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>
49 #include <linux/mlx4/driver.h>
50 #include <linux/mlx4/cmd.h>
51 #include <linux/mlx4/qp.h>
56 #define DRV_NAME MLX4_IB_DRV_NAME
57 #define DRV_VERSION "2.2-1"
58 #define DRV_RELDATE "Feb 2014"
60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
62 #define MLX4_IB_CARD_REV_A0 0xA0
64 MODULE_AUTHOR("Roland Dreier");
65 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
66 MODULE_LICENSE("Dual BSD/GPL");
67 MODULE_VERSION(DRV_VERSION);
69 int mlx4_ib_sm_guid_assign = 0;
70 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
71 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
73 static const char mlx4_ib_version[] =
74 DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
75 DRV_VERSION " (" DRV_RELDATE ")\n";
77 struct update_gid_work {
78 struct work_struct work;
79 union ib_gid gids[128];
80 struct mlx4_ib_dev *dev;
84 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
86 static struct workqueue_struct *wq;
88 static void init_query_mad(struct ib_smp *mad)
90 mad->base_version = 1;
91 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
92 mad->class_version = 1;
93 mad->method = IB_MGMT_METHOD_GET;
96 static union ib_gid zgid;
98 static int check_flow_steering_support(struct mlx4_dev *dev)
100 int eth_num_ports = 0;
101 int ib_num_ports = 0;
103 int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
107 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
109 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
111 dmfs &= (!ib_num_ports ||
112 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
114 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
115 if (ib_num_ports && mlx4_is_mfunc(dev)) {
116 pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
123 static int num_ib_ports(struct mlx4_dev *dev)
128 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
134 static int mlx4_ib_query_device(struct ib_device *ibdev,
135 struct ib_device_attr *props,
136 struct ib_udata *uhw)
138 struct mlx4_ib_dev *dev = to_mdev(ibdev);
139 struct ib_smp *in_mad = NULL;
140 struct ib_smp *out_mad = NULL;
143 struct mlx4_uverbs_ex_query_device cmd;
144 struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
145 struct mlx4_clock_params clock_params;
148 if (uhw->inlen < sizeof(cmd))
151 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
162 resp.response_length = offsetof(typeof(resp), response_length) +
163 sizeof(resp.response_length);
164 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
165 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
166 if (!in_mad || !out_mad)
169 init_query_mad(in_mad);
170 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
172 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
173 1, NULL, NULL, in_mad, out_mad);
177 memset(props, 0, sizeof *props);
179 have_ib_ports = num_ib_ports(dev->dev);
181 props->fw_ver = dev->dev->caps.fw_ver;
182 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
183 IB_DEVICE_PORT_ACTIVE_EVENT |
184 IB_DEVICE_SYS_IMAGE_GUID |
185 IB_DEVICE_RC_RNR_NAK_GEN |
186 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
187 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
188 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
189 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
190 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
191 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
192 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
193 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
194 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
195 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
196 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
197 if (dev->dev->caps.max_gso_sz &&
198 (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
199 (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
200 props->device_cap_flags |= IB_DEVICE_UD_TSO;
201 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
202 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
203 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
204 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
205 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
206 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
207 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
208 props->device_cap_flags |= IB_DEVICE_XRC;
209 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
210 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
211 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
212 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
213 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
215 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
216 if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
217 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
220 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
222 props->vendor_part_id = dev->dev->persist->pdev->device;
223 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
224 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
226 props->max_mr_size = ~0ull;
227 props->page_size_cap = dev->dev->caps.page_size_cap;
228 props->max_qp = dev->dev->quotas.qp;
229 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
230 props->max_sge = min(dev->dev->caps.max_sq_sg,
231 dev->dev->caps.max_rq_sg);
232 props->max_cq = dev->dev->quotas.cq;
233 props->max_cqe = dev->dev->caps.max_cqes;
234 props->max_mr = dev->dev->quotas.mpt;
235 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
236 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
237 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
238 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
239 props->max_srq = dev->dev->quotas.srq;
240 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
241 props->max_srq_sge = dev->dev->caps.max_srq_sge;
242 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
243 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
244 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
245 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
246 props->masked_atomic_cap = props->atomic_cap;
247 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
248 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
249 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
250 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
251 props->max_mcast_grp;
252 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
253 props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
254 props->timestamp_mask = 0xFFFFFFFFFFFFULL;
256 if (!mlx4_is_slave(dev->dev))
257 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
259 if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
260 resp.response_length += sizeof(resp.hca_core_clock_offset);
261 if (!err && !mlx4_is_slave(dev->dev)) {
262 resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
263 resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
268 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
279 static enum rdma_link_layer
280 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
282 struct mlx4_dev *dev = to_mdev(device)->dev;
284 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
285 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
288 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
289 struct ib_port_attr *props, int netw_view)
291 struct ib_smp *in_mad = NULL;
292 struct ib_smp *out_mad = NULL;
293 int ext_active_speed;
294 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
297 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
298 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
299 if (!in_mad || !out_mad)
302 init_query_mad(in_mad);
303 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
304 in_mad->attr_mod = cpu_to_be32(port);
306 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
307 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
309 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
315 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
316 props->lmc = out_mad->data[34] & 0x7;
317 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
318 props->sm_sl = out_mad->data[36] & 0xf;
319 props->state = out_mad->data[32] & 0xf;
320 props->phys_state = out_mad->data[33] >> 4;
321 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
323 props->gid_tbl_len = out_mad->data[50];
325 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
326 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
327 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
328 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
329 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
330 props->active_width = out_mad->data[31] & 0xf;
331 props->active_speed = out_mad->data[35] >> 4;
332 props->max_mtu = out_mad->data[41] & 0xf;
333 props->active_mtu = out_mad->data[36] >> 4;
334 props->subnet_timeout = out_mad->data[51] & 0x1f;
335 props->max_vl_num = out_mad->data[37] >> 4;
336 props->init_type_reply = out_mad->data[41] >> 4;
338 /* Check if extended speeds (EDR/FDR/...) are supported */
339 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
340 ext_active_speed = out_mad->data[62] >> 4;
342 switch (ext_active_speed) {
