9ecfa0d3ff22ccc3a03ef968cd0b03f252d77dfc
[firefly-linux-kernel-4.4.55.git] / drivers / infiniband / hw / mlx4 / main.c
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <net/ipv6.h>
43 #include <net/addrconf.h>
44
45 #include <rdma/ib_smi.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_addr.h>
48 #include <rdma/ib_cache.h>
49
50 #include <net/bonding.h>
51
52 #include <linux/mlx4/driver.h>
53 #include <linux/mlx4/cmd.h>
54 #include <linux/mlx4/qp.h>
55
56 #include "mlx4_ib.h"
57 #include "user.h"
58
59 #define DRV_NAME        MLX4_IB_DRV_NAME
60 #define DRV_VERSION     "2.2-1"
61 #define DRV_RELDATE     "Feb 2014"
62
63 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
64 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
65 #define MLX4_IB_CARD_REV_A0   0xA0
66
67 MODULE_AUTHOR("Roland Dreier");
68 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
69 MODULE_LICENSE("Dual BSD/GPL");
70 MODULE_VERSION(DRV_VERSION);
71
72 int mlx4_ib_sm_guid_assign = 0;
73 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
74 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
75
76 static const char mlx4_ib_version[] =
77         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
78         DRV_VERSION " (" DRV_RELDATE ")\n";
79
80 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
81
82 static struct workqueue_struct *wq;
83
84 static void init_query_mad(struct ib_smp *mad)
85 {
86         mad->base_version  = 1;
87         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
88         mad->class_version = 1;
89         mad->method        = IB_MGMT_METHOD_GET;
90 }
91
92 static int check_flow_steering_support(struct mlx4_dev *dev)
93 {
94         int eth_num_ports = 0;
95         int ib_num_ports = 0;
96
97         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
98
99         if (dmfs) {
100                 int i;
101                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
102                         eth_num_ports++;
103                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
104                         ib_num_ports++;
105                 dmfs &= (!ib_num_ports ||
106                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
107                         (!eth_num_ports ||
108                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
109                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
110                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
111                         dmfs = 0;
112                 }
113         }
114         return dmfs;
115 }
116
117 static int num_ib_ports(struct mlx4_dev *dev)
118 {
119         int ib_ports = 0;
120         int i;
121
122         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
123                 ib_ports++;
124
125         return ib_ports;
126 }
127
128 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
129 {
130         struct mlx4_ib_dev *ibdev = to_mdev(device);
131         struct net_device *dev;
132
133         rcu_read_lock();
134         dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
135
136         if (dev) {
137                 if (mlx4_is_bonded(ibdev->dev)) {
138                         struct net_device *upper = NULL;
139
140                         upper = netdev_master_upper_dev_get_rcu(dev);
141                         if (upper) {
142                                 struct net_device *active;
143
144                                 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
145                                 if (active)
146                                         dev = active;
147                         }
148                 }
149         }
150         if (dev)
151                 dev_hold(dev);
152
153         rcu_read_unlock();
154         return dev;
155 }
156
157 static int mlx4_ib_update_gids(struct gid_entry *gids,
158                                struct mlx4_ib_dev *ibdev,
159                                u8 port_num)
160 {
161         struct mlx4_cmd_mailbox *mailbox;
162         int err;
163         struct mlx4_dev *dev = ibdev->dev;
164         int i;
165         union ib_gid *gid_tbl;
166
167         mailbox = mlx4_alloc_cmd_mailbox(dev);
168         if (IS_ERR(mailbox))
169                 return -ENOMEM;
170
171         gid_tbl = mailbox->buf;
172
173         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
174                 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
175
176         err = mlx4_cmd(dev, mailbox->dma,
177                        MLX4_SET_PORT_GID_TABLE << 8 | port_num,
178                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
179                        MLX4_CMD_WRAPPED);
180         if (mlx4_is_bonded(dev))
181                 err += mlx4_cmd(dev, mailbox->dma,
182                                 MLX4_SET_PORT_GID_TABLE << 8 | 2,
183                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
184                                 MLX4_CMD_WRAPPED);
185
186         mlx4_free_cmd_mailbox(dev, mailbox);
187         return err;
188 }
189
190 static int mlx4_ib_add_gid(struct ib_device *device,
191                            u8 port_num,
192                            unsigned int index,
193                            const union ib_gid *gid,
194                            const struct ib_gid_attr *attr,
195                            void **context)
196 {
197         struct mlx4_ib_dev *ibdev = to_mdev(device);
198         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
199         struct mlx4_port_gid_table   *port_gid_table;
200         int free = -1, found = -1;
201         int ret = 0;
202         int hw_update = 0;
203         int i;
204         struct gid_entry *gids = NULL;
205
206         if (!rdma_cap_roce_gid_table(device, port_num))
207                 return -EINVAL;
208
209         if (port_num > MLX4_MAX_PORTS)
210                 return -EINVAL;
211
212         if (!context)
213                 return -EINVAL;
214
215         port_gid_table = &iboe->gids[port_num - 1];
216         spin_lock_bh(&iboe->lock);
217         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
218                 if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid))) {
219                         found = i;
220                         break;
221                 }
222                 if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
223                         free = i; /* HW has space */
224         }
225
226         if (found < 0) {
227                 if (free < 0) {
228                         ret = -ENOSPC;
229                 } else {
230                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
231                         if (!port_gid_table->gids[free].ctx) {
232                                 ret = -ENOMEM;
233                         } else {
234                                 *context = port_gid_table->gids[free].ctx;
235                                 memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
236                                 port_gid_table->gids[free].ctx->real_index = free;
237                                 port_gid_table->gids[free].ctx->refcount = 1;
238                                 hw_update = 1;
239                         }
240                 }
241         } else {
242                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
243                 *context = ctx;
244                 ctx->refcount++;
245         }
246         if (!ret && hw_update) {
247                 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
248                 if (!gids) {
249                         ret = -ENOMEM;
250                 } else {
251                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
252                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
253                 }
254         }
255         spin_unlock_bh(&iboe->lock);
256
257         if (!ret && hw_update) {
258                 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
259                 kfree(gids);
260         }
261
262         return ret;
263 }
264
265 static int mlx4_ib_del_gid(struct ib_device *device,
266                            u8 port_num,
267                            unsigned int index,
268                            void **context)
269 {
270         struct gid_cache_context *ctx = *context;
271         struct mlx4_ib_dev *ibdev = to_mdev(device);
272         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
273         struct mlx4_port_gid_table   *port_gid_table;
274         int ret = 0;
275         int hw_update = 0;
276         struct gid_entry *gids = NULL;
277
278         if (!rdma_cap_roce_gid_table(device, port_num))
279                 return -EINVAL;
280
281         if (port_num > MLX4_MAX_PORTS)
282                 return -EINVAL;
283
284         port_gid_table = &iboe->gids[port_num - 1];
285         spin_lock_bh(&iboe->lock);
286         if (ctx) {
287                 ctx->refcount--;
288                 if (!ctx->refcount) {
289                         unsigned int real_index = ctx->real_index;
290
291                         memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
292                         kfree(port_gid_table->gids[real_index].ctx);
293                         port_gid_table->gids[real_index].ctx = NULL;
294                         hw_update = 1;
295                 }
296         }
297         if (!ret && hw_update) {
298                 int i;
299
300                 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
301                 if (!gids) {
302                         ret = -ENOMEM;
303                 } else {
304                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
305                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
306                 }
307         }
308         spin_unlock_bh(&iboe->lock);
309
310         if (!ret && hw_update) {
311                 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
312                 kfree(gids);
313         }
314         return ret;
315 }
316
317 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
318                                     u8 port_num, int index)
319 {
320         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
321         struct gid_cache_context *ctx = NULL;
322         union ib_gid gid;
323         struct mlx4_port_gid_table   *port_gid_table;
324         int real_index = -EINVAL;
325         int i;
326         int ret;
327         unsigned long flags;
328
329         if (port_num > MLX4_MAX_PORTS)
330                 return -EINVAL;
331
332         if (mlx4_is_bonded(ibdev->dev))
333                 port_num = 1;
334
335         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
336                 return index;
337
338         ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, NULL);
339         if (ret)
340                 return ret;
341
342         if (!memcmp(&gid, &zgid, sizeof(gid)))
343                 return -EINVAL;
344
345         spin_lock_irqsave(&iboe->lock, flags);
346         port_gid_table = &iboe->gids[port_num - 1];
347
348         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
349                 if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid))) {
350                         ctx = port_gid_table->gids[i].ctx;
351                         break;
352                 }
353         if (ctx)
354                 real_index = ctx->real_index;
355         spin_unlock_irqrestore(&iboe->lock, flags);
356         return real_index;
357 }
358
359 static int mlx4_ib_query_device(struct ib_device *ibdev,
360                                 struct ib_device_attr *props,
361                                 struct ib_udata *uhw)
362 {
363         struct mlx4_ib_dev *dev = to_mdev(ibdev);
364         struct ib_smp *in_mad  = NULL;
365         struct ib_smp *out_mad = NULL;
366         int err = -ENOMEM;
367         int have_ib_ports;
368         struct mlx4_uverbs_ex_query_device cmd;
369         struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
370         struct mlx4_clock_params clock_params;
371
372         if (uhw->inlen) {
373                 if (uhw->inlen < sizeof(cmd))
374                         return -EINVAL;
375
376                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
377                 if (err)
378                         return err;
379
380                 if (cmd.comp_mask)
381                         return -EINVAL;
382
383                 if (cmd.reserved)
384                         return -EINVAL;
385         }
386
387         resp.response_length = offsetof(typeof(resp), response_length) +
388                 sizeof(resp.response_length);
389         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
390         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
391         if (!in_mad || !out_mad)
392                 goto out;
393
394         init_query_mad(in_mad);
395         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
396
397         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
398                            1, NULL, NULL, in_mad, out_mad);
399         if (err)
400                 goto out;
401
402         memset(props, 0, sizeof *props);
403
404         have_ib_ports = num_ib_ports(dev->dev);
405
406         props->fw_ver = dev->dev->caps.fw_ver;
407         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
408                 IB_DEVICE_PORT_ACTIVE_EVENT             |
409                 IB_DEVICE_SYS_IMAGE_GUID                |
410                 IB_DEVICE_RC_RNR_NAK_GEN                |
411                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
412         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
413                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
414         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
415                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
416         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
417                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
418         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
419                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
420         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
421                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
422         if (dev->dev->caps.max_gso_sz &&
423             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
424             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
425                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
426         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
427                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
428         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
429             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
430             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
431                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
432         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
433                 props->device_cap_flags |= IB_DEVICE_XRC;
434         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
435                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
436         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
437                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
438                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
439                 else
440                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
441         if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
