GNU Linux-libre 4.9.287-gnu1
[releases.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 #include <net/devlink.h>
45
46 #include <rdma/ib_smi.h>
47 #include <rdma/ib_user_verbs.h>
48 #include <rdma/ib_addr.h>
49 #include <rdma/ib_cache.h>
50
51 #include <net/bonding.h>
52
53 #include <linux/mlx4/driver.h>
54 #include <linux/mlx4/cmd.h>
55 #include <linux/mlx4/qp.h>
56
57 #include "mlx4_ib.h"
58 #include <rdma/mlx4-abi.h>
59
60 #define DRV_NAME        MLX4_IB_DRV_NAME
61 #define DRV_VERSION     "2.2-1"
62 #define DRV_RELDATE     "Feb 2014"
63
64 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
65 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
66 #define MLX4_IB_CARD_REV_A0   0xA0
67
68 MODULE_AUTHOR("Roland Dreier");
69 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
70 MODULE_LICENSE("Dual BSD/GPL");
71 MODULE_VERSION(DRV_VERSION);
72
73 int mlx4_ib_sm_guid_assign = 0;
74 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
75 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
76
77 static const char mlx4_ib_version[] =
78         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
79         DRV_VERSION " (" DRV_RELDATE ")\n";
80
81 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
82
83 static struct workqueue_struct *wq;
84
85 static void init_query_mad(struct ib_smp *mad)
86 {
87         mad->base_version  = 1;
88         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
89         mad->class_version = 1;
90         mad->method        = IB_MGMT_METHOD_GET;
91 }
92
93 static int check_flow_steering_support(struct mlx4_dev *dev)
94 {
95         int eth_num_ports = 0;
96         int ib_num_ports = 0;
97
98         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
99
100         if (dmfs) {
101                 int i;
102                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
103                         eth_num_ports++;
104                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
105                         ib_num_ports++;
106                 dmfs &= (!ib_num_ports ||
107                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
108                         (!eth_num_ports ||
109                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
110                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
111                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
112                         dmfs = 0;
113                 }
114         }
115         return dmfs;
116 }
117
118 static int num_ib_ports(struct mlx4_dev *dev)
119 {
120         int ib_ports = 0;
121         int i;
122
123         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
124                 ib_ports++;
125
126         return ib_ports;
127 }
128
129 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
130 {
131         struct mlx4_ib_dev *ibdev = to_mdev(device);
132         struct net_device *dev;
133
134         rcu_read_lock();
135         dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
136
137         if (dev) {
138                 if (mlx4_is_bonded(ibdev->dev)) {
139                         struct net_device *upper = NULL;
140
141                         upper = netdev_master_upper_dev_get_rcu(dev);
142                         if (upper) {
143                                 struct net_device *active;
144
145                                 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
146                                 if (active)
147                                         dev = active;
148                         }
149                 }
150         }
151         if (dev)
152                 dev_hold(dev);
153
154         rcu_read_unlock();
155         return dev;
156 }
157
158 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
159                                   struct mlx4_ib_dev *ibdev,
160                                   u8 port_num)
161 {
162         struct mlx4_cmd_mailbox *mailbox;
163         int err;
164         struct mlx4_dev *dev = ibdev->dev;
165         int i;
166         union ib_gid *gid_tbl;
167
168         mailbox = mlx4_alloc_cmd_mailbox(dev);
169         if (IS_ERR(mailbox))
170                 return -ENOMEM;
171
172         gid_tbl = mailbox->buf;
173
174         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
175                 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
176
177         err = mlx4_cmd(dev, mailbox->dma,
178                        MLX4_SET_PORT_GID_TABLE << 8 | port_num,
179                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
180                        MLX4_CMD_WRAPPED);
181         if (mlx4_is_bonded(dev))
182                 err += mlx4_cmd(dev, mailbox->dma,
183                                 MLX4_SET_PORT_GID_TABLE << 8 | 2,
184                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
185                                 MLX4_CMD_WRAPPED);
186
187         mlx4_free_cmd_mailbox(dev, mailbox);
188         return err;
189 }
190
191 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
192                                      struct mlx4_ib_dev *ibdev,
193                                      u8 port_num)
194 {
195         struct mlx4_cmd_mailbox *mailbox;
196         int err;
197         struct mlx4_dev *dev = ibdev->dev;
198         int i;
199         struct {
200                 union ib_gid    gid;
201                 __be32          rsrvd1[2];
202                 __be16          rsrvd2;
203                 u8              type;
204                 u8              version;
205                 __be32          rsrvd3;
206         } *gid_tbl;
207
208         mailbox = mlx4_alloc_cmd_mailbox(dev);
209         if (IS_ERR(mailbox))
210                 return -ENOMEM;
211
212         gid_tbl = mailbox->buf;
213         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
214                 memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
215                 if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
216                         gid_tbl[i].version = 2;
217                         if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
218                                 gid_tbl[i].type = 1;
219                 }
220         }
221
222         err = mlx4_cmd(dev, mailbox->dma,
223                        MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
224                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
225                        MLX4_CMD_WRAPPED);
226         if (mlx4_is_bonded(dev))
227                 err += mlx4_cmd(dev, mailbox->dma,
228                                 MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
229                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
230                                 MLX4_CMD_WRAPPED);
231
232         mlx4_free_cmd_mailbox(dev, mailbox);
233         return err;
234 }
235
236 static int mlx4_ib_update_gids(struct gid_entry *gids,
237                                struct mlx4_ib_dev *ibdev,
238                                u8 port_num)
239 {
240         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
241                 return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
242
243         return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
244 }
245
246 static int mlx4_ib_add_gid(struct ib_device *device,
247                            u8 port_num,
248                            unsigned int index,
249                            const union ib_gid *gid,