344 props->active_speed = IB_SPEED_FDR;
347 props->active_speed = IB_SPEED_EDR;
352 /* If reported active speed is QDR, check if is FDR-10 */
353 if (props->active_speed == IB_SPEED_QDR) {
354 init_query_mad(in_mad);
355 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
356 in_mad->attr_mod = cpu_to_be32(port);
358 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
359 NULL, NULL, in_mad, out_mad);
363 /* Checking LinkSpeedActive for FDR-10 */
364 if (out_mad->data[15] & 0x1)
365 props->active_speed = IB_SPEED_FDR10;
368 /* Avoid wrong speed value returned by FW if the IB link is down. */
369 if (props->state == IB_PORT_DOWN)
370 props->active_speed = IB_SPEED_SDR;
378 static u8 state_to_phys_state(enum ib_port_state state)
380 return state == IB_PORT_ACTIVE ? 5 : 3;
383 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
384 struct ib_port_attr *props, int netw_view)
387 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
388 struct mlx4_ib_iboe *iboe = &mdev->iboe;
389 struct net_device *ndev;
391 struct mlx4_cmd_mailbox *mailbox;
393 int is_bonded = mlx4_is_bonded(mdev->dev);
395 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
397 return PTR_ERR(mailbox);
399 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
400 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
405 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
406 IB_WIDTH_4X : IB_WIDTH_1X;
407 props->active_speed = IB_SPEED_QDR;
408 props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
409 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
410 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
411 props->pkey_tbl_len = 1;
412 props->max_mtu = IB_MTU_4096;
413 props->max_vl_num = 2;
414 props->state = IB_PORT_DOWN;
415 props->phys_state = state_to_phys_state(props->state);
416 props->active_mtu = IB_MTU_256;
418 rtnl_lock(); /* required to get upper dev */
419 spin_lock_bh(&iboe->lock);
420 ndev = iboe->netdevs[port - 1];
421 if (ndev && is_bonded)
422 ndev = netdev_master_upper_dev_get(ndev);
426 tmp = iboe_get_mtu(ndev->mtu);
427 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
429 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
430 IB_PORT_ACTIVE : IB_PORT_DOWN;
431 props->phys_state = state_to_phys_state(props->state);
433 spin_unlock_bh(&iboe->lock);
437 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
441 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
442 struct ib_port_attr *props, int netw_view)
446 memset(props, 0, sizeof *props);
448 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
449 ib_link_query_port(ibdev, port, props, netw_view) :
450 eth_link_query_port(ibdev, port, props, netw_view);
455 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
456 struct ib_port_attr *props)
458 /* returns host view */
459 return __mlx4_ib_query_port(ibdev, port, props, 0);
462 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
463 union ib_gid *gid, int netw_view)
465 struct ib_smp *in_mad = NULL;
466 struct ib_smp *out_mad = NULL;
468 struct mlx4_ib_dev *dev = to_mdev(ibdev);
470 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
472 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
473 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
474 if (!in_mad || !out_mad)
477 init_query_mad(in_mad);
478 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
479 in_mad->attr_mod = cpu_to_be32(port);
481 if (mlx4_is_mfunc(dev->dev) && netw_view)
482 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
484 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
488 memcpy(gid->raw, out_mad->data + 8, 8);
490 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
492 /* For any index > 0, return the null guid */
499 init_query_mad(in_mad);
500 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
501 in_mad->attr_mod = cpu_to_be32(index / 8);
503 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
504 NULL, NULL, in_mad, out_mad);
508 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
512 memset(gid->raw + 8, 0, 8);
518 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
521 struct mlx4_ib_dev *dev = to_mdev(ibdev);
523 *gid = dev->iboe.gid_table[port - 1][index];
528 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
531 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
532 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
534 return iboe_query_gid(ibdev, port, index, gid);
537 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
538 u16 *pkey, int netw_view)
540 struct ib_smp *in_mad = NULL;
541 struct ib_smp *out_mad = NULL;
542 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
545 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
546 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
547 if (!in_mad || !out_mad)
550 init_query_mad(in_mad);
551 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
552 in_mad->attr_mod = cpu_to_be32(index / 32);
554 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
555 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
557 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
562 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
570 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
572 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
575 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
576 struct ib_device_modify *props)
578 struct mlx4_cmd_mailbox *mailbox;
581 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
584 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
587 if (mlx4_is_slave(to_mdev(ibdev)->dev))
590 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
591 memcpy(ibdev->node_desc, props->node_desc, 64);
592 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
595 * If possible, pass node desc to FW, so it can generate
596 * a 144 trap. If cmd fails, just ignore.
598 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
602 memcpy(mailbox->buf, props->node_desc, 64);
603 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
604 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
606 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
611 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
614 struct mlx4_cmd_mailbox *mailbox;
617 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
619 return PTR_ERR(mailbox);
621 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
622 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
623 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
625 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
626 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
629 err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
630 MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
633 mlx4_free_cmd_mailbox(dev->dev, mailbox);
637 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
638 struct ib_port_modify *props)
640 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
641 u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
642 struct ib_port_attr attr;
646 /* return OK if this is RoCE. CM calls ib_modify_port() regardless
647 * of whether port link layer is ETH or IB. For ETH ports, qkey
648 * violations and port capabilities are not meaningful.
653 mutex_lock(&mdev->cap_mask_mutex);
655 err = mlx4_ib_query_port(ibdev, port, &attr);
659 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
660 ~props->clr_port_cap_mask;
662 err = mlx4_ib_SET_PORT(mdev, port,
663 !!(mask & IB_PORT_RESET_QKEY_CNTR),
667 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
671 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
672 struct ib_udata *udata)
674 struct mlx4_ib_dev *dev = to_mdev(ibdev);
675 struct mlx4_ib_ucontext *context;
676 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
677 struct mlx4_ib_alloc_ucontext_resp resp;
681 return ERR_PTR(-EAGAIN);
683 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
684 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
685 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
686 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
688 resp.dev_caps = dev->dev->caps.userspace_caps;
689 resp.qp_tab_size = dev->dev->caps.num_qps;
690 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
691 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
692 resp.cqe_size = dev->dev->caps.cqe_size;
695 context = kmalloc(sizeof *context, GFP_KERNEL);
697 return ERR_PTR(-ENOMEM);
699 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
705 INIT_LIST_HEAD(&context->db_page_list);
706 mutex_init(&context->db_page_mutex);
708 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
709 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
711 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
714 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
716 return ERR_PTR(-EFAULT);
719 return &context->ibucontext;
722 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
724 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
726 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
732 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
734 struct mlx4_ib_dev *dev = to_mdev(context->device);
736 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
739 if (vma->vm_pgoff == 0) {