442                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
443         }
444
445         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
446
447         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
448                 0xffffff;
449         props->vendor_part_id      = dev->dev->persist->pdev->device;
450         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
451         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
452
453         props->max_mr_size         = ~0ull;
454         props->page_size_cap       = dev->dev->caps.page_size_cap;
455         props->max_qp              = dev->dev->quotas.qp;
456         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
457         props->max_sge             = min(dev->dev->caps.max_sq_sg,
458                                          dev->dev->caps.max_rq_sg);
459         props->max_sge_rd          = MLX4_MAX_SGE_RD;
460         props->max_cq              = dev->dev->quotas.cq;
461         props->max_cqe             = dev->dev->caps.max_cqes;
462         props->max_mr              = dev->dev->quotas.mpt;
463         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
464         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
465         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
466         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
467         props->max_srq             = dev->dev->quotas.srq;
468         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
469         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
470         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
471         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
472         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
473                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
474         props->masked_atomic_cap   = props->atomic_cap;
475         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
476         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
477         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
478         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
479                                            props->max_mcast_grp;
480         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
481         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
482         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
483
484         if (!mlx4_is_slave(dev->dev))
485                 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
486
487         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
488                 resp.response_length += sizeof(resp.hca_core_clock_offset);
489                 if (!err && !mlx4_is_slave(dev->dev)) {
490                         resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
491                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
492                 }
493         }
494
495         if (uhw->outlen) {
496                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
497                 if (err)
498                         goto out;
499         }
500 out:
501         kfree(in_mad);
502         kfree(out_mad);
503
504         return err;
505 }
506
507 static enum rdma_link_layer
508 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
509 {
510         struct mlx4_dev *dev = to_mdev(device)->dev;
511
512         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
513                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
514 }
515
516 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
517                               struct ib_port_attr *props, int netw_view)
518 {
519         struct ib_smp *in_mad  = NULL;
520         struct ib_smp *out_mad = NULL;
521         int ext_active_speed;
522         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
523         int err = -ENOMEM;
524
525         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
526         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
527         if (!in_mad || !out_mad)
528                 goto out;
529
530         init_query_mad(in_mad);
531         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
532         in_mad->attr_mod = cpu_to_be32(port);
533
534         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
535                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
536
537         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
538                                 in_mad, out_mad);
539         if (err)
540                 goto out;
541
542
543         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
544         props->lmc              = out_mad->data[34] & 0x7;
545         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
546         props->sm_sl            = out_mad->data[36] & 0xf;
547         props->state            = out_mad->data[32] & 0xf;
548         props->phys_state       = out_mad->data[33] >> 4;
549         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
550         if (netw_view)
551                 props->gid_tbl_len = out_mad->data[50];
552         else
553                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
554         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
555         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
556         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
557         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
558         props->active_width     = out_mad->data[31] & 0xf;
559         props->active_speed     = out_mad->data[35] >> 4;
560         props->max_mtu          = out_mad->data[41] & 0xf;
561         props->active_mtu       = out_mad->data[36] >> 4;
562         props->subnet_timeout   = out_mad->data[51] & 0x1f;
563         props->max_vl_num       = out_mad->data[37] >> 4;
564         props->init_type_reply  = out_mad->data[41] >> 4;
565
566         /* Check if extended speeds (EDR/FDR/...) are supported */
567         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
568                 ext_active_speed = out_mad->data[62] >> 4;
569
570                 switch (ext_active_speed) {
571                 case 1:
572                         props->active_speed = IB_SPEED_FDR;
573                         break;
574                 case 2:
575                         props->active_speed = IB_SPEED_EDR;
576                         break;
577                 }
578         }
579
580         /* If reported active speed is QDR, check if is FDR-10 */
581         if (props->active_speed == IB_SPEED_QDR) {
582                 init_query_mad(in_mad);
583                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
584                 in_mad->attr_mod = cpu_to_be32(port);
585
586                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
587                                    NULL, NULL, in_mad, out_mad);
588                 if (err)
589                         goto out;
590
591                 /* Checking LinkSpeedActive for FDR-10 */
592                 if (out_mad->data[15] & 0x1)
593                         props->active_speed = IB_SPEED_FDR10;
594         }
595
596         /* Avoid wrong speed value returned by FW if the IB link is down. */
597         if (props->state == IB_PORT_DOWN)
598                  props->active_speed = IB_SPEED_SDR;
599
600 out:
601         kfree(in_mad);
602         kfree(out_mad);
603         return err;
604 }
605
606 static u8 state_to_phys_state(enum ib_port_state state)
607 {
608         return state == IB_PORT_ACTIVE ? 5 : 3;
609 }
610
611 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
612                                struct ib_port_attr *props, int netw_view)
613 {
614
615         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
616         struct mlx4_ib_iboe *iboe = &mdev->iboe;
617         struct net_device *ndev;
618         enum ib_mtu tmp;
619         struct mlx4_cmd_mailbox *mailbox;
620         int err = 0;
621         int is_bonded = mlx4_is_bonded(mdev->dev);
622
623         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
624         if (IS_ERR(mailbox))
625                 return PTR_ERR(mailbox);
626
627         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
628                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
629                            MLX4_CMD_WRAPPED);
630         if (err)
631                 goto out;
632
633         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
634                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
635                                            IB_WIDTH_4X : IB_WIDTH_1X;
636         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
637                                            IB_SPEED_FDR : IB_SPEED_QDR;
638         props->port_cap_flags   = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
639         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
640         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
641         props->pkey_tbl_len     = 1;
642         props->max_mtu          = IB_MTU_4096;
643         props->max_vl_num       = 2;
644         props->state            = IB_PORT_DOWN;
645         props->phys_state       = state_to_phys_state(props->state);
646         props->active_mtu       = IB_MTU_256;
647         spin_lock_bh(&iboe->lock);
648         ndev = iboe->netdevs[port - 1];
649         if (ndev && is_bonded) {
650                 rcu_read_lock(); /* required to get upper dev */
651                 ndev = netdev_master_upper_dev_get_rcu(ndev);
652                 rcu_read_unlock();
653         }
654         if (!ndev)
655                 goto out_unlock;
656
657         tmp = iboe_get_mtu(ndev->mtu);
658         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
659
660         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
661                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
662         props->phys_state       = state_to_phys_state(props->state);
663 out_unlock:
664         spin_unlock_bh(&iboe->lock);
665 out:
666         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
667         return err;
668 }
669
670 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
671                          struct ib_port_attr *props, int netw_view)
672 {
673         int err;
674
675         memset(props, 0, sizeof *props);
676
677         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
678                 ib_link_query_port(ibdev, port, props, netw_view) :
679                                 eth_link_query_port(ibdev, port, props, netw_view);
680
681         return err;
682 }
683
684 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
685                               struct ib_port_attr *props)
686 {
687         /* returns host view */
688         return __mlx4_ib_query_port(ibdev, port, props, 0);
689 }
690
691 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
692                         union ib_gid *gid, int netw_view)
693 {
694         struct ib_smp *in_mad  = NULL;
695         struct ib_smp *out_mad = NULL;
696         int err = -ENOMEM;
697         struct mlx4_ib_dev *dev = to_mdev(ibdev);
698         int clear = 0;
699         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
700
701         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
702         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
703         if (!in_mad || !out_mad)
704                 goto out;
705
706         init_query_mad(in_mad);
707         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
708         in_mad->attr_mod = cpu_to_be32(port);
709
710         if (mlx4_is_mfunc(dev->dev) && netw_view)
711                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
712
713         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
714         if (err)
715                 goto out;
716
717         memcpy(gid->raw, out_mad->data + 8, 8);
718
719         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
720                 if (index) {
721                         /* For any index > 0, return the null guid */
722                         err = 0;
723                         clear = 1;
724                         goto out;
725                 }
726         }
727
728         init_query_mad(in_mad);
729         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
730         in_mad->attr_mod = cpu_to_be32(index / 8);
731
732         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
733                            NULL, NULL, in_mad, out_mad);
734         if (err)
735                 goto out;
736
737         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
738
739 out:
740         if (clear)
741                 memset(gid->raw + 8, 0, 8);
742         kfree(in_mad);
743         kfree(out_mad);
744         return err;
745 }
746
747 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
748                              union ib_gid *gid)
749 {
750         int ret;
751
752         if (rdma_protocol_ib(ibdev, port))
753                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
754
755         if (!rdma_protocol_roce(ibdev, port))
756                 return -ENODEV;
757
758         if (!rdma_cap_roce_gid_table(ibdev, port))
759                 return -ENODEV;
760
761         ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
762         if (ret == -EAGAIN) {
763                 memcpy(gid, &zgid, sizeof(*gid));
764                 return 0;
765         }
766
767         return ret;
768 }
769
770 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
771                          u16 *pkey, int netw_view)
772 {
773         struct ib_smp *in_mad  = NULL;
774         struct ib_smp *out_mad = NULL;
775         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
776         int err = -ENOMEM;
777
778         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
779         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
780         if (!in_mad || !out_mad)
781                 goto out;
782
783         init_query_mad(in_mad);
784         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
785         in_mad->attr_mod = cpu_to_be32(index / 32);
786
787         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
788                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
789
790         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
791                            in_mad, out_mad);
792         if (err)
793                 goto out;
794
795         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
796
797 out:
798         kfree(in_mad);
799         kfree(out_mad);
800         return err;
801 }
802
803 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
804 {
805         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
806 }
807
808 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
809                                  struct ib_device_modify *props)
810 {
811         struct mlx4_cmd_mailbox *mailbox;
812         unsigned long flags;
813
814         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
815                 return -EOPNOTSUPP;
816
817         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
818                 return 0;
819
820         if (mlx4_is_slave(to_mdev(ibdev)->dev))
821                 return -EOPNOTSUPP;
822
823         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
824         memcpy(ibdev->node_desc, props->node_desc, 64);
825         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
826
827         /*
828          * If possible, pass node desc to FW, so it can generate
829          * a 144 trap.  If cmd fails, just ignore.
