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)
137 struct mlx4_ib_dev *dev = to_mdev(ibdev);
138 struct ib_smp *in_mad = NULL;
139 struct ib_smp *out_mad = NULL;
143 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
144 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
145 if (!in_mad || !out_mad)
148 init_query_mad(in_mad);
149 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
151 err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
152 1, NULL, NULL, in_mad, out_mad);
156 memset(props, 0, sizeof *props);
158 have_ib_ports = num_ib_ports(dev->dev);
160 props->fw_ver = dev->dev->caps.fw_ver;
161 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
162 IB_DEVICE_PORT_ACTIVE_EVENT |
163 IB_DEVICE_SYS_IMAGE_GUID |
164 IB_DEVICE_RC_RNR_NAK_GEN |
165 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
166 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
167 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
168 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
169 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
170 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
171 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
172 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
173 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
174 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
175 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
176 if (dev->dev->caps.max_gso_sz &&
177 (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
178 (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
179 props->device_cap_flags |= IB_DEVICE_UD_TSO;
180 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
181 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
182 if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
183 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
184 (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
185 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
186 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
187 props->device_cap_flags |= IB_DEVICE_XRC;
188 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
189 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
190 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
191 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
192 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
194 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
195 if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
196 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
199 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
201 props->vendor_part_id = dev->dev->persist->pdev->device;
202 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
203 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
205 props->max_mr_size = ~0ull;
206 props->page_size_cap = dev->dev->caps.page_size_cap;
207 props->max_qp = dev->dev->quotas.qp;
208 props->max_qp_wr = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
209 props->max_sge = min(dev->dev->caps.max_sq_sg,
210 dev->dev->caps.max_rq_sg);
211 props->max_cq = dev->dev->quotas.cq;
212 props->max_cqe = dev->dev->caps.max_cqes;
213 props->max_mr = dev->dev->quotas.mpt;
214 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
215 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
216 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
217 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
218 props->max_srq = dev->dev->quotas.srq;
219 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
220 props->max_srq_sge = dev->dev->caps.max_srq_sge;
221 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
222 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
223 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
224 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
225 props->masked_atomic_cap = props->atomic_cap;
226 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
227 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
228 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
229 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
230 props->max_mcast_grp;
231 props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
240 static enum rdma_link_layer
241 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
243 struct mlx4_dev *dev = to_mdev(device)->dev;
245 return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
246 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
249 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
250 struct ib_port_attr *props, int netw_view)
252 struct ib_smp *in_mad = NULL;
253 struct ib_smp *out_mad = NULL;
254 int ext_active_speed;
255 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
258 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
259 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
260 if (!in_mad || !out_mad)
263 init_query_mad(in_mad);
264 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
265 in_mad->attr_mod = cpu_to_be32(port);
267 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
268 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
270 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
276 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
277 props->lmc = out_mad->data[34] & 0x7;
278 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
279 props->sm_sl = out_mad->data[36] & 0xf;
280 props->state = out_mad->data[32] & 0xf;
281 props->phys_state = out_mad->data[33] >> 4;
282 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
284 props->gid_tbl_len = out_mad->data[50];
286 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
287 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
288 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
289 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
290 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
291 props->active_width = out_mad->data[31] & 0xf;
292 props->active_speed = out_mad->data[35] >> 4;
293 props->max_mtu = out_mad->data[41] & 0xf;
294 props->active_mtu = out_mad->data[36] >> 4;
295 props->subnet_timeout = out_mad->data[51] & 0x1f;
296 props->max_vl_num = out_mad->data[37] >> 4;
297 props->init_type_reply = out_mad->data[41] >> 4;
299 /* Check if extended speeds (EDR/FDR/...) are supported */
300 if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
301 ext_active_speed = out_mad->data[62] >> 4;
303 switch (ext_active_speed) {
305 props->active_speed = IB_SPEED_FDR;
308 props->active_speed = IB_SPEED_EDR;
313 /* If reported active speed is QDR, check if is FDR-10 */
314 if (props->active_speed == IB_SPEED_QDR) {
315 init_query_mad(in_mad);
316 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
317 in_mad->attr_mod = cpu_to_be32(port);
319 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
320 NULL, NULL, in_mad, out_mad);
324 /* Checking LinkSpeedActive for FDR-10 */
325 if (out_mad->data[15] & 0x1)
326 props->active_speed = IB_SPEED_FDR10;
329 /* Avoid wrong speed value returned by FW if the IB link is down. */
330 if (props->state == IB_PORT_DOWN)
331 props->active_speed = IB_SPEED_SDR;
339 static u8 state_to_phys_state(enum ib_port_state state)
341 return state == IB_PORT_ACTIVE ? 5 : 3;
344 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
345 struct ib_port_attr *props, int netw_view)
348 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
349 struct mlx4_ib_iboe *iboe = &mdev->iboe;
350 struct net_device *ndev;
352 struct mlx4_cmd_mailbox *mailbox;
354 int is_bonded = mlx4_is_bonded(mdev->dev);
356 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
358 return PTR_ERR(mailbox);
360 err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
361 MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
366 props->active_width = (((u8 *)mailbox->buf)[5] == 0x40) ?
367 IB_WIDTH_4X : IB_WIDTH_1X;
368 props->active_speed = IB_SPEED_QDR;
369 props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
370 props->gid_tbl_len = mdev->dev->caps.gid_table_len[port];
371 props->max_msg_sz = mdev->dev->caps.max_msg_sz;
372 props->pkey_tbl_len = 1;
373 props->max_mtu = IB_MTU_4096;
374 props->max_vl_num = 2;
375 props->state = IB_PORT_DOWN;
376 props->phys_state = state_to_phys_state(props->state);
377 props->active_mtu = IB_MTU_256;
379 rtnl_lock(); /* required to get upper dev */
380 spin_lock_bh(&iboe->lock);
381 ndev = iboe->netdevs[port - 1];
382 if (ndev && is_bonded)
383 ndev = netdev_master_upper_dev_get(ndev);
387 tmp = iboe_get_mtu(ndev->mtu);
388 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
390 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
391 IB_PORT_ACTIVE : IB_PORT_DOWN;
392 props->phys_state = state_to_phys_state(props->state);
394 spin_unlock_bh(&iboe->lock);
398 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
402 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
403 struct ib_port_attr *props, int netw_view)
407 memset(props, 0, sizeof *props);
409 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
410 ib_link_query_port(ibdev, port, props, netw_view) :
411 eth_link_query_port(ibdev, port, props, netw_view);
416 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
417 struct ib_port_attr *props)
419 /* returns host view */
420 return __mlx4_ib_query_port(ibdev, port, props, 0);
423 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
424 union ib_gid *gid, int netw_view)
426 struct ib_smp *in_mad = NULL;
427 struct ib_smp *out_mad = NULL;
429 struct mlx4_ib_dev *dev = to_mdev(ibdev);
431 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
433 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
434 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
435 if (!in_mad || !out_mad)
438 init_query_mad(in_mad);
439 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
440 in_mad->attr_mod = cpu_to_be32(port);
442 if (mlx4_is_mfunc(dev->dev) && netw_view)
443 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
445 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
449 memcpy(gid->raw, out_mad->data + 8, 8);
451 if (mlx4_is_mfunc(dev->dev) && !netw_view) {
453 /* For any index > 0, return the null guid */
460 init_query_mad(in_mad);
461 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
462 in_mad->attr_mod = cpu_to_be32(index / 8);
464 err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
465 NULL, NULL, in_mad, out_mad);
469 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
473 memset(gid->raw + 8, 0, 8);
479 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
482 struct mlx4_ib_dev *dev = to_mdev(ibdev);
484 *gid = dev->iboe.gid_table[port - 1][index];
489 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
492 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
493 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
495 return iboe_query_gid(ibdev, port, index, gid);
498 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
499 u16 *pkey, int netw_view)
501 struct ib_smp *in_mad = NULL;
502 struct ib_smp *out_mad = NULL;
503 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
506 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
507 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
508 if (!in_mad || !out_mad)
511 init_query_mad(in_mad);
512 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
513 in_mad->attr_mod = cpu_to_be32(index / 32);
515 if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
516 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
518 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
523 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
531 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
533 return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
536 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
537 struct ib_device_modify *props)
539 struct mlx4_cmd_mailbox *mailbox;
542 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
545 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
548 if (mlx4_is_slave(to_mdev(ibdev)->dev))
551 spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
552 memcpy(ibdev->node_desc, props->node_desc, 64);
553 spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
556 * If possible, pass node desc to FW, so it can generate
557 * a 144 trap. If cmd fails, just ignore.