250                            const struct ib_gid_attr *attr,
251                            void **context)
252 {
253         struct mlx4_ib_dev *ibdev = to_mdev(device);
254         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
255         struct mlx4_port_gid_table   *port_gid_table;
256         int free = -1, found = -1;
257         int ret = 0;
258         int hw_update = 0;
259         int i;
260         struct gid_entry *gids = NULL;
261
262         if (!rdma_cap_roce_gid_table(device, port_num))
263                 return -EINVAL;
264
265         if (port_num > MLX4_MAX_PORTS)
266                 return -EINVAL;
267
268         if (!context)
269                 return -EINVAL;
270
271         port_gid_table = &iboe->gids[port_num - 1];
272         spin_lock_bh(&iboe->lock);
273         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
274                 if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
275                     (port_gid_table->gids[i].gid_type == attr->gid_type))  {
276                         found = i;
277                         break;
278                 }
279                 if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
280                         free = i; /* HW has space */
281         }
282
283         if (found < 0) {
284                 if (free < 0) {
285                         ret = -ENOSPC;
286                 } else {
287                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
288                         if (!port_gid_table->gids[free].ctx) {
289                                 ret = -ENOMEM;
290                         } else {
291                                 *context = port_gid_table->gids[free].ctx;
292                                 memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
293                                 port_gid_table->gids[free].gid_type = attr->gid_type;
294                                 port_gid_table->gids[free].ctx->real_index = free;
295                                 port_gid_table->gids[free].ctx->refcount = 1;
296                                 hw_update = 1;
297                         }
298                 }
299         } else {
300                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
301                 *context = ctx;
302                 ctx->refcount++;
303         }
304         if (!ret && hw_update) {
305                 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
306                 if (!gids) {
307                         ret = -ENOMEM;
308                 } else {
309                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
310                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
311                                 gids[i].gid_type = port_gid_table->gids[i].gid_type;
312                         }
313                 }
314         }
315         spin_unlock_bh(&iboe->lock);
316
317         if (!ret && hw_update) {
318                 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
319                 kfree(gids);
320         }
321
322         return ret;
323 }
324
325 static int mlx4_ib_del_gid(struct ib_device *device,
326                            u8 port_num,
327                            unsigned int index,
328                            void **context)
329 {
330         struct gid_cache_context *ctx = *context;
331         struct mlx4_ib_dev *ibdev = to_mdev(device);
332         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
333         struct mlx4_port_gid_table   *port_gid_table;
334         int ret = 0;
335         int hw_update = 0;
336         struct gid_entry *gids = NULL;
337
338         if (!rdma_cap_roce_gid_table(device, port_num))
339                 return -EINVAL;
340
341         if (port_num > MLX4_MAX_PORTS)
342                 return -EINVAL;
343
344         port_gid_table = &iboe->gids[port_num - 1];
345         spin_lock_bh(&iboe->lock);
346         if (ctx) {
347                 ctx->refcount--;
348                 if (!ctx->refcount) {
349                         unsigned int real_index = ctx->real_index;
350
351                         memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
352                         kfree(port_gid_table->gids[real_index].ctx);
353                         port_gid_table->gids[real_index].ctx = NULL;
354                         hw_update = 1;
355                 }
356         }
357         if (!ret && hw_update) {
358                 int i;
359
360                 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
361                 if (!gids) {
362                         ret = -ENOMEM;
363                 } else {
364                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
365                                 memcpy(&gids[i].gid,
366                                        &port_gid_table->gids[i].gid,
367                                        sizeof(union ib_gid));
368                                 gids[i].gid_type =
369                                     port_gid_table->gids[i].gid_type;
370                         }
371                 }
372         }
373         spin_unlock_bh(&iboe->lock);
374
375         if (!ret && hw_update) {
376                 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
377                 kfree(gids);
378         }
379         return ret;
380 }
381
382 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
383                                     u8 port_num, int index)
384 {
385         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
386         struct gid_cache_context *ctx = NULL;
387         union ib_gid gid;
388         struct mlx4_port_gid_table   *port_gid_table;
389         int real_index = -EINVAL;
390         int i;
391         int ret;
392         unsigned long flags;
393         struct ib_gid_attr attr;
394
395         if (port_num > MLX4_MAX_PORTS)
396                 return -EINVAL;
397
398         if (mlx4_is_bonded(ibdev->dev))
399                 port_num = 1;
400
401         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
402                 return index;
403
404         ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
405         if (ret)
406                 return ret;
407
408         if (attr.ndev)
409                 dev_put(attr.ndev);
410
411         if (!memcmp(&gid, &zgid, sizeof(gid)))
412                 return -EINVAL;
413
414         spin_lock_irqsave(&iboe->lock, flags);
415         port_gid_table = &iboe->gids[port_num - 1];
416
417         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
418                 if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
419                     attr.gid_type == port_gid_table->gids[i].gid_type) {
420                         ctx = port_gid_table->gids[i].ctx;
421                         break;
422                 }
423         if (ctx)
424                 real_index = ctx->real_index;
425         spin_unlock_irqrestore(&iboe->lock, flags);
426         return real_index;
427 }
428
429 static int mlx4_ib_query_device(struct ib_device *ibdev,
430                                 struct ib_device_attr *props,
431                                 struct ib_udata *uhw)
432 {
433         struct mlx4_ib_dev *dev = to_mdev(ibdev);
434         struct ib_smp *in_mad  = NULL;
435         struct ib_smp *out_mad = NULL;
436         int err = -ENOMEM;
437         int have_ib_ports;
438         struct mlx4_uverbs_ex_query_device cmd;
439         struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