830          */
831         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
832         if (IS_ERR(mailbox))
833                 return 0;
834
835         memcpy(mailbox->buf, props->node_desc, 64);
836         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
837                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
838
839         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
840
841         return 0;
842 }
843
844 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
845                             u32 cap_mask)
846 {
847         struct mlx4_cmd_mailbox *mailbox;
848         int err;
849
850         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
851         if (IS_ERR(mailbox))
852                 return PTR_ERR(mailbox);
853
854         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
855                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
856                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
857         } else {
858                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
859                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
860         }
861
862         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
863                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
864                        MLX4_CMD_WRAPPED);
865
866         mlx4_free_cmd_mailbox(dev->dev, mailbox);
867         return err;
868 }
869
870 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
871                                struct ib_port_modify *props)
872 {
873         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
874         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
875         struct ib_port_attr attr;
876         u32 cap_mask;
877         int err;
878
879         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
880          * of whether port link layer is ETH or IB. For ETH ports, qkey
881          * violations and port capabilities are not meaningful.
882          */
883         if (is_eth)
884                 return 0;
885
886         mutex_lock(&mdev->cap_mask_mutex);
887
888         err = mlx4_ib_query_port(ibdev, port, &attr);
889         if (err)
890                 goto out;
891
892         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
893                 ~props->clr_port_cap_mask;
894
895         err = mlx4_ib_SET_PORT(mdev, port,
896                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
897                                cap_mask);
898
899 out:
900         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
901         return err;
902 }
903
904 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
905                                                   struct ib_udata *udata)
906 {
907         struct mlx4_ib_dev *dev = to_mdev(ibdev);
908         struct mlx4_ib_ucontext *context;
909         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
910         struct mlx4_ib_alloc_ucontext_resp resp;
911         int err;
912
913         if (!dev->ib_active)
914                 return ERR_PTR(-EAGAIN);
915
916         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
917                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
918                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
919                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
920         } else {
921                 resp.dev_caps         = dev->dev->caps.userspace_caps;
922                 resp.qp_tab_size      = dev->dev->caps.num_qps;
923                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
924                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
925                 resp.cqe_size         = dev->dev->caps.cqe_size;
926         }
927
928         context = kzalloc(sizeof(*context), GFP_KERNEL);
929         if (!context)
930                 return ERR_PTR(-ENOMEM);
931
932         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
933         if (err) {
934                 kfree(context);
935                 return ERR_PTR(err);
936         }
937
938         INIT_LIST_HEAD(&context->db_page_list);
939         mutex_init(&context->db_page_mutex);
940
941         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
942                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
943         else
944                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
945
946         if (err) {
947                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
948                 kfree(context);
949                 return ERR_PTR(-EFAULT);
950         }
951
952         return &context->ibucontext;
953 }
954
955 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
956 {
957         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
958
959         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
960         kfree(context);
961
962         return 0;
963 }
964
965 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
966 {
967         /* vma_open is called when a new VMA is created on top of our VMA.
968          * This is done through either mremap flow or split_vma (usually due
969          * to mlock, madvise, munmap, etc.). We do not support a clone of the
970          * vma, as this VMA is strongly hardware related. Therefore we set the
971          * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
972          * calling us again and trying to do incorrect actions. We assume that
973          * the original vma size is exactly a single page that there will be no
974          * "splitting" operations on.
975          */
976         area->vm_ops = NULL;
977 }
978
979 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
980 {
981         struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
982
983         /* It's guaranteed that all VMAs opened on a FD are closed before the
984          * file itself is closed, therefore no sync is needed with the regular
985          * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
986          * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
987          * The close operation is usually called under mm->mmap_sem except when
988          * process is exiting.  The exiting case is handled explicitly as part
989          * of mlx4_ib_disassociate_ucontext.
990          */
991         mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
992                                 area->vm_private_data;
993
994         /* set the vma context pointer to null in the mlx4_ib driver's private
995          * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
996          */
997         mlx4_ib_vma_priv_data->vma = NULL;
998 }
999
1000 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1001         .open = mlx4_ib_vma_open,
1002         .close = mlx4_ib_vma_close
1003 };
1004
1005 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1006 {
1007         int i;
1008         int ret = 0;
1009         struct vm_area_struct *vma;
1010         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1011         struct task_struct *owning_process  = NULL;
1012         struct mm_struct   *owning_mm       = NULL;
1013
1014         owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1015         if (!owning_process)
1016                 return;
1017
1018         owning_mm = get_task_mm(owning_process);
1019         if (!owning_mm) {
1020                 pr_info("no mm, disassociate ucontext is pending task termination\n");
1021                 while (1) {
1022                         /* make sure that task is dead before returning, it may
1023                          * prevent a rare case of module down in parallel to a
1024                          * call to mlx4_ib_vma_close.
1025                          */
1026                         put_task_struct(owning_process);
1027                         msleep(1);
1028                         owning_process = get_pid_task(ibcontext->tgid,
1029                                                       PIDTYPE_PID);
1030                         if (!owning_process ||
1031                             owning_process->state == TASK_DEAD) {
1032                                 pr_info("disassociate ucontext done, task was terminated\n");
1033                                 /* in case task was dead need to release the task struct */
1034                                 if (owning_process)
1035                                         put_task_struct(owning_process);
1036                                 return;
1037                         }
1038                 }
1039         }
1040
1041         /* need to protect from a race on closing the vma as part of
1042          * mlx4_ib_vma_close().