559 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
563 memcpy(mailbox->buf, props->node_desc, 64);
564 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
565 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
567 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
572 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
575 struct mlx4_cmd_mailbox *mailbox;
578 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
580 return PTR_ERR(mailbox);
582 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
583 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
584 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
586 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
587 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
590 err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
591 MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
594 mlx4_free_cmd_mailbox(dev->dev, mailbox);
598 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
599 struct ib_port_modify *props)
601 struct mlx4_ib_dev *mdev = to_mdev(ibdev);
602 u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
603 struct ib_port_attr attr;
607 /* return OK if this is RoCE. CM calls ib_modify_port() regardless
608 * of whether port link layer is ETH or IB. For ETH ports, qkey
609 * violations and port capabilities are not meaningful.
614 mutex_lock(&mdev->cap_mask_mutex);
616 err = mlx4_ib_query_port(ibdev, port, &attr);
620 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
621 ~props->clr_port_cap_mask;
623 err = mlx4_ib_SET_PORT(mdev, port,
624 !!(mask & IB_PORT_RESET_QKEY_CNTR),
628 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
632 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
633 struct ib_udata *udata)
635 struct mlx4_ib_dev *dev = to_mdev(ibdev);
636 struct mlx4_ib_ucontext *context;
637 struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
638 struct mlx4_ib_alloc_ucontext_resp resp;
642 return ERR_PTR(-EAGAIN);
644 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
645 resp_v3.qp_tab_size = dev->dev->caps.num_qps;
646 resp_v3.bf_reg_size = dev->dev->caps.bf_reg_size;
647 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
649 resp.dev_caps = dev->dev->caps.userspace_caps;
650 resp.qp_tab_size = dev->dev->caps.num_qps;
651 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
652 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
653 resp.cqe_size = dev->dev->caps.cqe_size;
656 context = kmalloc(sizeof *context, GFP_KERNEL);
658 return ERR_PTR(-ENOMEM);
660 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
666 INIT_LIST_HEAD(&context->db_page_list);
667 mutex_init(&context->db_page_mutex);
669 if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
670 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
672 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
675 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
677 return ERR_PTR(-EFAULT);
680 return &context->ibucontext;
683 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
685 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
687 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
693 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
695 struct mlx4_ib_dev *dev = to_mdev(context->device);
697 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
700 if (vma->vm_pgoff == 0) {
701 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
703 if (io_remap_pfn_range(vma, vma->vm_start,
704 to_mucontext(context)->uar.pfn,
705 PAGE_SIZE, vma->vm_page_prot))
707 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
708 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
710 if (io_remap_pfn_range(vma, vma->vm_start,
711 to_mucontext(context)->uar.pfn +
712 dev->dev->caps.num_uars,
713 PAGE_SIZE, vma->vm_page_prot))
721 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
722 struct ib_ucontext *context,
723 struct ib_udata *udata)
725 struct mlx4_ib_pd *pd;
728 pd = kmalloc(sizeof *pd, GFP_KERNEL);
730 return ERR_PTR(-ENOMEM);
732 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
739 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
740 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
742 return ERR_PTR(-EFAULT);
748 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
750 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
756 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
757 struct ib_ucontext *context,
758 struct ib_udata *udata)
760 struct mlx4_ib_xrcd *xrcd;
763 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
764 return ERR_PTR(-ENOSYS);
766 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
768 return ERR_PTR(-ENOMEM);
770 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
774 xrcd->pd = ib_alloc_pd(ibdev);
775 if (IS_ERR(xrcd->pd)) {
776 err = PTR_ERR(xrcd->pd);
780 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
781 if (IS_ERR(xrcd->cq)) {
782 err = PTR_ERR(xrcd->cq);
786 return &xrcd->ibxrcd;
789 ib_dealloc_pd(xrcd->pd);
791 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
797 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
799 ib_destroy_cq(to_mxrcd(xrcd)->cq);
800 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
801 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
807 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
809 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
810 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
811 struct mlx4_ib_gid_entry *ge;
813 ge = kzalloc(sizeof *ge, GFP_KERNEL);
818 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
819 ge->port = mqp->port;
823 mutex_lock(&mqp->mutex);
824 list_add_tail(&ge->list, &mqp->gid_list);
825 mutex_unlock(&mqp->mutex);
830 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
833 struct net_device *ndev;
839 spin_lock_bh(&mdev->iboe.lock);
840 ndev = mdev->iboe.netdevs[mqp->port - 1];
843 spin_unlock_bh(&mdev->iboe.lock);
853 struct mlx4_ib_steering {
854 struct list_head list;
855 struct mlx4_flow_reg_id reg_id;
859 static int parse_flow_attr(struct mlx4_dev *dev,
861 union ib_flow_spec *ib_spec,
862 struct _rule_hw *mlx4_spec)
864 enum mlx4_net_trans_rule_id type;
866 switch (ib_spec->type) {
867 case IB_FLOW_SPEC_ETH:
868 type = MLX4_NET_TRANS_RULE_ID_ETH;
869 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
871 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
873 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
874 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
876 case IB_FLOW_SPEC_IB:
877 type = MLX4_NET_TRANS_RULE_ID_IB;
878 mlx4_spec->ib.l3_qpn =
880 mlx4_spec->ib.qpn_mask =
881 cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
885 case IB_FLOW_SPEC_IPV4:
886 type = MLX4_NET_TRANS_RULE_ID_IPV4;
887 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
888 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
889 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
890 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
893 case IB_FLOW_SPEC_TCP:
894 case IB_FLOW_SPEC_UDP:
895 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
896 MLX4_NET_TRANS_RULE_ID_TCP :
897 MLX4_NET_TRANS_RULE_ID_UDP;
898 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
899 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
900 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
901 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
907 if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
908 mlx4_hw_rule_sz(dev, type) < 0)
910 mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
911 mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
912 return mlx4_hw_rule_sz(dev, type);
915 struct default_rules {
916 __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
917 __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