740 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
742 if (io_remap_pfn_range(vma, vma->vm_start,
743 to_mucontext(context)->uar.pfn,
744 PAGE_SIZE, vma->vm_page_prot))
746 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
747 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
749 if (io_remap_pfn_range(vma, vma->vm_start,
750 to_mucontext(context)->uar.pfn +
751 dev->dev->caps.num_uars,
752 PAGE_SIZE, vma->vm_page_prot))
754 } else if (vma->vm_pgoff == 3) {
755 struct mlx4_clock_params params;
756 int ret = mlx4_get_internal_clock_params(dev->dev, ¶ms);
761 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
762 if (io_remap_pfn_range(vma, vma->vm_start,
763 (pci_resource_start(dev->dev->persist->pdev,
767 PAGE_SIZE, vma->vm_page_prot))
776 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
777 struct ib_ucontext *context,
778 struct ib_udata *udata)
780 struct mlx4_ib_pd *pd;
783 pd = kmalloc(sizeof *pd, GFP_KERNEL);
785 return ERR_PTR(-ENOMEM);
787 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
794 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
795 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
797 return ERR_PTR(-EFAULT);
803 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
805 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
811 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
812 struct ib_ucontext *context,
813 struct ib_udata *udata)
815 struct mlx4_ib_xrcd *xrcd;
816 struct ib_cq_init_attr cq_attr = {};
819 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
820 return ERR_PTR(-ENOSYS);
822 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
824 return ERR_PTR(-ENOMEM);
826 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
830 xrcd->pd = ib_alloc_pd(ibdev);
831 if (IS_ERR(xrcd->pd)) {
832 err = PTR_ERR(xrcd->pd);
837 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
838 if (IS_ERR(xrcd->cq)) {
839 err = PTR_ERR(xrcd->cq);
843 return &xrcd->ibxrcd;
846 ib_dealloc_pd(xrcd->pd);
848 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
854 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
856 ib_destroy_cq(to_mxrcd(xrcd)->cq);
857 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
858 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
864 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
866 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
867 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
868 struct mlx4_ib_gid_entry *ge;
870 ge = kzalloc(sizeof *ge, GFP_KERNEL);
875 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
876 ge->port = mqp->port;
880 mutex_lock(&mqp->mutex);
881 list_add_tail(&ge->list, &mqp->gid_list);
882 mutex_unlock(&mqp->mutex);
887 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
890 struct net_device *ndev;
896 spin_lock_bh(&mdev->iboe.lock);
897 ndev = mdev->iboe.netdevs[mqp->port - 1];
900 spin_unlock_bh(&mdev->iboe.lock);
910 struct mlx4_ib_steering {
911 struct list_head list;
912 struct mlx4_flow_reg_id reg_id;
916 static int parse_flow_attr(struct mlx4_dev *dev,
918 union ib_flow_spec *ib_spec,
919 struct _rule_hw *mlx4_spec)
921 enum mlx4_net_trans_rule_id type;
923 switch (ib_spec->type) {
924 case IB_FLOW_SPEC_ETH:
925 type = MLX4_NET_TRANS_RULE_ID_ETH;
926 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
928 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
930 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
931 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
933 case IB_FLOW_SPEC_IB:
934 type = MLX4_NET_TRANS_RULE_ID_IB;
935 mlx4_spec->ib.l3_qpn =
937 mlx4_spec->ib.qpn_mask =
938 cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
942 case IB_FLOW_SPEC_IPV4:
943 type = MLX4_NET_TRANS_RULE_ID_IPV4;
944 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
945 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
946 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
947 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
950 case IB_FLOW_SPEC_TCP:
951 case IB_FLOW_SPEC_UDP:
952 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
953 MLX4_NET_TRANS_RULE_ID_TCP :
954 MLX4_NET_TRANS_RULE_ID_UDP;
955 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
956 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
957 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
958 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
964 if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
965 mlx4_hw_rule_sz(dev, type) < 0)
967 mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
968 mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
969 return mlx4_hw_rule_sz(dev, type);
972 struct default_rules {
973 __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
974 __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
975 __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
978 static const struct default_rules default_table[] = {
980 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
981 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
982 .rules_create_list = {IB_FLOW_SPEC_IB},
983 .link_layer = IB_LINK_LAYER_INFINIBAND
987 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
988 struct ib_flow_attr *flow_attr)
992 const struct default_rules *pdefault_rules = default_table;
993 u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
995 for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
996 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
997 memset(&field_types, 0, sizeof(field_types));
999 if (link_layer != pdefault_rules->link_layer)
1002 ib_flow = flow_attr + 1;
1003 /* we assume the specs are sorted */
1004 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1005 j < flow_attr->num_of_specs; k++) {
1006 union ib_flow_spec *current_flow =
1007 (union ib_flow_spec *)ib_flow;
1009 /* same layer but different type */
1010 if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1011 (pdefault_rules->mandatory_fields[k] &
1012 IB_FLOW_SPEC_LAYER_MASK)) &&
1013 (current_flow->type !=
1014 pdefault_rules->mandatory_fields[k]))
1017 /* same layer, try match next one */
1018 if (current_flow->type ==
1019 pdefault_rules->mandatory_fields[k]) {
1022 ((union ib_flow_spec *)ib_flow)->size;
1026 ib_flow = flow_attr + 1;
1027 for (j = 0; j < flow_attr->num_of_specs;
1028 j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1029 for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1030 /* same layer and same type */
1031 if (((union ib_flow_spec *)ib_flow)->type ==
1032 pdefault_rules->mandatory_not_fields[k])
1041 static int __mlx4_ib_create_default_rules(
1042 struct mlx4_ib_dev *mdev,
1044 const struct default_rules *pdefault_rules,
1045 struct _rule_hw *mlx4_spec) {
1049 for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1051 union ib_flow_spec ib_spec;
1052 switch (pdefault_rules->rules_create_list[i]) {
1056 case IB_FLOW_SPEC_IB:
1057 ib_spec.type = IB_FLOW_SPEC_IB;
1058 ib_spec.size = sizeof(struct ib_flow_spec_ib);
1065 /* We must put empty rule, qpn is being ignored */
1066 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1069 pr_info("invalid parsing\n");
1073 mlx4_spec = (void *)mlx4_spec + ret;
1079 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1081 enum mlx4_net_trans_promisc_mode flow_type,
1087 struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1088 struct mlx4_cmd_mailbox *mailbox;
1089 struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1092 static const u16 __mlx4_domain[] = {
1093 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1094 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1095 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1096 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1099 if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1100 pr_err("Invalid priority value %d\n", flow_attr->priority);
1104 if (domain >= IB_FLOW_DOMAIN_NUM) {
1105 pr_err("Invalid domain value %d\n", domain);
1109 if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1112 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1113 if (IS_ERR(mailbox))
1114 return PTR_ERR(mailbox);
1115 ctrl = mailbox->buf;
1117 ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1118 flow_attr->priority);
1119 ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1120 ctrl->port = flow_attr->port;
1121 ctrl->qpn = cpu_to_be32(qp->qp_num);
1123 ib_flow = flow_attr + 1;
1124 size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1125 /* Add default flows */
1126 default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1127 if (default_flow >= 0) {
1128 ret = __mlx4_ib_create_default_rules(
1129 mdev, qp, default_table + default_flow,
1130 mailbox->buf + size);
1132 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1137 for (i = 0; i < flow_attr->num_of_specs; i++) {
1138 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1139 mailbox->buf + size);
1141 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1144 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1148 ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1149 MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1152 pr_err("mcg table is full. Fail to register network rule.\n");
1153 else if (ret == -ENXIO)