1043          */
1044         down_read(&owning_mm->mmap_sem);
1045         for (i = 0; i < HW_BAR_COUNT; i++) {
1046                 vma = context->hw_bar_info[i].vma;
1047                 if (!vma)
1048                         continue;
1049
1050                 ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1051                                    context->hw_bar_info[i].vma->vm_start,
1052                                    PAGE_SIZE);
1053                 if (ret) {
1054                         pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1055                         BUG_ON(1);
1056                 }
1057
1058                 /* context going to be destroyed, should not access ops any more */
1059                 context->hw_bar_info[i].vma->vm_ops = NULL;
1060         }
1061
1062         up_read(&owning_mm->mmap_sem);
1063         mmput(owning_mm);
1064         put_task_struct(owning_process);
1065 }
1066
1067 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1068                                  struct mlx4_ib_vma_private_data *vma_private_data)
1069 {
1070         vma_private_data->vma = vma;
1071         vma->vm_private_data = vma_private_data;
1072         vma->vm_ops =  &mlx4_ib_vm_ops;
1073 }
1074
1075 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1076 {
1077         struct mlx4_ib_dev *dev = to_mdev(context->device);
1078         struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1079
1080         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1081                 return -EINVAL;
1082
1083         if (vma->vm_pgoff == 0) {
1084                 /* We prevent double mmaping on same context */
1085                 if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1086                         return -EINVAL;
1087
1088                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1089
1090                 if (io_remap_pfn_range(vma, vma->vm_start,
1091                                        to_mucontext(context)->uar.pfn,
1092                                        PAGE_SIZE, vma->vm_page_prot))
1093                         return -EAGAIN;
1094
1095                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1096
1097         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1098                 /* We prevent double mmaping on same context */
1099                 if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1100                         return -EINVAL;
1101
1102                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1103
1104                 if (io_remap_pfn_range(vma, vma->vm_start,
1105                                        to_mucontext(context)->uar.pfn +
1106                                        dev->dev->caps.num_uars,
1107                                        PAGE_SIZE, vma->vm_page_prot))
1108                         return -EAGAIN;
1109
1110                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1111
1112         } else if (vma->vm_pgoff == 3) {
1113                 struct mlx4_clock_params params;
1114                 int ret;
1115
1116                 /* We prevent double mmaping on same context */
1117                 if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1118                         return -EINVAL;
1119
1120                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1121
1122                 if (ret)
1123                         return ret;
1124
1125                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1126                 if (io_remap_pfn_range(vma, vma->vm_start,
1127                                        (pci_resource_start(dev->dev->persist->pdev,
1128                                                            params.bar) +
1129                                         params.offset)
1130                                        >> PAGE_SHIFT,
1131                                        PAGE_SIZE, vma->vm_page_prot))
1132                         return -EAGAIN;
1133
1134                 mlx4_ib_set_vma_data(vma,
1135                                      &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1136         } else {
1137                 return -EINVAL;
1138         }
1139
1140         return 0;
1141 }
1142
1143 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1144                                       struct ib_ucontext *context,
1145                                       struct ib_udata *udata)
1146 {
1147         struct mlx4_ib_pd *pd;
1148         int err;
1149
1150         pd = kmalloc(sizeof *pd, GFP_KERNEL);
1151         if (!pd)
1152                 return ERR_PTR(-ENOMEM);
1153
1154         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1155         if (err) {
1156                 kfree(pd);
1157                 return ERR_PTR(err);
1158         }
1159
1160         if (context)
1161                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1162                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1163                         kfree(pd);
1164                         return ERR_PTR(-EFAULT);
1165                 }
1166
1167         return &pd->ibpd;
1168 }
1169
1170 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1171 {
1172         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1173         kfree(pd);
1174
1175         return 0;
1176 }
1177
1178 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1179                                           struct ib_ucontext *context,
1180                                           struct ib_udata *udata)
1181 {
1182         struct mlx4_ib_xrcd *xrcd;
1183         struct ib_cq_init_attr cq_attr = {};
1184         int err;
1185
1186         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1187                 return ERR_PTR(-ENOSYS);
1188
1189         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1190         if (!xrcd)
1191                 return ERR_PTR(-ENOMEM);
1192
1193         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1194         if (err)
1195                 goto err1;
1196
1197         xrcd->pd = ib_alloc_pd(ibdev);
1198         if (IS_ERR(xrcd->pd)) {
1199                 err = PTR_ERR(xrcd->pd);
1200                 goto err2;
1201         }
1202
1203         cq_attr.cqe = 1;
1204         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1205         if (IS_ERR(xrcd->cq)) {
1206                 err = PTR_ERR(xrcd->cq);
1207                 goto err3;
1208         }
1209
1210         return &xrcd->ibxrcd;
1211
1212 err3:
1213         ib_dealloc_pd(xrcd->pd);
1214 err2:
1215         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1216 err1:
1217         kfree(xrcd);
1218         return ERR_PTR(err);
1219 }
1220
1221 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1222 {
1223         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1224         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1225         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1226         kfree(xrcd);
1227
1228         return 0;
1229 }
1230
1231 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1232 {
1233         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1234         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1235         struct mlx4_ib_gid_entry *ge;
1236
1237         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1238         if (!ge)
1239                 return -ENOMEM;
1240
1241         ge->gid = *gid;
1242         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1243                 ge->port = mqp->port;
1244                 ge->added = 1;
1245         }
1246
1247         mutex_lock(&mqp->mutex);
1248         list_add_tail(&ge->list, &mqp->gid_list);
1249         mutex_unlock(&mqp->mutex);
1250
1251         return 0;
1252 }
1253
1254 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1255                                           struct mlx4_ib_counters *ctr_table)
1256 {
1257         struct counter_index *counter, *tmp_count;
1258
1259         mutex_lock(&ctr_table->mutex);
1260         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1261                                  list) {
1262                 if (counter->allocated)
1263                         mlx4_counter_free(ibdev->dev, counter->index);
1264                 list_del(&counter->list);
1265                 kfree(counter);
1266         }
1267         mutex_unlock(&ctr_table->mutex);
1268 }
1269
1270 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1271                    union ib_gid *gid)
1272 {
1273         struct net_device *ndev;
1274         int ret = 0;
1275
1276         if (!mqp->port)
1277                 return 0;
1278
1279         spin_lock_bh(&mdev->iboe.lock);
1280         ndev = mdev->iboe.netdevs[mqp->port - 1];
1281         if (ndev)
1282                 dev_hold(ndev);
1283         spin_unlock_bh(&mdev->iboe.lock);
1284
1285         if (ndev) {
1286                 ret = 1;
1287                 dev_put(ndev);
1288         }
1289
1290         return ret;
1291 }
1292
1293 struct mlx4_ib_steering {
1294         struct list_head list;
1295         struct mlx4_flow_reg_id reg_id;
1296         union ib_gid gid;
1297 };
1298
1299 static int parse_flow_attr(struct mlx4_dev *dev,
1300                            u32 qp_num,
1301                            union ib_flow_spec *ib_spec,
1302                            struct _rule_hw *mlx4_spec)
1303 {
1304         enum mlx4_net_trans_rule_id type;
1305
1306         switch (ib_spec->type) {
1307         case IB_FLOW_SPEC_ETH:
1308                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1309                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1310                        ETH_ALEN);
1311                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1312                        ETH_ALEN);
1313                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1314                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1315                 break;
1316         case IB_FLOW_SPEC_IB:
1317                 type = MLX4_NET_TRANS_RULE_ID_IB;
1318                 mlx4_spec->ib.l3_qpn =
1319                         cpu_to_be32(qp_num);
1320                 mlx4_spec->ib.qpn_mask =
1321                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1322                 break;
1323
1324
1325         case IB_FLOW_SPEC_IPV4:
1326                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1327                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1328                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1329                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1330                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1331                 break;
1332
1333         case IB_FLOW_SPEC_TCP:
1334         case IB_FLOW_SPEC_UDP:
1335                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1336                                         MLX4_NET_TRANS_RULE_ID_TCP :
1337                                         MLX4_NET_TRANS_RULE_ID_UDP;
1338                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1339                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1340                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1341                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1342                 break;
1343
1344         default:
1345                 return -EINVAL;
1346         }
1347         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1348             mlx4_hw_rule_sz(dev, type) < 0)
1349                 return -EINVAL;
1350         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1351         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1352         return mlx4_hw_rule_sz(dev, type);
1353 }
1354
1355 struct default_rules {
1356         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1357         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1358         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1359         __u8  link_layer;
1360 };
1361 static const struct default_rules default_table[] = {
1362         {
1363                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1364                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1365                 .rules_create_list = {IB_FLOW_SPEC_IB},
1366                 .link_layer = IB_LINK_LAYER_INFINIBAND
1367         }
1368 };
1369
1370 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1371                                          struct ib_flow_attr *flow_attr)
1372 {
1373         int i, j, k;
1374         void *ib_flow;
1375         const struct default_rules *pdefault_rules = default_table;
1376         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1377
1378         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1379                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1380                 memset(&field_types, 0, sizeof(field_types));
1381
1382                 if (link_layer != pdefault_rules->link_layer)
1383                         continue;
1384
1385                 ib_flow = flow_attr + 1;
1386                 /* we assume the specs are sorted */
1387                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1388                      j < flow_attr->num_of_specs; k++) {
1389                         union ib_flow_spec *current_flow =
1390                                 (union ib_flow_spec *)ib_flow;
1391
1392                         /* same layer but different type */
1393                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1394                              (pdefault_rules->mandatory_fields[k] &
1395                               IB_FLOW_SPEC_LAYER_MASK)) &&
1396                             (current_flow->type !=
1397                              pdefault_rules->mandatory_fields[k]))
1398                                 goto out;
1399
1400                         /* same layer, try match next one */
1401                         if (current_flow->type ==
1402                             pdefault_rules->mandatory_fields[k]) {
1403                                 j++;
1404                                 ib_flow +=
1405                                         ((union ib_flow_spec *)ib_flow)->size;
1406                         }
1407                 }
1408
1409                 ib_flow = flow_attr + 1;
1410                 for (j = 0; j < flow_attr->num_of_specs;
1411                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1412                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1413                                 /* same layer and same type */
1414                                 if (((union ib_flow_spec *)ib_flow)->type ==
1415                                     pdefault_rules->mandatory_not_fields[k])
1416                                         goto out;
1417
1418                 return i;
1419         }
1420 out:
1421         return -1;
1422 }
1423
1424 static int __mlx4_ib_create_default_rules(
1425                 struct mlx4_ib_dev *mdev,
1426                 struct ib_qp *qp,
1427                 const struct default_rules *pdefault_rules,
1428                 struct _rule_hw *mlx4_spec) {
1429         int size = 0;
1430         int i;
1431