918 __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
921 static const struct default_rules default_table[] = {
923 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
924 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
925 .rules_create_list = {IB_FLOW_SPEC_IB},
926 .link_layer = IB_LINK_LAYER_INFINIBAND
930 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
931 struct ib_flow_attr *flow_attr)
935 const struct default_rules *pdefault_rules = default_table;
936 u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
938 for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
939 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
940 memset(&field_types, 0, sizeof(field_types));
942 if (link_layer != pdefault_rules->link_layer)
945 ib_flow = flow_attr + 1;
946 /* we assume the specs are sorted */
947 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
948 j < flow_attr->num_of_specs; k++) {
949 union ib_flow_spec *current_flow =
950 (union ib_flow_spec *)ib_flow;
952 /* same layer but different type */
953 if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
954 (pdefault_rules->mandatory_fields[k] &
955 IB_FLOW_SPEC_LAYER_MASK)) &&
956 (current_flow->type !=
957 pdefault_rules->mandatory_fields[k]))
960 /* same layer, try match next one */
961 if (current_flow->type ==
962 pdefault_rules->mandatory_fields[k]) {
965 ((union ib_flow_spec *)ib_flow)->size;
969 ib_flow = flow_attr + 1;
970 for (j = 0; j < flow_attr->num_of_specs;
971 j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
972 for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
973 /* same layer and same type */
974 if (((union ib_flow_spec *)ib_flow)->type ==
975 pdefault_rules->mandatory_not_fields[k])
984 static int __mlx4_ib_create_default_rules(
985 struct mlx4_ib_dev *mdev,
987 const struct default_rules *pdefault_rules,
988 struct _rule_hw *mlx4_spec) {
992 for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
994 union ib_flow_spec ib_spec;
995 switch (pdefault_rules->rules_create_list[i]) {
999 case IB_FLOW_SPEC_IB:
1000 ib_spec.type = IB_FLOW_SPEC_IB;
1001 ib_spec.size = sizeof(struct ib_flow_spec_ib);
1008 /* We must put empty rule, qpn is being ignored */
1009 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1012 pr_info("invalid parsing\n");
1016 mlx4_spec = (void *)mlx4_spec + ret;
1022 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1024 enum mlx4_net_trans_promisc_mode flow_type,
1030 struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1031 struct mlx4_cmd_mailbox *mailbox;
1032 struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1035 static const u16 __mlx4_domain[] = {
1036 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1037 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1038 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1039 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1042 if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1043 pr_err("Invalid priority value %d\n", flow_attr->priority);
1047 if (domain >= IB_FLOW_DOMAIN_NUM) {
1048 pr_err("Invalid domain value %d\n", domain);
1052 if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1055 mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1056 if (IS_ERR(mailbox))
1057 return PTR_ERR(mailbox);
1058 ctrl = mailbox->buf;
1060 ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1061 flow_attr->priority);
1062 ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1063 ctrl->port = flow_attr->port;
1064 ctrl->qpn = cpu_to_be32(qp->qp_num);
1066 ib_flow = flow_attr + 1;
1067 size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1068 /* Add default flows */
1069 default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1070 if (default_flow >= 0) {
1071 ret = __mlx4_ib_create_default_rules(
1072 mdev, qp, default_table + default_flow,
1073 mailbox->buf + size);
1075 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1080 for (i = 0; i < flow_attr->num_of_specs; i++) {
1081 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1082 mailbox->buf + size);
1084 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1087 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1091 ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1092 MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1095 pr_err("mcg table is full. Fail to register network rule.\n");
1096 else if (ret == -ENXIO)
1097 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1099 pr_err("Invalid argumant. Fail to register network rule.\n");
1101 mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1105 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1108 err = mlx4_cmd(dev, reg_id, 0, 0,
1109 MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1112 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1117 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1121 union ib_flow_spec *ib_spec;
1122 struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1125 if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1126 dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1127 return 0; /* do nothing */
1129 ib_flow = flow_attr + 1;
1130 ib_spec = (union ib_flow_spec *)ib_flow;
1132 if (ib_spec->type != IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1133 return 0; /* do nothing */
1135 err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1136 flow_attr->port, qp->qp_num,
1137 MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1142 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1143 struct ib_flow_attr *flow_attr,
1146 int err = 0, i = 0, j = 0;
1147 struct mlx4_ib_flow *mflow;
1148 enum mlx4_net_trans_promisc_mode type[2];
1149 struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1150 int is_bonded = mlx4_is_bonded(dev);
1152 memset(type, 0, sizeof(type));
1154 mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1160 switch (flow_attr->type) {
1161 case IB_FLOW_ATTR_NORMAL:
1162 type[0] = MLX4_FS_REGULAR;
1165 case IB_FLOW_ATTR_ALL_DEFAULT:
1166 type[0] = MLX4_FS_ALL_DEFAULT;
1169 case IB_FLOW_ATTR_MC_DEFAULT:
1170 type[0] = MLX4_FS_MC_DEFAULT;
1173 case IB_FLOW_ATTR_SNIFFER:
1174 type[0] = MLX4_FS_UC_SNIFFER;
1175 type[1] = MLX4_FS_MC_SNIFFER;
1183 while (i < ARRAY_SIZE(type) && type[i]) {
1184 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1185 &mflow->reg_id[i].id);
1187 goto err_create_flow;
1189 /* Application always sees one port so the mirror rule
1190 * must be on port #2
1192 flow_attr->port = 2;
1193 err = __mlx4_ib_create_flow(qp, flow_attr,
1195 &mflow->reg_id[j].mirror);
1196 flow_attr->port = 1;
1198 goto err_create_flow;
1205 if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1206 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1207 &mflow->reg_id[i].id);
1209 goto err_create_flow;
1212 flow_attr->port = 2;
1213 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1214 &mflow->reg_id[j].mirror);
1215 flow_attr->port = 1;
1217 goto err_create_flow;
1220 /* function to create mirror rule */
1224 return &mflow->ibflow;
1228 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1229 mflow->reg_id[i].id);
1234 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1235 mflow->reg_id[j].mirror);
1240 return ERR_PTR(err);
1243 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1247 struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1248 struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1250 while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1251 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1254 if (mflow->reg_id[i].mirror) {
1255 err = __mlx4_ib_destroy_flow(mdev->dev,
1256 mflow->reg_id[i].mirror);
1267 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1270 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1271 struct mlx4_dev *dev = mdev->dev;