1154 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1156 pr_err("Invalid argumant. Fail to register network rule.\n");
1158 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1162 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1165 err = mlx4_cmd(dev, reg_id, 0, 0,
1166 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1169 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1174 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1178 union ib_flow_spec *ib_spec;
1179 struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1182 if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1183 dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1184 return 0; /* do nothing */
1186 ib_flow = flow_attr + 1;
1187 ib_spec = (union ib_flow_spec *)ib_flow;
1189 if (ib_spec->type != IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1190 return 0; /* do nothing */
1192 err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1193 flow_attr->port, qp->qp_num,
1194 MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1199 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1200 struct ib_flow_attr *flow_attr,
1203 int err = 0, i = 0, j = 0;
1204 struct mlx4_ib_flow *mflow;
1205 enum mlx4_net_trans_promisc_mode type[2];
1206 struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1207 int is_bonded = mlx4_is_bonded(dev);
1209 memset(type, 0, sizeof(type));
1211 mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1217 switch (flow_attr->type) {
1218 case IB_FLOW_ATTR_NORMAL:
1219 type[0] = MLX4_FS_REGULAR;
1222 case IB_FLOW_ATTR_ALL_DEFAULT:
1223 type[0] = MLX4_FS_ALL_DEFAULT;
1226 case IB_FLOW_ATTR_MC_DEFAULT:
1227 type[0] = MLX4_FS_MC_DEFAULT;
1230 case IB_FLOW_ATTR_SNIFFER:
1231 type[0] = MLX4_FS_UC_SNIFFER;
1232 type[1] = MLX4_FS_MC_SNIFFER;
1240 while (i < ARRAY_SIZE(type) && type[i]) {
1241 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1242 &mflow->reg_id[i].id);
1244 goto err_create_flow;
1246 /* Application always sees one port so the mirror rule
1247 * must be on port #2
1249 flow_attr->port = 2;
1250 err = __mlx4_ib_create_flow(qp, flow_attr,
1252 &mflow->reg_id[j].mirror);
1253 flow_attr->port = 1;
1255 goto err_create_flow;
1262 if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1263 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1264 &mflow->reg_id[i].id);
1266 goto err_create_flow;
1269 flow_attr->port = 2;
1270 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1271 &mflow->reg_id[j].mirror);
1272 flow_attr->port = 1;
1274 goto err_create_flow;
1277 /* function to create mirror rule */
1281 return &mflow->ibflow;
1285 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1286 mflow->reg_id[i].id);
1291 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1292 mflow->reg_id[j].mirror);
1297 return ERR_PTR(err);
1300 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1304 struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1305 struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1307 while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1308 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1311 if (mflow->reg_id[i].mirror) {
1312 err = __mlx4_ib_destroy_flow(mdev->dev,
1313 mflow->reg_id[i].mirror);
1324 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1327 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1328 struct mlx4_dev *dev = mdev->dev;
1329 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1330 struct mlx4_ib_steering *ib_steering = NULL;
1331 enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1332 struct mlx4_flow_reg_id reg_id;
1334 if (mdev->dev->caps.steering_mode ==
1335 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1336 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1341 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1343 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1346 pr_err("multicast attach op failed, err %d\n", err);
1351 if (mlx4_is_bonded(dev)) {
1352 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1353 (mqp->port == 1) ? 2 : 1,
1355 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1356 prot, ®_id.mirror);
1361 err = add_gid_entry(ibqp, gid);
1366 memcpy(ib_steering->gid.raw, gid->raw, 16);
1367 ib_steering->reg_id = reg_id;
1368 mutex_lock(&mqp->mutex);
1369 list_add(&ib_steering->list, &mqp->steering_rules);
1370 mutex_unlock(&mqp->mutex);
1375 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1378 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1379 prot, reg_id.mirror);
1386 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1388 struct mlx4_ib_gid_entry *ge;
1389 struct mlx4_ib_gid_entry *tmp;
1390 struct mlx4_ib_gid_entry *ret = NULL;
1392 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1393 if (!memcmp(raw, ge->gid.raw, 16)) {
1402 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1405 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1406 struct mlx4_dev *dev = mdev->dev;
1407 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1408 struct net_device *ndev;
1409 struct mlx4_ib_gid_entry *ge;
1410 struct mlx4_flow_reg_id reg_id = {0, 0};
1411 enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1413 if (mdev->dev->caps.steering_mode ==
1414 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1415 struct mlx4_ib_steering *ib_steering;
1417 mutex_lock(&mqp->mutex);
1418 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1419 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1420 list_del(&ib_steering->list);
1424 mutex_unlock(&mqp->mutex);
1425 if (&ib_steering->list == &mqp->steering_rules) {
1426 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1429 reg_id = ib_steering->reg_id;
1433 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1438 if (mlx4_is_bonded(dev)) {
1439 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1440 prot, reg_id.mirror);
1445 mutex_lock(&mqp->mutex);
1446 ge = find_gid_entry(mqp, gid->raw);
1448 spin_lock_bh(&mdev->iboe.lock);
1449 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1452 spin_unlock_bh(&mdev->iboe.lock);
1455 list_del(&ge->list);
1458 pr_warn("could not find mgid entry\n");
1460 mutex_unlock(&mqp->mutex);
1465 static int init_node_data(struct mlx4_ib_dev *dev)
1467 struct ib_smp *in_mad = NULL;
1468 struct ib_smp *out_mad = NULL;
1469 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1472 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1473 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1474 if (!in_mad || !out_mad)
1477 init_query_mad(in_mad);
1478 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1479 if (mlx4_is_master(dev->dev))
1480 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1482 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1486 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1488 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1490 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1494 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1495 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1503 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1506 struct mlx4_ib_dev *dev =
1507 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1508 return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1511 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1514 struct mlx4_ib_dev *dev =
1515 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1516 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1517 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1518 (int) dev->dev->caps.fw_ver & 0xffff);
1521 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1524 struct mlx4_ib_dev *dev =
1525 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1526 return sprintf(buf, "%x\n", dev->dev->rev_id);
1529 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1532 struct mlx4_ib_dev *dev =
1533 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1534 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1535 dev->dev->board_id);
1538 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1539 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1540 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1541 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1543 static struct device_attribute *mlx4_class_attributes[] = {
1550 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1551 struct net_device *dev)
1553 memcpy(eui, dev->dev_addr, 3);
1554 memcpy(eui + 5, dev->dev_addr + 3, 3);
1555 if (vlan_id < 0x1000) {
1556 eui[3] = vlan_id >> 8;
1557 eui[4] = vlan_id & 0xff;
1565 static void update_gids_task(struct work_struct *work)
1567 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1568 struct mlx4_cmd_mailbox *mailbox;
1571 struct mlx4_dev *dev = gw->dev->dev;
1572 int is_bonded = mlx4_is_bonded(dev);
1574 if (!gw->dev->ib_active)
1577 mailbox = mlx4_alloc_cmd_mailbox(dev);
1578 if (IS_ERR(mailbox)) {
1579 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1583 gids = mailbox->buf;
1584 memcpy(gids, gw->gids, sizeof gw->gids);
1586 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1587 MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1588 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
1590 pr_warn("set port command failed\n");