1432         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1433                 int ret;
1434                 union ib_flow_spec ib_spec;
1435                 switch (pdefault_rules->rules_create_list[i]) {
1436                 case 0:
1437                         /* no rule */
1438                         continue;
1439                 case IB_FLOW_SPEC_IB:
1440                         ib_spec.type = IB_FLOW_SPEC_IB;
1441                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1442
1443                         break;
1444                 default:
1445                         /* invalid rule */
1446                         return -EINVAL;
1447                 }
1448                 /* We must put empty rule, qpn is being ignored */
1449                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1450                                       mlx4_spec);
1451                 if (ret < 0) {
1452                         pr_info("invalid parsing\n");
1453                         return -EINVAL;
1454                 }
1455
1456                 mlx4_spec = (void *)mlx4_spec + ret;
1457                 size += ret;
1458         }
1459         return size;
1460 }
1461
1462 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1463                           int domain,
1464                           enum mlx4_net_trans_promisc_mode flow_type,
1465                           u64 *reg_id)
1466 {
1467         int ret, i;
1468         int size = 0;
1469         void *ib_flow;
1470         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1471         struct mlx4_cmd_mailbox *mailbox;
1472         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1473         int default_flow;
1474
1475         static const u16 __mlx4_domain[] = {
1476                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1477                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1478                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1479                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1480         };
1481
1482         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1483                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1484                 return -EINVAL;
1485         }
1486
1487         if (domain >= IB_FLOW_DOMAIN_NUM) {
1488                 pr_err("Invalid domain value %d\n", domain);
1489                 return -EINVAL;
1490         }
1491
1492         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1493                 return -EINVAL;
1494
1495         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1496         if (IS_ERR(mailbox))
1497                 return PTR_ERR(mailbox);
1498         ctrl = mailbox->buf;
1499
1500         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1501                                  flow_attr->priority);
1502         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1503         ctrl->port = flow_attr->port;
1504         ctrl->qpn = cpu_to_be32(qp->qp_num);
1505
1506         ib_flow = flow_attr + 1;
1507         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1508         /* Add default flows */
1509         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1510         if (default_flow >= 0) {
1511                 ret = __mlx4_ib_create_default_rules(
1512                                 mdev, qp, default_table + default_flow,
1513                                 mailbox->buf + size);
1514                 if (ret < 0) {
1515                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1516                         return -EINVAL;
1517                 }
1518                 size += ret;
1519         }
1520         for (i = 0; i < flow_attr->num_of_specs; i++) {
1521                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1522                                       mailbox->buf + size);
1523                 if (ret < 0) {
1524                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1525                         return -EINVAL;
1526                 }
1527                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1528                 size += ret;
1529         }
1530
1531         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1532                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1533                            MLX4_CMD_WRAPPED);
1534         if (ret == -ENOMEM)
1535                 pr_err("mcg table is full. Fail to register network rule.\n");
1536         else if (ret == -ENXIO)
1537                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1538         else if (ret)
1539                 pr_err("Invalid argumant. Fail to register network rule.\n");
1540
1541         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1542         return ret;
1543 }
1544
1545 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1546 {
1547         int err;
1548         err = mlx4_cmd(dev, reg_id, 0, 0,
1549                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1550                        MLX4_CMD_WRAPPED);
1551         if (err)
1552                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1553                        reg_id);
1554         return err;
1555 }
1556
1557 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1558                                     u64 *reg_id)
1559 {
1560         void *ib_flow;
1561         union ib_flow_spec *ib_spec;
1562         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1563         int err = 0;
1564
1565         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1566             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1567                 return 0; /* do nothing */
1568
1569         ib_flow = flow_attr + 1;
1570         ib_spec = (union ib_flow_spec *)ib_flow;
1571
1572         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1573                 return 0; /* do nothing */
1574
1575         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1576                                     flow_attr->port, qp->qp_num,
1577                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1578                                     reg_id);
1579         return err;
1580 }
1581
1582 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1583                                     struct ib_flow_attr *flow_attr,
1584                                     int domain)
1585 {
1586         int err = 0, i = 0, j = 0;
1587         struct mlx4_ib_flow *mflow;
1588         enum mlx4_net_trans_promisc_mode type[2];
1589         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1590         int is_bonded = mlx4_is_bonded(dev);
1591
1592         memset(type, 0, sizeof(type));
1593
1594         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1595         if (!mflow) {
1596                 err = -ENOMEM;
1597                 goto err_free;
1598         }
1599
1600         switch (flow_attr->type) {
1601         case IB_FLOW_ATTR_NORMAL:
1602                 type[0] = MLX4_FS_REGULAR;
1603                 break;
1604
1605         case IB_FLOW_ATTR_ALL_DEFAULT:
1606                 type[0] = MLX4_FS_ALL_DEFAULT;
1607                 break;
1608
1609         case IB_FLOW_ATTR_MC_DEFAULT:
1610                 type[0] = MLX4_FS_MC_DEFAULT;
1611                 break;
1612
1613         case IB_FLOW_ATTR_SNIFFER:
1614                 type[0] = MLX4_FS_UC_SNIFFER;
1615                 type[1] = MLX4_FS_MC_SNIFFER;
1616                 break;
1617
1618         default:
1619                 err = -EINVAL;
1620                 goto err_free;
1621         }
1622
1623         while (i < ARRAY_SIZE(type) && type[i]) {
1624                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1625                                             &mflow->reg_id[i].id);
1626                 if (err)
1627                         goto err_create_flow;
1628                 if (is_bonded) {
1629                         /* Application always sees one port so the mirror rule
1630                          * must be on port #2
1631                          */
1632                         flow_attr->port = 2;
1633                         err = __mlx4_ib_create_flow(qp, flow_attr,
1634                                                     domain, type[j],
1635                                                     &mflow->reg_id[j].mirror);
1636                         flow_attr->port = 1;
1637                         if (err)
1638                                 goto err_create_flow;
1639                         j++;
1640                 }
1641
1642                 i++;
1643         }
1644
1645         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1646                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1647                                                &mflow->reg_id[i].id);
1648                 if (err)
1649                         goto err_create_flow;
1650
1651                 if (is_bonded) {
1652                         flow_attr->port = 2;
1653                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1654                                                        &mflow->reg_id[j].mirror);
1655                         flow_attr->port = 1;
1656                         if (err)
1657                                 goto err_create_flow;
1658                         j++;
1659                 }
1660                 /* function to create mirror rule */
1661                 i++;
1662         }
1663
1664         return &mflow->ibflow;
1665
1666 err_create_flow:
1667         while (i) {
1668                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1669                                              mflow->reg_id[i].id);
1670                 i--;
1671         }
1672
1673         while (j) {
1674                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1675                                              mflow->reg_id[j].mirror);
1676                 j--;
1677         }
1678 err_free:
1679         kfree(mflow);
1680         return ERR_PTR(err);
1681 }
1682
1683 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1684 {
1685         int err, ret = 0;
1686         int i = 0;
1687         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1688         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1689
1690         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1691                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1692                 if (err)
1693                         ret = err;
1694                 if (mflow->reg_id[i].mirror) {
1695                         err = __mlx4_ib_destroy_flow(mdev->dev,
1696                                                      mflow->reg_id[i].mirror);
1697                         if (err)
1698                                 ret = err;
1699                 }
1700                 i++;
1701         }
1702
1703         kfree(mflow);
1704         return ret;
1705 }
1706
1707 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1708 {
1709         int err;
1710         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1711         struct mlx4_dev *dev = mdev->dev;
1712         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1713         struct mlx4_ib_steering *ib_steering = NULL;
1714         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1715         struct mlx4_flow_reg_id reg_id;
1716
1717         if (mdev->dev->caps.steering_mode ==
1718             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1719                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1720                 if (!ib_steering)
1721                         return -ENOMEM;
1722         }
1723
1724         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1725                                     !!(mqp->flags &
1726                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1727                                     prot, &reg_id.id);
1728         if (err) {
1729                 pr_err("multicast attach op failed, err %d\n", err);
1730                 goto err_malloc;
1731         }
1732
1733         reg_id.mirror = 0;
1734         if (mlx4_is_bonded(dev)) {
1735                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1736                                             (mqp->port == 1) ? 2 : 1,
1737                                             !!(mqp->flags &
1738                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1739                                             prot, &reg_id.mirror);
1740                 if (err)
1741                         goto err_add;
1742         }
1743
1744         err = add_gid_entry(ibqp, gid);
1745         if (err)
1746                 goto err_add;
1747
1748         if (ib_steering) {
1749                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1750                 ib_steering->reg_id = reg_id;
1751                 mutex_lock(&mqp->mutex);
1752                 list_add(&ib_steering->list, &mqp->steering_rules);
1753                 mutex_unlock(&mqp->mutex);
1754         }
1755         return 0;
1756
1757 err_add:
1758         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1759                               prot, reg_id.id);
1760         if (reg_id.mirror)
1761                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1762                                       prot, reg_id.mirror);
1763 err_malloc:
1764         kfree(ib_steering);
1765
1766         return err;
1767 }
1768
1769 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1770 {
1771         struct mlx4_ib_gid_entry *ge;
1772         struct mlx4_ib_gid_entry *tmp;
1773         struct mlx4_ib_gid_entry *ret = NULL;
1774
1775         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1776                 if (!memcmp(raw, ge->gid.raw, 16)) {
1777                         ret = ge;
1778                         break;
1779                 }
1780         }
1781
1782         return ret;
1783 }
1784
1785 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1786 {
1787         int err;
1788         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1789         struct mlx4_dev *dev = mdev->dev;
1790         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1791         struct net_device *ndev;
1792         struct mlx4_ib_gid_entry *ge;
1793         struct mlx4_flow_reg_id reg_id = {0, 0};
1794         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1795
1796         if (mdev->dev->caps.steering_mode ==
1797             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1798                 struct mlx4_ib_steering *ib_steering;
1799
1800                 mutex_lock(&mqp->mutex);
1801                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1802                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1803                                 list_del(&ib_steering->list);
1804                                 break;
1805                         }
1806                 }
1807                 mutex_unlock(&mqp->mutex);
1808                 if (&ib_steering->list == &mqp->steering_rules) {
1809                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1810                         return -EINVAL;
1811                 }
1812                 reg_id = ib_steering->reg_id;
1813                 kfree(ib_steering);
1814         }
1815