1272 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1273 struct mlx4_ib_steering *ib_steering = NULL;
1274 enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1275 struct mlx4_flow_reg_id reg_id;
1277 if (mdev->dev->caps.steering_mode ==
1278 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1279 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1284 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1286 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1289 pr_err("multicast attach op failed, err %d\n", err);
1294 if (mlx4_is_bonded(dev)) {
1295 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1296 (mqp->port == 1) ? 2 : 1,
1298 MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1299 prot, ®_id.mirror);
1304 err = add_gid_entry(ibqp, gid);
1309 memcpy(ib_steering->gid.raw, gid->raw, 16);
1310 ib_steering->reg_id = reg_id;
1311 mutex_lock(&mqp->mutex);
1312 list_add(&ib_steering->list, &mqp->steering_rules);
1313 mutex_unlock(&mqp->mutex);
1318 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1321 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1322 prot, reg_id.mirror);
1329 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1331 struct mlx4_ib_gid_entry *ge;
1332 struct mlx4_ib_gid_entry *tmp;
1333 struct mlx4_ib_gid_entry *ret = NULL;
1335 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1336 if (!memcmp(raw, ge->gid.raw, 16)) {
1345 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1348 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1349 struct mlx4_dev *dev = mdev->dev;
1350 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1351 struct net_device *ndev;
1352 struct mlx4_ib_gid_entry *ge;
1353 struct mlx4_flow_reg_id reg_id = {0, 0};
1354 enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1356 if (mdev->dev->caps.steering_mode ==
1357 MLX4_STEERING_MODE_DEVICE_MANAGED) {
1358 struct mlx4_ib_steering *ib_steering;
1360 mutex_lock(&mqp->mutex);
1361 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1362 if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1363 list_del(&ib_steering->list);
1367 mutex_unlock(&mqp->mutex);
1368 if (&ib_steering->list == &mqp->steering_rules) {
1369 pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1372 reg_id = ib_steering->reg_id;
1376 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1381 if (mlx4_is_bonded(dev)) {
1382 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1383 prot, reg_id.mirror);
1388 mutex_lock(&mqp->mutex);
1389 ge = find_gid_entry(mqp, gid->raw);
1391 spin_lock_bh(&mdev->iboe.lock);
1392 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1395 spin_unlock_bh(&mdev->iboe.lock);
1398 list_del(&ge->list);
1401 pr_warn("could not find mgid entry\n");
1403 mutex_unlock(&mqp->mutex);
1408 static int init_node_data(struct mlx4_ib_dev *dev)
1410 struct ib_smp *in_mad = NULL;
1411 struct ib_smp *out_mad = NULL;
1412 int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1415 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1416 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1417 if (!in_mad || !out_mad)
1420 init_query_mad(in_mad);
1421 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1422 if (mlx4_is_master(dev->dev))
1423 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1425 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1429 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1431 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1433 err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1437 dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1438 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1446 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1449 struct mlx4_ib_dev *dev =
1450 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1451 return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1454 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1457 struct mlx4_ib_dev *dev =
1458 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1459 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1460 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1461 (int) dev->dev->caps.fw_ver & 0xffff);
1464 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1467 struct mlx4_ib_dev *dev =
1468 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1469 return sprintf(buf, "%x\n", dev->dev->rev_id);
1472 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1475 struct mlx4_ib_dev *dev =
1476 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1477 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1478 dev->dev->board_id);
1481 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1482 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
1483 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1484 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1486 static struct device_attribute *mlx4_class_attributes[] = {
1493 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1494 struct net_device *dev)
1496 memcpy(eui, dev->dev_addr, 3);
1497 memcpy(eui + 5, dev->dev_addr + 3, 3);
1498 if (vlan_id < 0x1000) {
1499 eui[3] = vlan_id >> 8;
1500 eui[4] = vlan_id & 0xff;
1508 static void update_gids_task(struct work_struct *work)
1510 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1511 struct mlx4_cmd_mailbox *mailbox;
1514 struct mlx4_dev *dev = gw->dev->dev;
1515 int is_bonded = mlx4_is_bonded(dev);
1517 if (!gw->dev->ib_active)
1520 mailbox = mlx4_alloc_cmd_mailbox(dev);
1521 if (IS_ERR(mailbox)) {
1522 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1526 gids = mailbox->buf;
1527 memcpy(gids, gw->gids, sizeof gw->gids);
1529 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1530 MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1531 MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
1533 pr_warn("set port command failed\n");
1535 if ((gw->port == 1) || !is_bonded)
1536 mlx4_ib_dispatch_event(gw->dev,
1537 is_bonded ? 1 : gw->port,
1538 IB_EVENT_GID_CHANGE);
1540 mlx4_free_cmd_mailbox(dev, mailbox);
1544 static void reset_gids_task(struct work_struct *work)
1546 struct update_gid_work *gw =
1547 container_of(work, struct update_gid_work, work);
1548 struct mlx4_cmd_mailbox *mailbox;
1551 struct mlx4_dev *dev = gw->dev->dev;
1553 if (!gw->dev->ib_active)
1556 mailbox = mlx4_alloc_cmd_mailbox(dev);
1557 if (IS_ERR(mailbox)) {
1558 pr_warn("reset gid table failed\n");
1562 gids = mailbox->buf;
1563 memcpy(gids, gw->gids, sizeof(gw->gids));
1565 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1566 IB_LINK_LAYER_ETHERNET) {
1567 err = mlx4_cmd(dev, mailbox->dma,
1568 MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1569 MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1570 MLX4_CMD_TIME_CLASS_B,
1573 pr_warn("set port %d command failed\n", gw->port);
1576 mlx4_free_cmd_mailbox(dev, mailbox);
1581 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1582 union ib_gid *gid, int clear,
1585 struct update_gid_work *work;
1587 int need_update = 0;
1595 max_gids = dev->dev->caps.gid_table_len[port];
1596 for (i = 1; i < max_gids; ++i) {
1597 if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1604 dev->iboe.gid_table[port - 1][found] =
1613 !memcmp(&dev->iboe.gid_table[port - 1][i],
1614 &zgid, sizeof(*gid)))
1620 if (found == -1 && !clear && free >= 0) {
1621 dev->iboe.gid_table[port - 1][free] = *gid;
1628 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1632 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1633 INIT_WORK(&work->work, update_gids_task);
1636 queue_work(wq, &work->work);
1641 static void mlx4_make_default_gid(struct net_device *dev, union ib_gid *gid)