1592 if ((gw->port == 1) || !is_bonded)
1593 mlx4_ib_dispatch_event(gw->dev,
1594 is_bonded ? 1 : gw->port,
1595 IB_EVENT_GID_CHANGE);
1597 mlx4_free_cmd_mailbox(dev, mailbox);
1601 static void reset_gids_task(struct work_struct *work)
1603 struct update_gid_work *gw =
1604 container_of(work, struct update_gid_work, work);
1605 struct mlx4_cmd_mailbox *mailbox;
1608 struct mlx4_dev *dev = gw->dev->dev;
1610 if (!gw->dev->ib_active)
1613 mailbox = mlx4_alloc_cmd_mailbox(dev);
1614 if (IS_ERR(mailbox)) {
1615 pr_warn("reset gid table failed\n");
1619 gids = mailbox->buf;
1620 memcpy(gids, gw->gids, sizeof(gw->gids));
1622 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1623 IB_LINK_LAYER_ETHERNET) {
1624 err = mlx4_cmd(dev, mailbox->dma,
1625 MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1626 MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1627 MLX4_CMD_TIME_CLASS_B,
1630 pr_warn("set port %d command failed\n", gw->port);
1633 mlx4_free_cmd_mailbox(dev, mailbox);
1638 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1639 union ib_gid *gid, int clear,
1642 struct update_gid_work *work;
1644 int need_update = 0;
1652 max_gids = dev->dev->caps.gid_table_len[port];
1653 for (i = 1; i < max_gids; ++i) {
1654 if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1661 dev->iboe.gid_table[port - 1][found] =
1670 !memcmp(&dev->iboe.gid_table[port - 1][i],
1671 &zgid, sizeof(*gid)))
1677 if (found == -1 && !clear && free >= 0) {
1678 dev->iboe.gid_table[port - 1][free] = *gid;
1685 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1689 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1690 INIT_WORK(&work->work, update_gids_task);
1693 queue_work(wq, &work->work);
1698 static void mlx4_make_default_gid(struct net_device *dev, union ib_gid *gid)
1700 gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1701 mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1705 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1707 struct update_gid_work *work;
1709 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1713 memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1714 memset(work->gids, 0, sizeof(work->gids));
1715 INIT_WORK(&work->work, reset_gids_task);
1718 queue_work(wq, &work->work);
1722 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1723 struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1725 struct mlx4_ib_iboe *iboe;
1727 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1728 rdma_vlan_dev_real_dev(event_netdev) :
1730 union ib_gid default_gid;
1732 mlx4_make_default_gid(real_dev, &default_gid);
1734 if (!memcmp(gid, &default_gid, sizeof(*gid)))
1737 if (event != NETDEV_DOWN && event != NETDEV_UP)
1740 if ((real_dev != event_netdev) &&
1741 (event == NETDEV_DOWN) &&
1742 rdma_link_local_addr((struct in6_addr *)gid))
1745 iboe = &ibdev->iboe;
1746 spin_lock_bh(&iboe->lock);
1748 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1749 if ((netif_is_bond_master(real_dev) &&
1750 (real_dev == iboe->masters[port - 1])) ||
1751 (!netif_is_bond_master(real_dev) &&
1752 (real_dev == iboe->netdevs[port - 1])))
1753 update_gid_table(ibdev, port, gid,
1754 event == NETDEV_DOWN, 0);
1756 spin_unlock_bh(&iboe->lock);
1761 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1762 struct mlx4_ib_dev *ibdev)
1765 struct mlx4_ib_iboe *iboe;
1766 struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1767 rdma_vlan_dev_real_dev(dev) : dev;
1769 iboe = &ibdev->iboe;
1771 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1772 if ((netif_is_bond_master(real_dev) &&
1773 (real_dev == iboe->masters[port - 1])) ||
1774 (!netif_is_bond_master(real_dev) &&
1775 (real_dev == iboe->netdevs[port - 1])))
1778 if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1784 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1787 struct mlx4_ib_dev *ibdev;
1788 struct in_ifaddr *ifa = ptr;
1790 struct net_device *event_netdev = ifa->ifa_dev->dev;
1792 ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1794 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1796 mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1800 #if IS_ENABLED(CONFIG_IPV6)
1801 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1804 struct mlx4_ib_dev *ibdev;
1805 struct inet6_ifaddr *ifa = ptr;
1806 union ib_gid *gid = (union ib_gid *)&ifa->addr;
1807 struct net_device *event_netdev = ifa->idev->dev;
1809 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1811 mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1816 #define MLX4_IB_INVALID_MAC ((u64)-1)
1817 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1818 struct net_device *dev,
1822 u64 release_mac = MLX4_IB_INVALID_MAC;
1823 struct mlx4_ib_qp *qp;
1825 read_lock(&dev_base_lock);
1826 new_smac = mlx4_mac_to_u64(dev->dev_addr);
1827 read_unlock(&dev_base_lock);
1829 atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1831 /* no need for update QP1 and mac registration in non-SRIOV */
1832 if (!mlx4_is_mfunc(ibdev->dev))
1835 mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1836 qp = ibdev->qp1_proxy[port - 1];
1840 struct mlx4_update_qp_params update_params;
1842 mutex_lock(&qp->mutex);
1843 old_smac = qp->pri.smac;
1844 if (new_smac == old_smac)
1847 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1849 if (new_smac_index < 0)
1852 update_params.smac_index = new_smac_index;
1853 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1855 release_mac = new_smac;
1858 /* if old port was zero, no mac was yet registered for this QP */
1859 if (qp->pri.smac_port)
1860 release_mac = old_smac;
1861 qp->pri.smac = new_smac;
1862 qp->pri.smac_port = port;
1863 qp->pri.smac_index = new_smac_index;
1867 if (release_mac != MLX4_IB_INVALID_MAC)
1868 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1870 mutex_unlock(&qp->mutex);
1871 mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1874 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1875 struct mlx4_ib_dev *ibdev, u8 port)
1877 struct in_device *in_dev;
1878 #if IS_ENABLED(CONFIG_IPV6)
1879 struct inet6_dev *in6_dev;
1881 struct inet6_ifaddr *ifp;
1882 union ib_gid default_gid;
1887 if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1891 in_dev = in_dev_get(dev);
1894 /*ifa->ifa_address;*/
1895 ipv6_addr_set_v4mapped(ifa->ifa_address,
1896 (struct in6_addr *)&gid);
1897 update_gid_table(ibdev, port, &gid, 0, 0);
1902 #if IS_ENABLED(CONFIG_IPV6)
1903 mlx4_make_default_gid(dev, &default_gid);
1905 in6_dev = in6_dev_get(dev);
1907 read_lock_bh(&in6_dev->lock);
1908 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1909 pgid = (union ib_gid *)&ifp->addr;
1910 if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1912 update_gid_table(ibdev, port, pgid, 0, 0);
1914 read_unlock_bh(&in6_dev->lock);
1915 in6_dev_put(in6_dev);
1920 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1921 struct net_device *dev, u8 port)
1924 mlx4_make_default_gid(dev, &gid);
1925 update_gid_table(ibdev, port, &gid, 0, 1);
1928 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1930 struct net_device *dev;
1931 struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1935 for (i = 1; i <= ibdev->num_ports; ++i) {
1936 if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1937 IB_LINK_LAYER_ETHERNET) {
1938 err = reset_gid_table(ibdev, i);
1944 read_lock(&dev_base_lock);
1945 spin_lock_bh(&iboe->lock);
1947 for_each_netdev(&init_net, dev) {
1948 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1949 /* port will be non-zero only for ETH ports */
1951 mlx4_ib_set_default_gid(ibdev, dev, port);
1952 mlx4_ib_get_dev_addr(dev, ibdev, port);
1956 spin_unlock_bh(&iboe->lock);
1957 read_unlock(&dev_base_lock);
1962 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1963 struct net_device *dev,
1964 unsigned long event)
1967 struct mlx4_ib_iboe *iboe;
1968 int update_qps_port = -1;
1971 iboe = &ibdev->iboe;
1973 spin_lock_bh(&iboe->lock);
1974 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1975 enum ib_port_state port_state = IB_PORT_NOP;
1976 struct net_device *old_master = iboe->masters[port - 1];
1977 struct net_device *curr_netdev;
1978 struct net_device *curr_master;
1980 iboe->netdevs[port - 1] =
1981 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1982 if (iboe->netdevs[port - 1])
1983 mlx4_ib_set_default_gid(ibdev,
1984 iboe->netdevs[port - 1], port);
1985 curr_netdev = iboe->netdevs[port - 1];
1987 if (iboe->netdevs[port - 1] &&
1988 netif_is_bond_slave(iboe->netdevs[port - 1])) {
1989 iboe->masters[port - 1] = netdev_master_upper_dev_get(
1990 iboe->netdevs[port - 1]);
1992 iboe->masters[port - 1] = NULL;
1994 curr_master = iboe->masters[port - 1];
1996 if (dev == iboe->netdevs[port - 1] &&
1997 (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1998 event == NETDEV_UP || event == NETDEV_CHANGE))
1999 update_qps_port = port;
2002 port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
2003 IB_PORT_ACTIVE : IB_PORT_DOWN;
2004 mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
2006 /* if using bonding/team and a slave port is down, we
2007 * don't want the bond IP based gids in the table since
2008 * flows that select port by gid may get the down port.