1816         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1817                                     prot, reg_id.id);
1818         if (err)
1819                 return err;
1820
1821         if (mlx4_is_bonded(dev)) {
1822                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1823                                             prot, reg_id.mirror);
1824                 if (err)
1825                         return err;
1826         }
1827
1828         mutex_lock(&mqp->mutex);
1829         ge = find_gid_entry(mqp, gid->raw);
1830         if (ge) {
1831                 spin_lock_bh(&mdev->iboe.lock);
1832                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1833                 if (ndev)
1834                         dev_hold(ndev);
1835                 spin_unlock_bh(&mdev->iboe.lock);
1836                 if (ndev)
1837                         dev_put(ndev);
1838                 list_del(&ge->list);
1839                 kfree(ge);
1840         } else
1841                 pr_warn("could not find mgid entry\n");
1842
1843         mutex_unlock(&mqp->mutex);
1844
1845         return 0;
1846 }
1847
1848 static int init_node_data(struct mlx4_ib_dev *dev)
1849 {
1850         struct ib_smp *in_mad  = NULL;
1851         struct ib_smp *out_mad = NULL;
1852         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1853         int err = -ENOMEM;
1854
1855         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1856         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1857         if (!in_mad || !out_mad)
1858                 goto out;
1859
1860         init_query_mad(in_mad);
1861         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1862         if (mlx4_is_master(dev->dev))
1863                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1864
1865         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1866         if (err)
1867                 goto out;
1868
1869         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1870
1871         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1872
1873         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1874         if (err)
1875                 goto out;
1876
1877         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1878         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1879
1880 out:
1881         kfree(in_mad);
1882         kfree(out_mad);
1883         return err;
1884 }
1885
1886 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1887                         char *buf)
1888 {
1889         struct mlx4_ib_dev *dev =
1890                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1891         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1892 }
1893
1894 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1895                            char *buf)
1896 {
1897         struct mlx4_ib_dev *dev =
1898                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1899         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1900                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1901                        (int) dev->dev->caps.fw_ver & 0xffff);
1902 }
1903
1904 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1905                         char *buf)
1906 {
1907         struct mlx4_ib_dev *dev =
1908                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1909         return sprintf(buf, "%x\n", dev->dev->rev_id);
1910 }
1911
1912 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1913                           char *buf)
1914 {
1915         struct mlx4_ib_dev *dev =
1916                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1917         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1918                        dev->dev->board_id);
1919 }
1920
1921 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1922 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1923 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1924 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1925
1926 static struct device_attribute *mlx4_class_attributes[] = {
1927         &dev_attr_hw_rev,
1928         &dev_attr_fw_ver,
1929         &dev_attr_hca_type,
1930         &dev_attr_board_id
1931 };
1932
1933 #define MLX4_IB_INVALID_MAC     ((u64)-1)
1934 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1935                                struct net_device *dev,
1936                                int port)
1937 {
1938         u64 new_smac = 0;
1939         u64 release_mac = MLX4_IB_INVALID_MAC;
1940         struct mlx4_ib_qp *qp;
1941
1942         read_lock(&dev_base_lock);
1943         new_smac = mlx4_mac_to_u64(dev->dev_addr);
1944         read_unlock(&dev_base_lock);
1945
1946         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1947
1948         /* no need for update QP1 and mac registration in non-SRIOV */
1949         if (!mlx4_is_mfunc(ibdev->dev))
1950                 return;
1951
1952         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1953         qp = ibdev->qp1_proxy[port - 1];
1954         if (qp) {
1955                 int new_smac_index;
1956                 u64 old_smac;
1957                 struct mlx4_update_qp_params update_params;
1958
1959                 mutex_lock(&qp->mutex);
1960                 old_smac = qp->pri.smac;
1961                 if (new_smac == old_smac)
1962                         goto unlock;
1963
1964                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1965
1966                 if (new_smac_index < 0)
1967                         goto unlock;
1968
1969                 update_params.smac_index = new_smac_index;
1970                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1971                                    &update_params)) {
1972                         release_mac = new_smac;
1973                         goto unlock;
1974                 }
1975                 /* if old port was zero, no mac was yet registered for this QP */
1976                 if (qp->pri.smac_port)
1977                         release_mac = old_smac;
1978                 qp->pri.smac = new_smac;
1979                 qp->pri.smac_port = port;
1980                 qp->pri.smac_index = new_smac_index;
1981         }
1982
1983 unlock:
1984         if (release_mac != MLX4_IB_INVALID_MAC)
1985                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1986         if (qp)
1987                 mutex_unlock(&qp->mutex);
1988         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1989 }
1990
1991 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1992                                  struct net_device *dev,
1993                                  unsigned long event)
1994
1995 {
1996         struct mlx4_ib_iboe *iboe;
1997         int update_qps_port = -1;
1998         int port;
1999
2000         ASSERT_RTNL();
2001
2002         iboe = &ibdev->iboe;
2003
2004         spin_lock_bh(&iboe->lock);
2005         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2006
2007                 iboe->netdevs[port - 1] =
2008                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2009
2010                 if (dev == iboe->netdevs[port - 1] &&
2011                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2012                      event == NETDEV_UP || event == NETDEV_CHANGE))
2013                         update_qps_port = port;
2014
2015         }
2016         spin_unlock_bh(&iboe->lock);
2017
2018         if (update_qps_port > 0)
2019                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2020 }
2021
2022 static int mlx4_ib_netdev_event(struct notifier_block *this,
2023                                 unsigned long event, void *ptr)
2024 {
2025         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2026         struct mlx4_ib_dev *ibdev;
2027
2028         if (!net_eq(dev_net(dev), &init_net))
2029                 return NOTIFY_DONE;
2030
2031         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2032         mlx4_ib_scan_netdevs(ibdev, dev, event);
2033
2034         return NOTIFY_DONE;
2035 }
2036
2037 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2038 {
2039         int port;
2040         int slave;
2041         int i;
2042
2043         if (mlx4_is_master(ibdev->dev)) {
2044                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2045                      ++slave) {
2046                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2047                                 for (i = 0;
2048                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2049                                      ++i) {
2050                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2051                                         /* master has the identity virt2phys pkey mapping */
2052                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2053                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2054                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2055                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2056                                 }
2057                         }
2058                 }
2059                 /* initialize pkey cache */
2060                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2061                         for (i = 0;
2062                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2063                              ++i)
2064                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2065                                         (i) ? 0 : 0xFFFF;
2066                 }
2067         }
2068 }
2069
2070 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2071 {
2072         int i, j, eq = 0, total_eqs = 0;
2073
2074         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2075                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2076         if (!ibdev->eq_table)
2077                 return;
2078
2079         for (i = 1; i <= dev->caps.num_ports; i++) {
2080                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2081                      j++, total_eqs++) {
2082                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2083                                 continue;
2084                         ibdev->eq_table[eq] = total_eqs;
2085                         if (!mlx4_assign_eq(dev, i,
2086                                             &ibdev->eq_table[eq]))
2087                                 eq++;
2088                         else
2089                                 ibdev->eq_table[eq] = -1;
2090                 }
2091         }
2092
2093         for (i = eq; i < dev->caps.num_comp_vectors;
2094              ibdev->eq_table[i++] = -1)
2095                 ;
2096
2097         /* Advertise the new number of EQs to clients */
2098         ibdev->ib_dev.num_comp_vectors = eq;
2099 }
2100
2101 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2102 {
2103         int i;
2104         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2105
2106         /* no eqs were allocated */
2107         if (!ibdev->eq_table)
2108                 return;
2109
2110         /* Reset the advertised EQ number */
2111         ibdev->ib_dev.num_comp_vectors = 0;
2112
2113         for (i = 0; i < total_eqs; i++)
2114                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2115
2116         kfree(ibdev->eq_table);
2117         ibdev->eq_table = NULL;
2118 }
2119
2120 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2121                                struct ib_port_immutable *immutable)
2122 {
2123         struct ib_port_attr attr;
2124         int err;
2125
2126         err = mlx4_ib_query_port(ibdev, port_num, &attr);
2127         if (err)
2128                 return err;
2129
2130         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2131         immutable->gid_tbl_len = attr.gid_tbl_len;
2132
2133         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2134                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2135         else
2136                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2137
2138         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2139
2140         return 0;
2141 }
2142
2143 static void *mlx4_ib_add(struct mlx4_dev *dev)
2144 {
2145         struct mlx4_ib_dev *ibdev;
2146         int num_ports = 0;
2147         int i, j;
2148         int err;
2149         struct mlx4_ib_iboe *iboe;
2150         int ib_num_ports = 0;
2151         int num_req_counters;
2152         int allocated;
2153         u32 counter_index;
2154         struct counter_index *new_counter_index = NULL;
2155
2156         pr_info_once("%s", mlx4_ib_version);
2157
2158         num_ports = 0;
2159         mlx4_foreach_ib_transport_port(i, dev)
2160                 num_ports++;
2161
2162         /* No point in registering a device with no ports... */
2163         if (num_ports == 0)
2164                 return NULL;
2165
2166         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2167         if (!ibdev) {
2168                 dev_err(&dev->persist->pdev->dev,
2169                         "Device struct alloc failed\n");
2170                 return NULL;
2171         }
2172
2173         iboe = &ibdev->iboe;
2174
2175         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2176                 goto err_dealloc;
2177
2178         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2179                 goto err_pd;
2180
2181         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2182                                  PAGE_SIZE);
2183         if (!ibdev->uar_map)
2184                 goto err_uar;
2185         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2186
2187         ibdev->dev = dev;
2188         ibdev->bond_next_port   = 0;
2189
2190         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2191         ibdev->ib_dev.owner             = THIS_MODULE;
2192         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2193         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2194         ibdev->num_ports                = num_ports;
2195         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2196                                                 1 : ibdev->num_ports;
2197         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2198         ibdev->ib_dev.dma_device        = &dev->persist->pdev->dev;
2199         ibdev->ib_dev.get_netdev        = mlx4_ib_get_netdev;
2200         ibdev->ib_dev.add_gid           = mlx4_ib_add_gid;
2201         ibdev->ib_dev.del_gid           = mlx4_ib_del_gid;
2202
2203         if (dev->caps.userspace_caps)
2204                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2205         else
2206                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2207
2208         ibdev->ib_dev.uverbs_cmd_mask   =