1643 gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1644 mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1648 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1650 struct update_gid_work *work;
1652 work = kzalloc(sizeof(*work), GFP_ATOMIC);
1656 memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1657 memset(work->gids, 0, sizeof(work->gids));
1658 INIT_WORK(&work->work, reset_gids_task);
1661 queue_work(wq, &work->work);
1665 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1666 struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1668 struct mlx4_ib_iboe *iboe;
1670 struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1671 rdma_vlan_dev_real_dev(event_netdev) :
1673 union ib_gid default_gid;
1675 mlx4_make_default_gid(real_dev, &default_gid);
1677 if (!memcmp(gid, &default_gid, sizeof(*gid)))
1680 if (event != NETDEV_DOWN && event != NETDEV_UP)
1683 if ((real_dev != event_netdev) &&
1684 (event == NETDEV_DOWN) &&
1685 rdma_link_local_addr((struct in6_addr *)gid))
1688 iboe = &ibdev->iboe;
1689 spin_lock_bh(&iboe->lock);
1691 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1692 if ((netif_is_bond_master(real_dev) &&
1693 (real_dev == iboe->masters[port - 1])) ||
1694 (!netif_is_bond_master(real_dev) &&
1695 (real_dev == iboe->netdevs[port - 1])))
1696 update_gid_table(ibdev, port, gid,
1697 event == NETDEV_DOWN, 0);
1699 spin_unlock_bh(&iboe->lock);
1704 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1705 struct mlx4_ib_dev *ibdev)
1708 struct mlx4_ib_iboe *iboe;
1709 struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1710 rdma_vlan_dev_real_dev(dev) : dev;
1712 iboe = &ibdev->iboe;
1714 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1715 if ((netif_is_bond_master(real_dev) &&
1716 (real_dev == iboe->masters[port - 1])) ||
1717 (!netif_is_bond_master(real_dev) &&
1718 (real_dev == iboe->netdevs[port - 1])))
1721 if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1727 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1730 struct mlx4_ib_dev *ibdev;
1731 struct in_ifaddr *ifa = ptr;
1733 struct net_device *event_netdev = ifa->ifa_dev->dev;
1735 ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1737 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1739 mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1743 #if IS_ENABLED(CONFIG_IPV6)
1744 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1747 struct mlx4_ib_dev *ibdev;
1748 struct inet6_ifaddr *ifa = ptr;
1749 union ib_gid *gid = (union ib_gid *)&ifa->addr;
1750 struct net_device *event_netdev = ifa->idev->dev;
1752 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1754 mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1759 #define MLX4_IB_INVALID_MAC ((u64)-1)
1760 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1761 struct net_device *dev,
1765 u64 release_mac = MLX4_IB_INVALID_MAC;
1766 struct mlx4_ib_qp *qp;
1768 read_lock(&dev_base_lock);
1769 new_smac = mlx4_mac_to_u64(dev->dev_addr);
1770 read_unlock(&dev_base_lock);
1772 atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1774 /* no need for update QP1 and mac registration in non-SRIOV */
1775 if (!mlx4_is_mfunc(ibdev->dev))
1778 mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1779 qp = ibdev->qp1_proxy[port - 1];
1783 struct mlx4_update_qp_params update_params;
1785 mutex_lock(&qp->mutex);
1786 old_smac = qp->pri.smac;
1787 if (new_smac == old_smac)
1790 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1792 if (new_smac_index < 0)
1795 update_params.smac_index = new_smac_index;
1796 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1798 release_mac = new_smac;
1801 /* if old port was zero, no mac was yet registered for this QP */
1802 if (qp->pri.smac_port)
1803 release_mac = old_smac;
1804 qp->pri.smac = new_smac;
1805 qp->pri.smac_port = port;
1806 qp->pri.smac_index = new_smac_index;
1810 if (release_mac != MLX4_IB_INVALID_MAC)
1811 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1813 mutex_unlock(&qp->mutex);
1814 mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1817 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1818 struct mlx4_ib_dev *ibdev, u8 port)
1820 struct in_device *in_dev;
1821 #if IS_ENABLED(CONFIG_IPV6)
1822 struct inet6_dev *in6_dev;
1824 struct inet6_ifaddr *ifp;
1825 union ib_gid default_gid;
1830 if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1834 in_dev = in_dev_get(dev);
1837 /*ifa->ifa_address;*/
1838 ipv6_addr_set_v4mapped(ifa->ifa_address,
1839 (struct in6_addr *)&gid);
1840 update_gid_table(ibdev, port, &gid, 0, 0);
1845 #if IS_ENABLED(CONFIG_IPV6)
1846 mlx4_make_default_gid(dev, &default_gid);
1848 in6_dev = in6_dev_get(dev);
1850 read_lock_bh(&in6_dev->lock);
1851 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1852 pgid = (union ib_gid *)&ifp->addr;
1853 if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1855 update_gid_table(ibdev, port, pgid, 0, 0);
1857 read_unlock_bh(&in6_dev->lock);
1858 in6_dev_put(in6_dev);
1863 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1864 struct net_device *dev, u8 port)
1867 mlx4_make_default_gid(dev, &gid);
1868 update_gid_table(ibdev, port, &gid, 0, 1);
1871 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1873 struct net_device *dev;
1874 struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1878 for (i = 1; i <= ibdev->num_ports; ++i) {
1879 if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1880 IB_LINK_LAYER_ETHERNET) {
1881 err = reset_gid_table(ibdev, i);
1887 read_lock(&dev_base_lock);
1888 spin_lock_bh(&iboe->lock);
1890 for_each_netdev(&init_net, dev) {
1891 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1892 /* port will be non-zero only for ETH ports */
1894 mlx4_ib_set_default_gid(ibdev, dev, port);
1895 mlx4_ib_get_dev_addr(dev, ibdev, port);
1899 spin_unlock_bh(&iboe->lock);
1900 read_unlock(&dev_base_lock);
1905 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1906 struct net_device *dev,
1907 unsigned long event)
1910 struct mlx4_ib_iboe *iboe;
1911 int update_qps_port = -1;
1914 iboe = &ibdev->iboe;
1916 spin_lock_bh(&iboe->lock);
1917 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1918 enum ib_port_state port_state = IB_PORT_NOP;
1919 struct net_device *old_master = iboe->masters[port - 1];
1920 struct net_device *curr_netdev;
1921 struct net_device *curr_master;
1923 iboe->netdevs[port - 1] =
1924 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1925 if (iboe->netdevs[port - 1])
1926 mlx4_ib_set_default_gid(ibdev,
1927 iboe->netdevs[port - 1], port);
1928 curr_netdev = iboe->netdevs[port - 1];
1930 if (iboe->netdevs[port - 1] &&
1931 netif_is_bond_slave(iboe->netdevs[port - 1])) {
1932 iboe->masters[port - 1] = netdev_master_upper_dev_get(
1933 iboe->netdevs[port - 1]);
1935 iboe->masters[port - 1] = NULL;
1937 curr_master = iboe->masters[port - 1];
1939 if (dev == iboe->netdevs[port - 1] &&
1940 (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1941 event == NETDEV_UP || event == NETDEV_CHANGE))
1942 update_qps_port = port;
1945 port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1946 IB_PORT_ACTIVE : IB_PORT_DOWN;
1947 mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1949 /* if using bonding/team and a slave port is down, we
1950 * don't want the bond IP based gids in the table since
1951 * flows that select port by gid may get the down port.
1953 if (port_state == IB_PORT_DOWN &&
1954 !mlx4_is_bonded(ibdev->dev)) {
1955 reset_gid_table(ibdev, port);
1956 mlx4_ib_set_default_gid(ibdev,
1960 /* gids from the upper dev (bond/team)
1961 * should appear in port's gid table
1963 mlx4_ib_get_dev_addr(curr_master,
1967 /* if bonding is used it is possible that we add it to
1968 * masters only after IP address is assigned to the
1969 * net bonding interface.