2010 if (port_state == IB_PORT_DOWN &&
2011 !mlx4_is_bonded(ibdev->dev)) {
2012 reset_gid_table(ibdev, port);
2013 mlx4_ib_set_default_gid(ibdev,
2017 /* gids from the upper dev (bond/team)
2018 * should appear in port's gid table
2020 mlx4_ib_get_dev_addr(curr_master,
2024 /* if bonding is used it is possible that we add it to
2025 * masters only after IP address is assigned to the
2026 * net bonding interface.
2028 if (curr_master && (old_master != curr_master)) {
2029 reset_gid_table(ibdev, port);
2030 mlx4_ib_set_default_gid(ibdev,
2032 mlx4_ib_get_dev_addr(curr_master, ibdev, port);
2035 if (!curr_master && (old_master != curr_master)) {
2036 reset_gid_table(ibdev, port);
2037 mlx4_ib_set_default_gid(ibdev,
2039 mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
2042 reset_gid_table(ibdev, port);
2046 spin_unlock_bh(&iboe->lock);
2048 if (update_qps_port > 0)
2049 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2052 static int mlx4_ib_netdev_event(struct notifier_block *this,
2053 unsigned long event, void *ptr)
2055 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2056 struct mlx4_ib_dev *ibdev;
2058 if (!net_eq(dev_net(dev), &init_net))
2061 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2062 mlx4_ib_scan_netdevs(ibdev, dev, event);
2067 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2073 if (mlx4_is_master(ibdev->dev)) {
2074 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2076 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2078 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2080 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2081 /* master has the identity virt2phys pkey mapping */
2082 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2083 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2084 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2085 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2089 /* initialize pkey cache */
2090 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2092 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2094 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2100 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2102 int i, j, eq = 0, total_eqs = 0;
2104 ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2105 sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2106 if (!ibdev->eq_table)
2109 for (i = 1; i <= dev->caps.num_ports; i++) {
2110 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2112 if (i > 1 && mlx4_is_eq_shared(dev, total_eqs))
2114 ibdev->eq_table[eq] = total_eqs;
2115 if (!mlx4_assign_eq(dev, i,
2116 &ibdev->eq_table[eq]))
2119 ibdev->eq_table[eq] = -1;
2123 for (i = eq; i < dev->caps.num_comp_vectors;
2124 ibdev->eq_table[i++] = -1)
2127 /* Advertise the new number of EQs to clients */
2128 ibdev->ib_dev.num_comp_vectors = eq;
2131 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2134 int total_eqs = ibdev->ib_dev.num_comp_vectors;
2136 /* no eqs were allocated */
2137 if (!ibdev->eq_table)
2140 /* Reset the advertised EQ number */
2141 ibdev->ib_dev.num_comp_vectors = 0;
2143 for (i = 0; i < total_eqs; i++)
2144 mlx4_release_eq(dev, ibdev->eq_table[i]);
2146 kfree(ibdev->eq_table);
2147 ibdev->eq_table = NULL;
2150 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2151 struct ib_port_immutable *immutable)
2153 struct ib_port_attr attr;
2156 err = mlx4_ib_query_port(ibdev, port_num, &attr);
2160 immutable->pkey_tbl_len = attr.pkey_tbl_len;
2161 immutable->gid_tbl_len = attr.gid_tbl_len;
2163 if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2164 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2166 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2168 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2173 static void *mlx4_ib_add(struct mlx4_dev *dev)
2175 struct mlx4_ib_dev *ibdev;
2179 struct mlx4_ib_iboe *iboe;
2180 int ib_num_ports = 0;
2181 int num_req_counters;
2185 pr_info_once("%s", mlx4_ib_version);
2188 mlx4_foreach_ib_transport_port(i, dev)
2191 /* No point in registering a device with no ports... */
2195 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2197 dev_err(&dev->persist->pdev->dev,
2198 "Device struct alloc failed\n");
2202 iboe = &ibdev->iboe;
2204 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2207 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2210 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2212 if (!ibdev->uar_map)
2214 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2217 ibdev->bond_next_port = 0;
2219 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2220 ibdev->ib_dev.owner = THIS_MODULE;
2221 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
2222 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
2223 ibdev->num_ports = num_ports;
2224 ibdev->ib_dev.phys_port_cnt = mlx4_is_bonded(dev) ?
2225 1 : ibdev->num_ports;
2226 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2227 ibdev->ib_dev.dma_device = &dev->persist->pdev->dev;
2229 if (dev->caps.userspace_caps)
2230 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2232 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2234 ibdev->ib_dev.uverbs_cmd_mask =
2235 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
2236 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
2237 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
2238 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
2239 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
2240 (1ull << IB_USER_VERBS_CMD_REG_MR) |
2241 (1ull << IB_USER_VERBS_CMD_REREG_MR) |
2242 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
2243 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2244 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
2245 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
2246 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
2247 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
2248 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
2249 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
2250 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
2251 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
2252 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
2253 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
2254 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
2255 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
2256 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
2257 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
2258 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2260 ibdev->ib_dev.query_device = mlx4_ib_query_device;
2261 ibdev->ib_dev.query_port = mlx4_ib_query_port;
2262 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
2263 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
2264 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
2265 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
2266 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
2267 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
2268 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
2269 ibdev->ib_dev.mmap = mlx4_ib_mmap;
2270 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
2271 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
2272 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
2273 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
2274 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
2275 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
2276 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
2277 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
2278 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
2279 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
2280 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
2281 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
2282 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
2283 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
2284 ibdev->ib_dev.post_send = mlx4_ib_post_send;
2285 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
2286 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
2287 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
2288 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
2289 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
2290 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
2291 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
2292 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