2209                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2210                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2211                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2212                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2213                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2214                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2215                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2216                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2217                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2218                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2219                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2220                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2221                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2222                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2223                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2224                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2225                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2226                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2227                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2228                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2229                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2230                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2231                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2232                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2233
2234         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2235         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2236         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2237         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2238         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2239         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2240         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2241         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2242         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2243         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2244         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2245         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2246         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2247         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2248         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2249         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2250         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2251         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2252         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2253         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2254         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2255         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2256         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2257         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2258         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2259         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2260         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2261         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2262         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2263         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2264         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2265         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2266         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2267         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2268         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2269         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2270         ibdev->ib_dev.alloc_mr          = mlx4_ib_alloc_mr;
2271         ibdev->ib_dev.map_mr_sg         = mlx4_ib_map_mr_sg;
2272         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2273         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2274         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2275         ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2276         ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2277
2278         if (!mlx4_is_slave(ibdev->dev)) {
2279                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2280                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2281                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2282                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2283         }
2284
2285         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2286             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2287                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2288                 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2289                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2290
2291                 ibdev->ib_dev.uverbs_cmd_mask |=
2292                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2293                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2294         }
2295
2296         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2297                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2298                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2299                 ibdev->ib_dev.uverbs_cmd_mask |=
2300                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2301                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2302         }
2303
2304         if (check_flow_steering_support(dev)) {
2305                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2306                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2307                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2308
2309                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2310                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2311                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2312         }
2313
2314         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2315                 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2316                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2317                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2318
2319         mlx4_ib_alloc_eqs(dev, ibdev);
2320
2321         spin_lock_init(&iboe->lock);
2322
2323         if (init_node_data(ibdev))
2324                 goto err_map;
2325
2326         for (i = 0; i < ibdev->num_ports; ++i) {
2327                 mutex_init(&ibdev->counters_table[i].mutex);
2328                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2329         }
2330
2331         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2332         for (i = 0; i < num_req_counters; ++i) {
2333                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2334                 allocated = 0;
2335                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2336                                                 IB_LINK_LAYER_ETHERNET) {
2337                         err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2338                         /* if failed to allocate a new counter, use default */
2339                         if (err)
2340                                 counter_index =
2341                                         mlx4_get_default_counter_index(dev,
2342                                                                        i + 1);
2343                         else
2344                                 allocated = 1;
2345                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2346                         counter_index = mlx4_get_default_counter_index(dev,
2347                                                                        i + 1);
2348                 }
2349                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2350                                             GFP_KERNEL);
2351                 if (!new_counter_index) {
2352                         if (allocated)
2353                                 mlx4_counter_free(ibdev->dev, counter_index);
2354                         goto err_counter;
2355                 }
2356                 new_counter_index->index = counter_index;
2357                 new_counter_index->allocated = allocated;
2358                 list_add_tail(&new_counter_index->list,
2359                               &ibdev->counters_table[i].counters_list);
2360                 ibdev->counters_table[i].default_counter = counter_index;
2361                 pr_info("counter index %d for port %d allocated %d\n",
2362                         counter_index, i + 1, allocated);
2363         }
2364         if (mlx4_is_bonded(dev))
2365                 for (i = 1; i < ibdev->num_ports ; ++i) {
2366                         new_counter_index =
2367                                         kmalloc(sizeof(struct counter_index),
2368                                                 GFP_KERNEL);
2369                         if (!new_counter_index)
2370                                 goto err_counter;
2371                         new_counter_index->index = counter_index;
2372                         new_counter_index->allocated = 0;
2373                         list_add_tail(&new_counter_index->list,
2374                                       &ibdev->counters_table[i].counters_list);
2375                         ibdev->counters_table[i].default_counter =
2376                                                                 counter_index;
2377                 }
2378
2379         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2380                 ib_num_ports++;
2381
2382         spin_lock_init(&ibdev->sm_lock);
2383         mutex_init(&ibdev->cap_mask_mutex);
2384         INIT_LIST_HEAD(&ibdev->qp_list);
2385         spin_lock_init(&ibdev->reset_flow_resource_lock);
2386
2387         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2388             ib_num_ports) {
2389                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2390                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2391                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2392                                             &ibdev->steer_qpn_base, 0);
2393                 if (err)
2394                         goto err_counter;
2395
2396                 ibdev->ib_uc_qpns_bitmap =
2397                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2398                                 sizeof(long),
2399                                 GFP_KERNEL);
2400                 if (!ibdev->ib_uc_qpns_bitmap) {
2401                         dev_err(&dev->persist->pdev->dev,
2402                                 "bit map alloc failed\n");
2403                         goto err_steer_qp_release;
2404                 }
2405
2406                 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2407
2408                 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2409                                 dev, ibdev->steer_qpn_base,
2410                                 ibdev->steer_qpn_base +
2411                                 ibdev->steer_qpn_count - 1);
2412                 if (err)
2413                         goto err_steer_free_bitmap;
2414         }
2415
2416         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2417                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2418
2419         if (ib_register_device(&ibdev->ib_dev, NULL))
2420                 goto err_steer_free_bitmap;
2421
2422         if (mlx4_ib_mad_init(ibdev))
2423                 goto err_reg;
2424
2425         if (mlx4_ib_init_sriov(ibdev))
2426                 goto err_mad;
2427
2428         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2429                 if (!iboe->nb.notifier_call) {
2430                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2431                         err = register_netdevice_notifier(&iboe->nb);
2432                         if (err) {
2433                                 iboe->nb.notifier_call = NULL;
2434                                 goto err_notif;
2435                         }
2436                 }
2437         }
2438
2439         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2440                 if (device_create_file(&ibdev->ib_dev.dev,
2441                                        mlx4_class_attributes[j]))
2442                         goto err_notif;
2443         }
2444
2445         ibdev->ib_active = true;
2446
2447         if (mlx4_is_mfunc(ibdev->dev))
2448                 init_pkeys(ibdev);
2449
2450         /* create paravirt contexts for any VFs which are active */
2451         if (mlx4_is_master(ibdev->dev)) {
2452                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2453                         if (j == mlx4_master_func_num(ibdev->dev))
2454                                 continue;
2455                         if (mlx4_is_slave_active(ibdev->dev, j))
2456                                 do_slave_init(ibdev, j, 1);
2457                 }
2458         }
2459         return ibdev;
2460
2461 err_notif:
2462         if (ibdev->iboe.nb.notifier_call) {
2463                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2464                         pr_warn("failure unregistering notifier\n");
2465                 ibdev->iboe.nb.notifier_call = NULL;
2466         }
2467         flush_workqueue(wq);
2468
2469         mlx4_ib_close_sriov(ibdev);
2470
2471 err_mad:
2472         mlx4_ib_mad_cleanup(ibdev);
2473
2474 err_reg:
2475         ib_unregister_device(&ibdev->ib_dev);
2476
2477 err_steer_free_bitmap:
2478         kfree(ibdev->ib_uc_qpns_bitmap);
2479
2480 err_steer_qp_release:
2481         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2482                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2483                                       ibdev->steer_qpn_count);
2484 err_counter:
2485         for (i = 0; i < ibdev->num_ports; ++i)
2486                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2487
2488 err_map:
2489         iounmap(ibdev->uar_map);
2490
2491 err_uar:
2492         mlx4_uar_free(dev, &ibdev->priv_uar);
2493
2494 err_pd:
2495         mlx4_pd_free(dev, ibdev->priv_pdn);
2496
2497 err_dealloc:
2498         ib_dealloc_device(&ibdev->ib_dev);
2499
2500         return NULL;
2501 }
2502
2503 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2504 {
2505         int offset;
2506
2507         WARN_ON(!dev->ib_uc_qpns_bitmap);
2508
2509         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2510                                          dev->steer_qpn_count,
2511                                          get_count_order(count));
2512         if (offset < 0)
2513                 return offset;
2514
2515         *qpn = dev->steer_qpn_base + offset;
2516         return 0;
2517 }
2518
2519 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2520 {
2521         if (!qpn ||
2522             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2523                 return;
2524
2525         BUG_ON(qpn < dev->steer_qpn_base);
2526
2527         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2528                               qpn - dev->steer_qpn_base,
2529                               get_count_order(count));