1971 if (curr_master && (old_master != curr_master)) {
1972 reset_gid_table(ibdev, port);
1973 mlx4_ib_set_default_gid(ibdev,
1975 mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1978 if (!curr_master && (old_master != curr_master)) {
1979 reset_gid_table(ibdev, port);
1980 mlx4_ib_set_default_gid(ibdev,
1982 mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1985 reset_gid_table(ibdev, port);
1989 spin_unlock_bh(&iboe->lock);
1991 if (update_qps_port > 0)
1992 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1995 static int mlx4_ib_netdev_event(struct notifier_block *this,
1996 unsigned long event, void *ptr)
1998 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1999 struct mlx4_ib_dev *ibdev;
2001 if (!net_eq(dev_net(dev), &init_net))
2004 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2005 mlx4_ib_scan_netdevs(ibdev, dev, event);
2010 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2016 if (mlx4_is_master(ibdev->dev)) {
2017 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2019 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2021 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2023 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2024 /* master has the identity virt2phys pkey mapping */
2025 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2026 ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2027 mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2028 ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2032 /* initialize pkey cache */
2033 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2035 i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2037 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2043 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2046 int eq_per_port = 0;
2051 /* Legacy mode or comp_pool is not large enough */
2052 if (dev->caps.comp_pool == 0 ||
2053 dev->caps.num_ports > dev->caps.comp_pool)
2056 eq_per_port = dev->caps.comp_pool / dev->caps.num_ports;
2060 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2061 added_eqs += eq_per_port;
2063 total_eqs = dev->caps.num_comp_vectors + added_eqs;
2065 ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
2066 if (!ibdev->eq_table)
2069 ibdev->eq_added = added_eqs;
2072 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
2073 for (j = 0; j < eq_per_port; j++) {
2074 snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
2075 i, j, dev->persist->pdev->bus->name);
2076 /* Set IRQ for specific name (per ring) */
2077 if (mlx4_assign_eq(dev, name, NULL,
2078 &ibdev->eq_table[eq])) {
2079 /* Use legacy (same as mlx4_en driver) */
2080 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
2081 ibdev->eq_table[eq] =
2082 (eq % dev->caps.num_comp_vectors);
2088 /* Fill the reset of the vector with legacy EQ */
2089 for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2090 ibdev->eq_table[eq++] = i;
2092 /* Advertise the new number of EQs to clients */
2093 ibdev->ib_dev.num_comp_vectors = total_eqs;
2096 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2100 /* no additional eqs were added */
2101 if (!ibdev->eq_table)
2104 /* Reset the advertised EQ number */
2105 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2107 /* Free only the added eqs */
2108 for (i = 0; i < ibdev->eq_added; i++) {
2109 /* Don't free legacy eqs if used */
2110 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2112 mlx4_release_eq(dev, ibdev->eq_table[i]);
2115 kfree(ibdev->eq_table);
2118 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2119 struct ib_port_immutable *immutable)
2121 struct ib_port_attr attr;
2124 err = mlx4_ib_query_port(ibdev, port_num, &attr);
2128 immutable->pkey_tbl_len = attr.pkey_tbl_len;
2129 immutable->gid_tbl_len = attr.gid_tbl_len;
2131 if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2132 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2134 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2139 static void *mlx4_ib_add(struct mlx4_dev *dev)
2141 struct mlx4_ib_dev *ibdev;
2145 struct mlx4_ib_iboe *iboe;
2146 int ib_num_ports = 0;
2147 int num_req_counters;
2149 pr_info_once("%s", mlx4_ib_version);
2152 mlx4_foreach_ib_transport_port(i, dev)
2155 /* No point in registering a device with no ports... */
2159 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2161 dev_err(&dev->persist->pdev->dev,
2162 "Device struct alloc failed\n");
2166 iboe = &ibdev->iboe;
2168 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2171 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2174 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2176 if (!ibdev->uar_map)
2178 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2181 ibdev->bond_next_port = 0;
2183 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2184 ibdev->ib_dev.owner = THIS_MODULE;
2185 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
2186 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
2187 ibdev->num_ports = num_ports;
2188 ibdev->ib_dev.phys_port_cnt = mlx4_is_bonded(dev) ?
2189 1 : ibdev->num_ports;
2190 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2191 ibdev->ib_dev.dma_device = &dev->persist->pdev->dev;
2193 if (dev->caps.userspace_caps)
2194 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2196 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2198 ibdev->ib_dev.uverbs_cmd_mask =
2199 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
2200 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
2201 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
2202 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
2203 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
2204 (1ull << IB_USER_VERBS_CMD_REG_MR) |
2205 (1ull << IB_USER_VERBS_CMD_REREG_MR) |
2206 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
2207 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2208 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
2209 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
2210 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
2211 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
2212 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
2213 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
2214 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
2215 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
2216 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
2217 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
2218 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
2219 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
2220 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
2221 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
2222 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2224 ibdev->ib_dev.query_device = mlx4_ib_query_device;
2225 ibdev->ib_dev.query_port = mlx4_ib_query_port;
2226 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
2227 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
2228 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
2229 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
2230 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
2231 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
2232 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
2233 ibdev->ib_dev.mmap = mlx4_ib_mmap;
2234 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
2235 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
2236 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
2237 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
2238 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
2239 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
2240 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
2241 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
2242 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
2243 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
2244 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
2245 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
2246 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
2247 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
2248 ibdev->ib_dev.post_send = mlx4_ib_post_send;
2249 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
2250 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