2293 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
2294 ibdev->ib_dev.rereg_user_mr = mlx4_ib_rereg_user_mr;
2295 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
2296 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2297 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2298 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
2299 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
2300 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
2301 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
2302 ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2304 if (!mlx4_is_slave(ibdev->dev)) {
2305 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
2306 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
2307 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
2308 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
2311 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2312 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2313 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2314 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2315 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2317 ibdev->ib_dev.uverbs_cmd_mask |=
2318 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2319 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2322 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2323 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2324 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2325 ibdev->ib_dev.uverbs_cmd_mask |=
2326 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2327 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2330 if (check_flow_steering_support(dev)) {
2331 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2332 ibdev->ib_dev.create_flow = mlx4_ib_create_flow;
2333 ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow;
2335 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2336 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2337 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2340 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2341 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2342 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ);
2344 mlx4_ib_alloc_eqs(dev, ibdev);
2346 spin_lock_init(&iboe->lock);
2348 if (init_node_data(ibdev))
2351 num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2352 for (i = 0; i < num_req_counters; ++i) {
2353 mutex_init(&ibdev->qp1_proxy_lock[i]);
2355 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2356 IB_LINK_LAYER_ETHERNET) {
2357 err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2358 /* if failed to allocate a new counter, use default */
2361 mlx4_get_default_counter_index(dev,
2365 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2366 counter_index = mlx4_get_default_counter_index(dev,
2369 ibdev->counters[i].index = counter_index;
2370 ibdev->counters[i].allocated = allocated;
2371 pr_info("counter index %d for port %d allocated %d\n",
2372 counter_index, i + 1, allocated);
2374 if (mlx4_is_bonded(dev))
2375 for (i = 1; i < ibdev->num_ports ; ++i) {
2376 ibdev->counters[i].index = ibdev->counters[0].index;
2377 ibdev->counters[i].allocated = 0;
2380 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2383 spin_lock_init(&ibdev->sm_lock);
2384 mutex_init(&ibdev->cap_mask_mutex);
2385 INIT_LIST_HEAD(&ibdev->qp_list);
2386 spin_lock_init(&ibdev->reset_flow_resource_lock);
2388 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2390 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2391 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2392 MLX4_IB_UC_STEER_QPN_ALIGN,
2393 &ibdev->steer_qpn_base, 0);
2397 ibdev->ib_uc_qpns_bitmap =
2398 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2401 if (!ibdev->ib_uc_qpns_bitmap) {
2402 dev_err(&dev->persist->pdev->dev,
2403 "bit map alloc failed\n");
2404 goto err_steer_qp_release;
2407 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2409 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2410 dev, ibdev->steer_qpn_base,
2411 ibdev->steer_qpn_base +
2412 ibdev->steer_qpn_count - 1);
2414 goto err_steer_free_bitmap;
2417 for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2418 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2420 if (ib_register_device(&ibdev->ib_dev, NULL))
2421 goto err_steer_free_bitmap;
2423 if (mlx4_ib_mad_init(ibdev))
2426 if (mlx4_ib_init_sriov(ibdev))
2429 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2430 if (!iboe->nb.notifier_call) {
2431 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2432 err = register_netdevice_notifier(&iboe->nb);
2434 iboe->nb.notifier_call = NULL;
2438 if (!iboe->nb_inet.notifier_call) {
2439 iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2440 err = register_inetaddr_notifier(&iboe->nb_inet);
2442 iboe->nb_inet.notifier_call = NULL;
2446 #if IS_ENABLED(CONFIG_IPV6)
2447 if (!iboe->nb_inet6.notifier_call) {
2448 iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2449 err = register_inet6addr_notifier(&iboe->nb_inet6);
2451 iboe->nb_inet6.notifier_call = NULL;
2456 if (mlx4_ib_init_gid_table(ibdev))
2460 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2461 if (device_create_file(&ibdev->ib_dev.dev,
2462 mlx4_class_attributes[j]))
2466 ibdev->ib_active = true;
2468 if (mlx4_is_mfunc(ibdev->dev))
2471 /* create paravirt contexts for any VFs which are active */
2472 if (mlx4_is_master(ibdev->dev)) {
2473 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2474 if (j == mlx4_master_func_num(ibdev->dev))
2476 if (mlx4_is_slave_active(ibdev->dev, j))
2477 do_slave_init(ibdev, j, 1);
2483 if (ibdev->iboe.nb.notifier_call) {
2484 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2485 pr_warn("failure unregistering notifier\n");
2486 ibdev->iboe.nb.notifier_call = NULL;
2488 if (ibdev->iboe.nb_inet.notifier_call) {
2489 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2490 pr_warn("failure unregistering notifier\n");
2491 ibdev->iboe.nb_inet.notifier_call = NULL;
2493 #if IS_ENABLED(CONFIG_IPV6)
2494 if (ibdev->iboe.nb_inet6.notifier_call) {
2495 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2496 pr_warn("failure unregistering notifier\n");
2497 ibdev->iboe.nb_inet6.notifier_call = NULL;
2500 flush_workqueue(wq);
2502 mlx4_ib_close_sriov(ibdev);
2505 mlx4_ib_mad_cleanup(ibdev);
2508 ib_unregister_device(&ibdev->ib_dev);
2510 err_steer_free_bitmap:
2511 kfree(ibdev->ib_uc_qpns_bitmap);
2513 err_steer_qp_release:
2514 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2515 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2516 ibdev->steer_qpn_count);
2518 for (i = 0; i < ibdev->num_ports; ++i) {
2519 if (ibdev->counters[i].index != -1 &&
2520 ibdev->counters[i].allocated)
2521 mlx4_counter_free(ibdev->dev,
2522 ibdev->counters[i].index);
2525 iounmap(ibdev->uar_map);
2528 mlx4_uar_free(dev, &ibdev->priv_uar);
2531 mlx4_pd_free(dev, ibdev->priv_pdn);
2534 ib_dealloc_device(&ibdev->ib_dev);
2539 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2543 WARN_ON(!dev->ib_uc_qpns_bitmap);
2545 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2546 dev->steer_qpn_count,
2547 get_count_order(count));
2551 *qpn = dev->steer_qpn_base + offset;
2555 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2558 dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2561 BUG_ON(qpn < dev->steer_qpn_base);
2563 bitmap_release_region(dev->ib_uc_qpns_bitmap,
2564 qpn - dev->steer_qpn_base,
2565 get_count_order(count));
2568 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2573 struct ib_flow_attr *flow = NULL;
2574 struct ib_flow_spec_ib *ib_spec;
2577 flow_size = sizeof(struct ib_flow_attr) +
2578 sizeof(struct ib_flow_spec_ib);
2579 flow = kzalloc(flow_size, GFP_KERNEL);
2582 flow->port = mqp->port;
2583 flow->num_of_specs = 1;
2584 flow->size = flow_size;
2585 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2586 ib_spec->type = IB_FLOW_SPEC_IB;
2587 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2588 /* Add an empty rule for IB L2 */
2589 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2591 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2596 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2602 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2604 struct mlx4_ib_dev *ibdev = ibdev_ptr;
2607 ibdev->ib_active = false;
2608 flush_workqueue(wq);
2610 mlx4_ib_close_sriov(ibdev);
2611 mlx4_ib_mad_cleanup(ibdev);
2612 ib_unregister_device(&ibdev->ib_dev);