2530 }
2531
2532 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2533                          int is_attach)
2534 {
2535         int err;
2536         size_t flow_size;
2537         struct ib_flow_attr *flow = NULL;
2538         struct ib_flow_spec_ib *ib_spec;
2539
2540         if (is_attach) {
2541                 flow_size = sizeof(struct ib_flow_attr) +
2542                             sizeof(struct ib_flow_spec_ib);
2543                 flow = kzalloc(flow_size, GFP_KERNEL);
2544                 if (!flow)
2545                         return -ENOMEM;
2546                 flow->port = mqp->port;
2547                 flow->num_of_specs = 1;
2548                 flow->size = flow_size;
2549                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2550                 ib_spec->type = IB_FLOW_SPEC_IB;
2551                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2552                 /* Add an empty rule for IB L2 */
2553                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2554
2555                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2556                                             IB_FLOW_DOMAIN_NIC,
2557                                             MLX4_FS_REGULAR,
2558                                             &mqp->reg_id);
2559         } else {
2560                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2561         }
2562         kfree(flow);
2563         return err;
2564 }
2565
2566 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2567 {
2568         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2569         int p;
2570
2571         ibdev->ib_active = false;
2572         flush_workqueue(wq);
2573
2574         mlx4_ib_close_sriov(ibdev);
2575         mlx4_ib_mad_cleanup(ibdev);
2576         ib_unregister_device(&ibdev->ib_dev);
2577         if (ibdev->iboe.nb.notifier_call) {
2578                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2579                         pr_warn("failure unregistering notifier\n");
2580                 ibdev->iboe.nb.notifier_call = NULL;
2581         }
2582
2583         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2584                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2585                                       ibdev->steer_qpn_count);
2586                 kfree(ibdev->ib_uc_qpns_bitmap);
2587         }
2588
2589         iounmap(ibdev->uar_map);
2590         for (p = 0; p < ibdev->num_ports; ++p)
2591                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2592
2593         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2594                 mlx4_CLOSE_PORT(dev, p);
2595
2596         mlx4_ib_free_eqs(dev, ibdev);
2597
2598         mlx4_uar_free(dev, &ibdev->priv_uar);
2599         mlx4_pd_free(dev, ibdev->priv_pdn);
2600         ib_dealloc_device(&ibdev->ib_dev);
2601 }
2602
2603 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2604 {
2605         struct mlx4_ib_demux_work **dm = NULL;
2606         struct mlx4_dev *dev = ibdev->dev;
2607         int i;
2608         unsigned long flags;
2609         struct mlx4_active_ports actv_ports;
2610         unsigned int ports;
2611         unsigned int first_port;
2612
2613         if (!mlx4_is_master(dev))
2614                 return;
2615
2616         actv_ports = mlx4_get_active_ports(dev, slave);
2617         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2618         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2619
2620         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2621         if (!dm) {
2622                 pr_err("failed to allocate memory for tunneling qp update\n");
2623                 return;
2624         }
2625
2626         for (i = 0; i < ports; i++) {
2627                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2628                 if (!dm[i]) {
2629                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2630                         while (--i >= 0)
2631                                 kfree(dm[i]);
2632                         goto out;
2633                 }
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;
2638                 dm[i]->dev = ibdev;
2639         }
2640         /* initialize or tear down tunnel QPs for the slave */
2641         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2642         if (!ibdev->sriov.is_going_down) {
2643                 for (i = 0; i < ports; i++)
2644                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2645                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2646         } else {
2647                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2648                 for (i = 0; i < ports; i++)
2649                         kfree(dm[i]);
2650         }
2651 out:
2652         kfree(dm);
2653         return;
2654 }
2655
2656 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2657 {
2658         struct mlx4_ib_qp *mqp;
2659         unsigned long flags_qp;
2660         unsigned long flags_cq;
2661         struct mlx4_ib_cq *send_mcq, *recv_mcq;
2662         struct list_head    cq_notify_list;
2663         struct mlx4_cq *mcq;
2664         unsigned long flags;
2665
2666         pr_warn("mlx4_ib_handle_catas_error was started\n");
2667         INIT_LIST_HEAD(&cq_notify_list);
2668
2669         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2670         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2671
2672         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2673                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2674                 if (mqp->sq.tail != mqp->sq.head) {
2675                         send_mcq = to_mcq(mqp->ibqp.send_cq);
2676                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
2677                         if (send_mcq->mcq.comp &&
2678                             mqp->ibqp.send_cq->comp_handler) {
2679                                 if (!send_mcq->mcq.reset_notify_added) {
2680                                         send_mcq->mcq.reset_notify_added = 1;
2681                                         list_add_tail(&send_mcq->mcq.reset_notify,
2682                                                       &cq_notify_list);
2683                                 }
2684                         }
2685                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2686                 }
2687                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2688                 /* Now, handle the QP's receive queue */
2689                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2690                 /* no handling is needed for SRQ */
2691                 if (!mqp->ibqp.srq) {
2692                         if (mqp->rq.tail != mqp->rq.head) {
2693                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2694                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2695                                 if (recv_mcq->mcq.comp &&
2696                                     mqp->ibqp.recv_cq->comp_handler) {
2697                                         if (!recv_mcq->mcq.reset_notify_added) {
2698                                                 recv_mcq->mcq.reset_notify_added = 1;
2699                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
2700                                                               &cq_notify_list);
2701                                         }
2702                                 }
2703                                 spin_unlock_irqrestore(&recv_mcq->lock,
2704                                                        flags_cq);
2705                         }
2706                 }
2707                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2708         }
2709
2710         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2711                 mcq->comp(mcq);
2712         }
2713         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2714         pr_warn("mlx4_ib_handle_catas_error ended\n");
2715 }
2716
2717 static void handle_bonded_port_state_event(struct work_struct *work)
2718 {
2719         struct ib_event_work *ew =
2720                 container_of(work, struct ib_event_work, work);
2721         struct mlx4_ib_dev *ibdev = ew->ib_dev;
2722         enum ib_port_state bonded_port_state = IB_PORT_NOP;
2723         int i;
2724         struct ib_event ibev;
2725
2726         kfree(ew);
2727         spin_lock_bh(&ibdev->iboe.lock);
2728         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2729                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2730                 enum ib_port_state curr_port_state;
2731
2732                 if (!curr_netdev)
2733                         continue;
2734
2735                 curr_port_state =
2736                         (netif_running(curr_netdev) &&
2737                          netif_carrier_ok(curr_netdev)) ?
2738                         IB_PORT_ACTIVE : IB_PORT_DOWN;
2739
2740                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2741                         curr_port_state : IB_PORT_ACTIVE;
2742         }
2743         spin_unlock_bh(&ibdev->iboe.lock);
2744
2745         ibev.device = &ibdev->ib_dev;
2746         ibev.element.port_num = 1;
2747         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2748                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2749
2750         ib_dispatch_event(&ibev);
2751 }
2752
2753 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2754                           enum mlx4_dev_event event, unsigned long param)
2755 {
2756         struct ib_event ibev;
2757         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2758         struct mlx4_eqe *eqe = NULL;
2759         struct ib_event_work *ew;
2760         int p = 0;
2761
2762         if (mlx4_is_bonded(dev) &&
2763             ((event == MLX4_DEV_EVENT_PORT_UP) ||
2764             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2765                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2766                 if (!ew)
2767                         return;
2768                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
2769                 ew->ib_dev = ibdev;
2770                 queue_work(wq, &ew->work);
2771                 return;
2772         }
2773
2774         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2775                 eqe = (struct mlx4_eqe *)param;
2776         else
2777                 p = (int) param;
2778
2779         switch (event) {
2780         case MLX4_DEV_EVENT_PORT_UP:
2781                 if (p > ibdev->num_ports)
2782                         return;
2783                 if (mlx4_is_master(dev) &&
2784                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2785                         IB_LINK_LAYER_INFINIBAND) {
2786                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
2787                 }
2788                 ibev.event = IB_EVENT_PORT_ACTIVE;
2789                 break;
2790
2791         case MLX4_DEV_EVENT_PORT_DOWN:
2792                 if (p > ibdev->num_ports)
2793                         return;
2794                 ibev.event = IB_EVENT_PORT_ERR;
2795                 break;
2796
2797         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2798                 ibdev->ib_active = false;
2799                 ibev.event = IB_EVENT_DEVICE_FATAL;
2800                 mlx4_ib_handle_catas_error(ibdev);
2801                 break;
2802
2803         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2804                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2805                 if (!ew) {
2806                         pr_err("failed to allocate memory for events work\n");
2807                         break;
2808                 }
2809
2810                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2811                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2812                 ew->ib_dev = ibdev;
2813                 /* need to queue only for port owner, which uses GEN_EQE */
2814                 if (mlx4_is_master(dev))
2815                         queue_work(wq, &ew->work);
2816                 else
2817                         handle_port_mgmt_change_event(&ew->work);
2818                 return;
2819
2820         case MLX4_DEV_EVENT_SLAVE_INIT:
2821                 /* here, p is the slave id */
2822                 do_slave_init(ibdev, p, 1);
2823                 if (mlx4_is_master(dev)) {
2824                         int i;
2825
2826                         for (i = 1; i <= ibdev->num_ports; i++) {
2827                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2828                                         == IB_LINK_LAYER_INFINIBAND)
2829                                         mlx4_ib_slave_alias_guid_event(ibdev,
2830                                                                        p, i,
2831                                                                        1);
2832                         }
2833                 }
2834                 return;
2835
2836         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2837                 if (mlx4_is_master(dev)) {
2838                         int i;
2839
2840                         for (i = 1; i <= ibdev->num_ports; i++) {
2841                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2842                                         == IB_LINK_LAYER_INFINIBAND)
2843                                         mlx4_ib_slave_alias_guid_event(ibdev,
2844                                                                        p, i,
2845                                                                        0);
2846                         }
2847                 }
2848                 /* here, p is the slave id */
2849                 do_slave_init(ibdev, p, 0);
2850                 return;
2851
2852         default:
2853                 return;
2854         }
2855
2856         ibev.device           = ibdev_ptr;
2857         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2858
2859         ib_dispatch_event(&ibev);
2860 }
2861
2862 static struct mlx4_interface mlx4_ib_interface = {
2863         .add            = mlx4_ib_add,
2864         .remove         = mlx4_ib_remove,
2865         .event          = mlx4_ib_event,
2866         .protocol       = MLX4_PROT_IB_IPV6,
2867         .flags          = MLX4_INTFF_BONDING
2868 };
2869
2870 static int __init mlx4_ib_init(void)
2871 {
2872         int err;
2873
2874         wq = create_singlethread_workqueue("mlx4_ib");
2875         if (!wq)
2876                 return -ENOMEM;
2877
2878         err = mlx4_ib_mcg_init();
2879         if (err)
2880                 goto clean_wq;
2881
2882         err = mlx4_register_interface(&mlx4_ib_interface);
2883         if (err)
2884                 goto clean_mcg;
2885
2886         return 0;
2887
2888 clean_mcg:
2889         mlx4_ib_mcg_destroy();
2890
2891 clean_wq:
2892         destroy_workqueue(wq);
2893         return err;
2894 }
2895
2896 static void __exit mlx4_ib_cleanup(void)
2897 {
2898         mlx4_unregister_interface(&mlx4_ib_interface);
2899         mlx4_ib_mcg_destroy();
2900         destroy_workqueue(wq);
2901 }
2902
2903 module_init(mlx4_ib_init);
2904 module_exit(mlx4_ib_cleanup);