2251 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
2252 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
2253 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
2254 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
2255 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
2256 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
2257 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
2258 ibdev->ib_dev.rereg_user_mr = mlx4_ib_rereg_user_mr;
2259 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
2260 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2261 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2262 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
2263 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
2264 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
2265 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
2266 ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2268 if (!mlx4_is_slave(ibdev->dev)) {
2269 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
2270 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
2271 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
2272 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
2275 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2276 dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2277 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2278 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2279 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2281 ibdev->ib_dev.uverbs_cmd_mask |=
2282 (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2283 (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2286 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2287 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2288 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2289 ibdev->ib_dev.uverbs_cmd_mask |=
2290 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2291 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2294 if (check_flow_steering_support(dev)) {
2295 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2296 ibdev->ib_dev.create_flow = mlx4_ib_create_flow;
2297 ibdev->ib_dev.destroy_flow = mlx4_ib_destroy_flow;
2299 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2300 (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2301 (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2304 mlx4_ib_alloc_eqs(dev, ibdev);
2306 spin_lock_init(&iboe->lock);
2308 if (init_node_data(ibdev))
2311 num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2312 for (i = 0; i < num_req_counters; ++i) {
2313 mutex_init(&ibdev->qp1_proxy_lock[i]);
2314 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2315 IB_LINK_LAYER_ETHERNET) {
2316 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2318 ibdev->counters[i] = -1;
2320 ibdev->counters[i] = -1;
2323 if (mlx4_is_bonded(dev))
2324 for (i = 1; i < ibdev->num_ports ; ++i)
2325 ibdev->counters[i] = ibdev->counters[0];
2328 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2331 spin_lock_init(&ibdev->sm_lock);
2332 mutex_init(&ibdev->cap_mask_mutex);
2333 INIT_LIST_HEAD(&ibdev->qp_list);
2334 spin_lock_init(&ibdev->reset_flow_resource_lock);
2336 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2338 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2339 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2340 MLX4_IB_UC_STEER_QPN_ALIGN,
2341 &ibdev->steer_qpn_base, 0);
2345 ibdev->ib_uc_qpns_bitmap =
2346 kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2349 if (!ibdev->ib_uc_qpns_bitmap) {
2350 dev_err(&dev->persist->pdev->dev,
2351 "bit map alloc failed\n");
2352 goto err_steer_qp_release;
2355 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2357 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2358 dev, ibdev->steer_qpn_base,
2359 ibdev->steer_qpn_base +
2360 ibdev->steer_qpn_count - 1);
2362 goto err_steer_free_bitmap;
2365 for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2366 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2368 if (ib_register_device(&ibdev->ib_dev, NULL))
2369 goto err_steer_free_bitmap;
2371 if (mlx4_ib_mad_init(ibdev))
2374 if (mlx4_ib_init_sriov(ibdev))
2377 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2378 if (!iboe->nb.notifier_call) {
2379 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2380 err = register_netdevice_notifier(&iboe->nb);
2382 iboe->nb.notifier_call = NULL;
2386 if (!iboe->nb_inet.notifier_call) {
2387 iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2388 err = register_inetaddr_notifier(&iboe->nb_inet);
2390 iboe->nb_inet.notifier_call = NULL;
2394 #if IS_ENABLED(CONFIG_IPV6)
2395 if (!iboe->nb_inet6.notifier_call) {
2396 iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2397 err = register_inet6addr_notifier(&iboe->nb_inet6);
2399 iboe->nb_inet6.notifier_call = NULL;
2404 if (mlx4_ib_init_gid_table(ibdev))
2408 for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2409 if (device_create_file(&ibdev->ib_dev.dev,
2410 mlx4_class_attributes[j]))
2414 ibdev->ib_active = true;
2416 if (mlx4_is_mfunc(ibdev->dev))
2419 /* create paravirt contexts for any VFs which are active */
2420 if (mlx4_is_master(ibdev->dev)) {
2421 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2422 if (j == mlx4_master_func_num(ibdev->dev))
2424 if (mlx4_is_slave_active(ibdev->dev, j))
2425 do_slave_init(ibdev, j, 1);
2431 if (ibdev->iboe.nb.notifier_call) {
2432 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2433 pr_warn("failure unregistering notifier\n");
2434 ibdev->iboe.nb.notifier_call = NULL;
2436 if (ibdev->iboe.nb_inet.notifier_call) {
2437 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2438 pr_warn("failure unregistering notifier\n");
2439 ibdev->iboe.nb_inet.notifier_call = NULL;
2441 #if IS_ENABLED(CONFIG_IPV6)
2442 if (ibdev->iboe.nb_inet6.notifier_call) {
2443 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2444 pr_warn("failure unregistering notifier\n");
2445 ibdev->iboe.nb_inet6.notifier_call = NULL;
2448 flush_workqueue(wq);
2450 mlx4_ib_close_sriov(ibdev);
2453 mlx4_ib_mad_cleanup(ibdev);
2456 ib_unregister_device(&ibdev->ib_dev);
2458 err_steer_free_bitmap:
2459 kfree(ibdev->ib_uc_qpns_bitmap);
2461 err_steer_qp_release:
2462 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2463 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2464 ibdev->steer_qpn_count);
2467 if (ibdev->counters[i - 1] != -1)
2468 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2471 iounmap(ibdev->uar_map);
2474 mlx4_uar_free(dev, &ibdev->priv_uar);
2477 mlx4_pd_free(dev, ibdev->priv_pdn);
2480 ib_dealloc_device(&ibdev->ib_dev);
2485 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2489 WARN_ON(!dev->ib_uc_qpns_bitmap);
2491 offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2492 dev->steer_qpn_count,
2493 get_count_order(count));
2497 *qpn = dev->steer_qpn_base + offset;
2501 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2504 dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2507 BUG_ON(qpn < dev->steer_qpn_base);
2509 bitmap_release_region(dev->ib_uc_qpns_bitmap,
2510 qpn - dev->steer_qpn_base,
2511 get_count_order(count));
2514 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2519 struct ib_flow_attr *flow = NULL;
2520 struct ib_flow_spec_ib *ib_spec;
2523 flow_size = sizeof(struct ib_flow_attr) +
2524 sizeof(struct ib_flow_spec_ib);
2525 flow = kzalloc(flow_size, GFP_KERNEL);
2528 flow->port = mqp->port;
2529 flow->num_of_specs = 1;
2530 flow->size = flow_size;
2531 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2532 ib_spec->type = IB_FLOW_SPEC_IB;
2533 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2534 /* Add an empty rule for IB L2 */
2535 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2537 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2542 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2548 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2550 struct mlx4_ib_dev *ibdev = ibdev_ptr;
2553 ibdev->ib_active = false;
2554 flush_workqueue(wq);
2556 mlx4_ib_close_sriov(ibdev);
2557 mlx4_ib_mad_cleanup(ibdev);
2558 ib_unregister_device(&ibdev->ib_dev);
2559 if (ibdev->iboe.nb.notifier_call) {
2560 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2561 pr_warn("failure unregistering notifier\n");
