr_warn("failure unregistering notifier\n");
2591 ibdev->iboe.nb.notifier_call = NULL;
2594 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2595 ibdev->steer_qpn_count);
2596 kfree(ibdev->ib_uc_qpns_bitmap);
2598 iounmap(ibdev->uar_map);
2599 for (p = 0; p < ibdev->num_ports; ++p)
2600 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2602 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2603 mlx4_CLOSE_PORT(dev, p);
2605 mlx4_ib_free_eqs(dev, ibdev);
2607 mlx4_uar_free(dev, &ibdev->priv_uar);
2608 mlx4_pd_free(dev, ibdev->priv_pdn);
2609 ib_dealloc_device(&ibdev->ib_dev);
2612 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2614 struct mlx4_ib_demux_work **dm = NULL;
2615 struct mlx4_dev *dev = ibdev->dev;
2617 unsigned long flags;
2618 struct mlx4_active_ports actv_ports;
2620 unsigned int first_port;
2622 if (!mlx4_is_master(dev))
2625 actv_ports = mlx4_get_active_ports(dev, slave);
2626 ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2627 first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2629 dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2631 pr_err("failed to allocate memory for tunneling qp update\n");
2635 for (i = 0; i < ports; i++) {
2636 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2638 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2643 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2644 dm[i]->port = first_port + i + 1;
2645 dm[i]->slave = slave;
2646 dm[i]->do_init = do_init;
2649 /* initialize or tear down tunnel QPs for the slave */
2650 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2651 if (!ibdev->sriov.is_going_down) {
2652 for (i = 0; i < ports; i++)
2653 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2654 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2656 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2657 for (i = 0; i < ports; i++)
2665 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2667 struct mlx4_ib_qp *mqp;
2668 unsigned long flags_qp;
2669 unsigned long flags_cq;
2670 struct mlx4_ib_cq *send_mcq, *recv_mcq;
2671 struct list_head cq_notify_list;
2672 struct mlx4_cq *mcq;
2673 unsigned long flags;
2675 pr_warn("mlx4_ib_handle_catas_error was started\n");
2676 INIT_LIST_HEAD(&cq_notify_list);
2678 /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2679 spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2681 list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2682 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2683 if (mqp->sq.tail != mqp->sq.head) {
2684 send_mcq = to_mcq(mqp->ibqp.send_cq);
2685 spin_lock_irqsave(&send_mcq->lock, flags_cq);
2686 if (send_mcq->mcq.comp &&
2687 mqp->ibqp.send_cq->comp_handler) {
2688 if (!send_mcq->mcq.reset_notify_added) {
2689 send_mcq->mcq.reset_notify_added = 1;
2690 list_add_tail(&send_mcq->mcq.reset_notify,
2694 spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2696 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2697 /* Now, handle the QP's receive queue */
2698 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2699 /* no handling is needed for SRQ */
2700 if (!mqp->ibqp.srq) {
2701 if (mqp->rq.tail != mqp->rq.head) {
2702 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2703 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2704 if (recv_mcq->mcq.comp &&
2705 mqp->ibqp.recv_cq->comp_handler) {
2706 if (!recv_mcq->mcq.reset_notify_added) {
2707 recv_mcq->mcq.reset_notify_added = 1;
2708 list_add_tail(&recv_mcq->mcq.reset_notify,
2712 spin_unlock_irqrestore(&recv_mcq->lock,
2716 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2719 list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2722 spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2723 pr_warn("mlx4_ib_handle_catas_error ended\n");
2726 static void handle_bonded_port_state_event(struct work_struct *work)
2728 struct ib_event_work *ew =
2729 container_of(work, struct ib_event_work, work);
2730 struct mlx4_ib_dev *ibdev = ew->ib_dev;
2731 enum ib_port_state bonded_port_state = IB_PORT_NOP;
2733 struct ib_event ibev;
2736 spin_lock_bh(&ibdev->iboe.lock);
2737 for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2738 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2739 enum ib_port_state curr_port_state;
2745 (netif_running(curr_netdev) &&
2746 netif_carrier_ok(curr_netdev)) ?