2613 if (ibdev->iboe.nb.notifier_call) {
2614 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2615 pr_warn("failure unregistering notifier\n");
2616 ibdev->iboe.nb.notifier_call = NULL;
2619 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2620 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2621 ibdev->steer_qpn_count);
2622 kfree(ibdev->ib_uc_qpns_bitmap);
2625 if (ibdev->iboe.nb_inet.notifier_call) {
2626 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2627 pr_warn("failure unregistering notifier\n");
2628 ibdev->iboe.nb_inet.notifier_call = NULL;
2630 #if IS_ENABLED(CONFIG_IPV6)
2631 if (ibdev->iboe.nb_inet6.notifier_call) {
2632 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2633 pr_warn("failure unregistering notifier\n");
2634 ibdev->iboe.nb_inet6.notifier_call = NULL;
2638 iounmap(ibdev->uar_map);
2639 for (p = 0; p < ibdev->num_ports; ++p)
2640 if (ibdev->counters[p].index != -1 &&
2641 ibdev->counters[p].allocated)
2642 mlx4_counter_free(ibdev->dev, ibdev->counters[p].index);
2643 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2644 mlx4_CLOSE_PORT(dev, p);
2646 mlx4_ib_free_eqs(dev, ibdev);
2648 mlx4_uar_free(dev, &ibdev->priv_uar);
2649 mlx4_pd_free(dev, ibdev->priv_pdn);
2650 ib_dealloc_device(&ibdev->ib_dev);
2653 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2655 struct mlx4_ib_demux_work **dm = NULL;
2656 struct mlx4_dev *dev = ibdev->dev;
2658 unsigned long flags;
2659 struct mlx4_active_ports actv_ports;
2661 unsigned int first_port;
2663 if (!mlx4_is_master(dev))
2666 actv_ports = mlx4_get_active_ports(dev, slave);
2667 ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2668 first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2670 dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2672 pr_err("failed to allocate memory for tunneling qp update\n");
2676 for (i = 0; i < ports; i++) {
2677 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2679 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2684 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2685 dm[i]->port = first_port + i + 1;
2686 dm[i]->slave = slave;
2687 dm[i]->do_init = do_init;
2690 /* initialize or tear down tunnel QPs for the slave */
2691 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2692 if (!ibdev->sriov.is_going_down) {
2693 for (i = 0; i < ports; i++)
2694 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2695 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2697 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2698 for (i = 0; i < ports; i++)
2706 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2708 struct mlx4_ib_qp *mqp;
2709 unsigned long flags_qp;
2710 unsigned long flags_cq;
2711 struct mlx4_ib_cq *send_mcq, *recv_mcq;
2712 struct list_head cq_notify_list;
2713 struct mlx4_cq *mcq;
2714 unsigned long flags;
2716 pr_warn("mlx4_ib_handle_catas_error was started\n");
2717 INIT_LIST_HEAD(&cq_notify_list);
2719 /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2720 spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2722 list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2723 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2724 if (mqp->sq.tail != mqp->sq.head) {
2725 send_mcq = to_mcq(mqp->ibqp.send_cq);
2726 spin_lock_irqsave(&send_mcq->lock, flags_cq);
2727 if (send_mcq->mcq.comp &&
2728 mqp->ibqp.send_cq->comp_handler) {
2729 if (!send_mcq->mcq.reset_notify_added) {
2730 send_mcq->mcq.reset_notify_added = 1;
2731 list_add_tail(&send_mcq->mcq.reset_notify,
2735 spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2737 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2738 /* Now, handle the QP's receive queue */
2739 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2740 /* no handling is needed for SRQ */
2741 if (!mqp->ibqp.srq) {
2742 if (mqp->rq.tail != mqp->rq.head) {
2743 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2744 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2745 if (recv_mcq->mcq.comp &&
2746 mqp->ibqp.recv_cq->comp_handler) {
2747 if (!recv_mcq->mcq.reset_notify_added) {
2748 recv_mcq->mcq.reset_notify_added = 1;
2749 list_add_tail(&recv_mcq->mcq.reset_notify,
2753 spin_unlock_irqrestore(&recv_mcq->lock,
2757 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2760 list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2763 spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2764 pr_warn("mlx4_ib_handle_catas_error ended\n");
2767 static void handle_bonded_port_state_event(struct work_struct *work)
2769 struct ib_event_work *ew =
2770 container_of(work, struct ib_event_work, work);
2771 struct mlx4_ib_dev *ibdev = ew->ib_dev;
2772 enum ib_port_state bonded_port_state = IB_PORT_NOP;
2774 struct ib_event ibev;
2777 spin_lock_bh(&ibdev->iboe.lock);
2778 for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2779 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2780 enum ib_port_state curr_port_state;
2786 (netif_running(curr_netdev) &&
2787 netif_carrier_ok(curr_netdev)) ?
2788 IB_PORT_ACTIVE : IB_PORT_DOWN;
2790 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2791 curr_port_state : IB_PORT_ACTIVE;
2793 spin_unlock_bh(&ibdev->iboe.lock);
2795 ibev.device = &ibdev->ib_dev;
2796 ibev.element.port_num = 1;
2797 ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2798 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2800 ib_dispatch_event(&ibev);
2803 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2804 enum mlx4_dev_event event, unsigned long param)
2806 struct ib_event ibev;
2807 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2808 struct mlx4_eqe *eqe = NULL;
2809 struct ib_event_work *ew;
2812 if (mlx4_is_bonded(dev) &&
2813 ((event == MLX4_DEV_EVENT_PORT_UP) ||
2814 (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2815 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2818 INIT_WORK(&ew->work, handle_bonded_port_state_event);
2820 queue_work(wq, &ew->work);
2824 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2825 eqe = (struct mlx4_eqe *)param;
2830 case MLX4_DEV_EVENT_PORT_UP:
2831 if (p > ibdev->num_ports)
2833 if (mlx4_is_master(dev) &&
2834 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2835 IB_LINK_LAYER_INFINIBAND) {
2836 mlx4_ib_invalidate_all_guid_record(ibdev, p);
2838 ibev.event = IB_EVENT_PORT_ACTIVE;
2841 case MLX4_DEV_EVENT_PORT_DOWN:
2842 if (p > ibdev->num_ports)
2844 ibev.event = IB_EVENT_PORT_ERR;
2847 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2848 ibdev->ib_active = false;
2849 ibev.event = IB_EVENT_DEVICE_FATAL;
2850 mlx4_ib_handle_catas_error(ibdev);
2853 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2854 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2856 pr_err("failed to allocate memory for events work\n");
2860 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2861 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2863 /* need to queue only for port owner, which uses GEN_EQE */
2864 if (mlx4_is_master(dev))
2865 queue_work(wq, &ew->work);
2867 handle_port_mgmt_change_event(&ew->work);
2870 case MLX4_DEV_EVENT_SLAVE_INIT:
2871 /* here, p is the slave id */
2872 do_slave_init(ibdev, p, 1);
2873 if (mlx4_is_master(dev)) {
2876 for (i = 1; i <= ibdev->num_ports; i++) {
2877 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2878 == IB_LINK_LAYER_INFINIBAND)
2879 mlx4_ib_slave_alias_guid_event(ibdev,
2886 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2887 if (mlx4_is_master(dev)) {
2890 for (i = 1; i <= ibdev->num_ports; i++) {
2891 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2892 == IB_LINK_LAYER_INFINIBAND)
2893 mlx4_ib_slave_alias_guid_event(ibdev,
2898 /* here, p is the slave id */
2899 do_slave_init(ibdev, p, 0);
2906 ibev.device = ibdev_ptr;
2907 ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2909 ib_dispatch_event(&ibev);
2912 static struct mlx4_interface mlx4_ib_interface = {
2914 .remove = mlx4_ib_remove,
2915 .event = mlx4_ib_event,
2916 .protocol = MLX4_PROT_IB_IPV6,
2917 .flags = MLX4_INTFF_BONDING
2920 static int __init mlx4_ib_init(void)
2924 wq = create_singlethread_workqueue("mlx4_ib");
2928 err = mlx4_ib_mcg_init();
2932 err = mlx4_register_interface(&mlx4_ib_interface);
2939 mlx4_ib_mcg_destroy();
2942 destroy_workqueue(wq);
2946 static void __exit mlx4_ib_cleanup(void)
2948 mlx4_unregister_interface(&mlx4_ib_interface);
2949 mlx4_ib_mcg_destroy();
2950 destroy_workqueue(wq);
2953 module_init(mlx4_ib_init);
2954 module_exit(mlx4_ib_cleanup);