2562 ibdev->iboe.nb.notifier_call = NULL;
2565 if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2566 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2567 ibdev->steer_qpn_count);
2568 kfree(ibdev->ib_uc_qpns_bitmap);
2571 if (ibdev->iboe.nb_inet.notifier_call) {
2572 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2573 pr_warn("failure unregistering notifier\n");
2574 ibdev->iboe.nb_inet.notifier_call = NULL;
2576 #if IS_ENABLED(CONFIG_IPV6)
2577 if (ibdev->iboe.nb_inet6.notifier_call) {
2578 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2579 pr_warn("failure unregistering notifier\n");
2580 ibdev->iboe.nb_inet6.notifier_call = NULL;
2584 iounmap(ibdev->uar_map);
2585 for (p = 0; p < ibdev->num_ports; ++p)
2586 if (ibdev->counters[p] != -1)
2587 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2588 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2589 mlx4_CLOSE_PORT(dev, p);
2591 mlx4_ib_free_eqs(dev, ibdev);
2593 mlx4_uar_free(dev, &ibdev->priv_uar);
2594 mlx4_pd_free(dev, ibdev->priv_pdn);
2595 ib_dealloc_device(&ibdev->ib_dev);
2598 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2600 struct mlx4_ib_demux_work **dm = NULL;
2601 struct mlx4_dev *dev = ibdev->dev;
2603 unsigned long flags;
2604 struct mlx4_active_ports actv_ports;
2606 unsigned int first_port;
2608 if (!mlx4_is_master(dev))
2611 actv_ports = mlx4_get_active_ports(dev, slave);
2612 ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2613 first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2615 dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2617 pr_err("failed to allocate memory for tunneling qp update\n");
2621 for (i = 0; i < ports; i++) {
2622 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2624 pr_err("failed to allocate memory for tunneling qp update work struct\n");
2625 for (i = 0; i < dev->caps.num_ports; i++) {
2632 /* initialize or tear down tunnel QPs for the slave */
2633 for (i = 0; i < ports; i++) {
2634 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2635 dm[i]->port = first_port + i + 1;
2636 dm[i]->slave = slave;
2637 dm[i]->do_init = do_init;
2639 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2640 if (!ibdev->sriov.is_going_down)
2641 queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2642 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2649 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2651 struct mlx4_ib_qp *mqp;
2652 unsigned long flags_qp;
2653 unsigned long flags_cq;
2654 struct mlx4_ib_cq *send_mcq, *recv_mcq;
2655 struct list_head cq_notify_list;
2656 struct mlx4_cq *mcq;
2657 unsigned long flags;
2659 pr_warn("mlx4_ib_handle_catas_error was started\n");
2660 INIT_LIST_HEAD(&cq_notify_list);
2662 /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2663 spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2665 list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2666 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2667 if (mqp->sq.tail != mqp->sq.head) {
2668 send_mcq = to_mcq(mqp->ibqp.send_cq);
2669 spin_lock_irqsave(&send_mcq->lock, flags_cq);
2670 if (send_mcq->mcq.comp &&
2671 mqp->ibqp.send_cq->comp_handler) {
2672 if (!send_mcq->mcq.reset_notify_added) {
2673 send_mcq->mcq.reset_notify_added = 1;
2674 list_add_tail(&send_mcq->mcq.reset_notify,
2678 spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2680 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2681 /* Now, handle the QP's receive queue */
2682 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2683 /* no handling is needed for SRQ */
2684 if (!mqp->ibqp.srq) {
2685 if (mqp->rq.tail != mqp->rq.head) {
2686 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2687 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2688 if (recv_mcq->mcq.comp &&
2689 mqp->ibqp.recv_cq->comp_handler) {
2690 if (!recv_mcq->mcq.reset_notify_added) {
2691 recv_mcq->mcq.reset_notify_added = 1;
2692 list_add_tail(&recv_mcq->mcq.reset_notify,
2696 spin_unlock_irqrestore(&recv_mcq->lock,
2700 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2703 list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2706 spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2707 pr_warn("mlx4_ib_handle_catas_error ended\n");
2710 static void handle_bonded_port_state_event(struct work_struct *work)
2712 struct ib_event_work *ew =
2713 container_of(work, struct ib_event_work, work);
2714 struct mlx4_ib_dev *ibdev = ew->ib_dev;
2715 enum ib_port_state bonded_port_state = IB_PORT_NOP;
2717 struct ib_event ibev;
2720 spin_lock_bh(&ibdev->iboe.lock);
2721 for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2722 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2723 enum ib_port_state curr_port_state;
2729 (netif_running(curr_netdev) &&
2730 netif_carrier_ok(curr_netdev)) ?
2731 IB_PORT_ACTIVE : IB_PORT_DOWN;
2733 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2734 curr_port_state : IB_PORT_ACTIVE;
2736 spin_unlock_bh(&ibdev->iboe.lock);
2738 ibev.device = &ibdev->ib_dev;
2739 ibev.element.port_num = 1;
2740 ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2741 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2743 ib_dispatch_event(&ibev);
2746 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2747 enum mlx4_dev_event event, unsigned long param)
2749 struct ib_event ibev;
2750 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2751 struct mlx4_eqe *eqe = NULL;
2752 struct ib_event_work *ew;
2755 if (mlx4_is_bonded(dev) &&
2756 ((event == MLX4_DEV_EVENT_PORT_UP) ||
2757 (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2758 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2761 INIT_WORK(&ew->work, handle_bonded_port_state_event);
2763 queue_work(wq, &ew->work);
2767 if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2768 eqe = (struct mlx4_eqe *)param;
2773 case MLX4_DEV_EVENT_PORT_UP:
2774 if (p > ibdev->num_ports)
2776 if (mlx4_is_master(dev) &&
2777 rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2778 IB_LINK_LAYER_INFINIBAND) {
2779 mlx4_ib_invalidate_all_guid_record(ibdev, p);
2781 ibev.event = IB_EVENT_PORT_ACTIVE;
2784 case MLX4_DEV_EVENT_PORT_DOWN:
2785 if (p > ibdev->num_ports)
2787 ibev.event = IB_EVENT_PORT_ERR;
2790 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2791 ibdev->ib_active = false;
2792 ibev.event = IB_EVENT_DEVICE_FATAL;
2793 mlx4_ib_handle_catas_error(ibdev);
2796 case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2797 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2799 pr_err("failed to allocate memory for events work\n");
2803 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2804 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2806 /* need to queue only for port owner, which uses GEN_EQE */
2807 if (mlx4_is_master(dev))
2808 queue_work(wq, &ew->work);
2810 handle_port_mgmt_change_event(&ew->work);
2813 case MLX4_DEV_EVENT_SLAVE_INIT:
2814 /* here, p is the slave id */
2815 do_slave_init(ibdev, p, 1);
2816 if (mlx4_is_master(dev)) {
2819 for (i = 1; i <= ibdev->num_ports; i++) {
2820 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2821 == IB_LINK_LAYER_INFINIBAND)
2822 mlx4_ib_slave_alias_guid_event(ibdev,
2829 case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2830 if (mlx4_is_master(dev)) {
2833 for (i = 1; i <= ibdev->num_ports; i++) {
2834 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2835 == IB_LINK_LAYER_INFINIBAND)
2836 mlx4_ib_slave_alias_guid_event(ibdev,
2841 /* here, p is the slave id */
2842 do_slave_init(ibdev, p, 0);
2849 ibev.device = ibdev_ptr;
2850 ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2852 ib_dispatch_event(&ibev);
2855 static struct mlx4_interface mlx4_ib_interface = {
2857 .remove = mlx4_ib_remove,
2858 .event = mlx4_ib_event,
2859 .protocol = MLX4_PROT_IB_IPV6,
2860 .flags = MLX4_INTFF_BONDING
2863 static int __init mlx4_ib_init(void)
2867 wq = create_singlethread_workqueue("mlx4_ib");
2871 err = mlx4_ib_mcg_init();
2875 err = mlx4_register_interface(&mlx4_ib_interface);
2882 mlx4_ib_mcg_destroy();
2885 destroy_workqueue(wq);
2889 static void __exit mlx4_ib_cleanup(void)
2891 mlx4_unregister_interface(&mlx4_ib_interface);
2892 mlx4_ib_mcg_destroy();
2893 destroy_workqueue(wq);
2896 module_init(mlx4_ib_init);
2897 module_exit(mlx4_ib_cleanup);