2747 IB_PORT_ACTIVE : IB_PORT_DOWN;
2749 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2750 curr_port_state : IB_PORT_ACTIVE;
2752 spin_unlock_bh(&ibdev->iboe.lock);
2754 ibev.device = &ibdev->ib_dev;
2755 ibev.element.port_num = 1;
2756 ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2757 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2759 ib_dispatch_event(&ibev);
2762 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2763 enum mlx4_dev_event event, unsigned long param)
2765 struct ib_event ibev;
2766 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2767 struct mlx4_eqe *eqe = NULL;
2768 struct ib_event_work *ew;
2771 if (mlx4_is_bonded(dev) &&
2772 ((event == MLX4_DEV_EVENT_PORT_UP) ||
2773 (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2774 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2777 INIT_WORK(&ew->work, handle_bonded_port_state_event);
2779 queue_work(wq, &ew->work);
2783 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2784 eqe = (struct mlx4_eqe *)param;
2789 case MLX4_DEV_EVENT_PORT_UP:
2790 if (p > ibdev->num_ports)
2792 if (mlx4_is_master(dev) &&
2793 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2794 IB_LINK_LAYER_INFINIBAND) {
2795 mlx4_ib_invalidate_all_guid_record(ibdev, p);
2797 ibev.event = IB_EVENT_PORT_ACTIVE;
2800 case MLX4_DEV_EVENT_PORT_DOWN:
2801 if (p > ibdev->num_ports)
2803 ibev.event = IB_EVENT_PORT_ERR;
2806 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2807 ibdev->ib_active = false;
2808 ibev.event = IB_EVENT_DEVICE_FATAL;
2809 mlx4_ib_handle_catas_error(ibdev);
2812 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2813 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2815 pr_err("failed to allocate memory for events work\n");
2819 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2820 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2822 /* need to queue only for port owner, which uses GEN_EQE */
2823 if (mlx4_is_master(dev))
2824 queue_work(wq, &ew->work);
2826 handle_port_mgmt_change_event(&ew->work);
2829 case MLX4_DEV_EVENT_SLAVE_INIT:
2830 /* here, p is the slave id */
2831 do_slave_init(ibdev, p, 1);
2832 if (mlx4_is_master(dev)) {
2835 for (i = 1; i <= ibdev->num_ports; i++) {
2836 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2837 == IB_LINK_LAYER_INFINIBAND)
2838 mlx4_ib_slave_alias_guid_event(ibdev,
2845 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2846 if (mlx4_is_master(dev)) {
2849 for (i = 1; i <= ibdev->num_ports; i++) {
2850 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2851 == IB_LINK_LAYER_INFINIBAND)
2852 mlx4_ib_slave_alias_guid_event(ibdev,
2857 /* here, p is the slave id */
2858 do_slave_init(ibdev, p, 0);
2865 ibev.device = ibdev_ptr;
2866 ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2868 ib_dispatch_event(&ibev);
2871 static struct mlx4_interface mlx4_ib_interface = {
2873 .remove = mlx4_ib_remove,
2874 .event = mlx4_ib_event,
2875 .protocol = MLX4_PROT_IB_IPV6,
2876 .flags = MLX4_INTFF_BONDING
2879 static int __init mlx4_ib_init(void)
2883 wq = create_singlethread_workqueue("mlx4_ib");
2887 err = mlx4_ib_mcg_init();
2891 err = mlx4_register_interface(&mlx4_ib_interface);
2898 mlx4_ib_mcg_destroy();
2901 destroy_workqueue(wq);
2905 static void __exit mlx4_ib_cleanup(void)
2907 mlx4_unregister_interface(&mlx4_ib_interface);
2908 mlx4_ib_mcg_destroy();
2909 destroy_workqueue(wq);
2912 module_init(mlx4_ib_init);
2913 module_exit(mlx4_ib_cleanup);