440         struct mlx4_clock_params clock_params;
441
442         if (uhw->inlen) {
443                 if (uhw->inlen < sizeof(cmd))
444                         return -EINVAL;
445
446                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
447                 if (err)
448                         return err;
449
450                 if (cmd.comp_mask)
451                         return -EINVAL;
452
453                 if (cmd.reserved)
454                         return -EINVAL;
455         }
456
457         resp.response_length = offsetof(typeof(resp), response_length) +
458                 sizeof(resp.response_length);
459         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
460         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
461         if (!in_mad || !out_mad)
462                 goto out;
463
464         init_query_mad(in_mad);
465         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
466
467         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
468                            1, NULL, NULL, in_mad, out_mad);
469         if (err)
470                 goto out;
471
472         memset(props, 0, sizeof *props);
473
474         have_ib_ports = num_ib_ports(dev->dev);
475
476         props->fw_ver = dev->dev->caps.fw_ver;
477         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
478                 IB_DEVICE_PORT_ACTIVE_EVENT             |
479                 IB_DEVICE_SYS_IMAGE_GUID                |
480                 IB_DEVICE_RC_RNR_NAK_GEN                |
481                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
482         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
483                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
484         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
485                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
486         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
487                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
488         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
489                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
490         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
491                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
492         if (dev->dev->caps.max_gso_sz &&
493             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
494             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
495                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
496         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
497                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
498         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
499             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
500             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
501                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
502         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
503                 props->device_cap_flags |= IB_DEVICE_XRC;
504         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
505                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
506         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
507                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
508                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
509                 else
510                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
511         }
512         if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
513                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
514
515         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
516
517         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
518                 0xffffff;
519         props->vendor_part_id      = dev->dev->persist->pdev->device;
520         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
521         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
522
523         props->max_mr_size         = ~0ull;
524         props->page_size_cap       = dev->dev->caps.page_size_cap;
525         props->max_qp              = dev->dev->quotas.qp;
526         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
527         props->max_sge             = min(dev->dev->caps.max_sq_sg,
528                                          dev->dev->caps.max_rq_sg);
529         props->max_sge_rd          = MLX4_MAX_SGE_RD;
530         props->max_cq              = dev->dev->quotas.cq;
531         props->max_cqe             = dev->dev->caps.max_cqes;
532         props->max_mr              = dev->dev->quotas.mpt;
533         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
534         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
535         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
536         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
537         props->max_srq             = dev->dev->quotas.srq;
538         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
539         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
540         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
541         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
542         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
543                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
544         props->masked_atomic_cap   = props->atomic_cap;
545         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
546         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
547         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
548         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
549                                            props->max_mcast_grp;
550         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
551         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
552         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
553
554         if (!mlx4_is_slave(dev->dev))
555                 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
556
557         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
558                 resp.response_length += sizeof(resp.hca_core_clock_offset);
559                 if (!err && !mlx4_is_slave(dev->dev)) {
560                         resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
561                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
562                 }
563         }
564
565         if (uhw->outlen) {
566                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
567                 if (err)
568                         goto out;
569         }
570 out:
571         kfree(in_mad);
572         kfree(out_mad);
573
574         return err;
575 }
576
577 static enum rdma_link_layer
578 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
579 {
580         struct mlx4_dev *dev = to_mdev(device)->dev;
581
582         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
583                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
584 }
585
586 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
587                               struct ib_port_attr *props, int netw_view)
588 {
589         struct ib_smp *in_mad  = NULL;
590         struct ib_smp *out_mad = NULL;
591         int ext_active_speed;
592         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
593         int err = -ENOMEM;
594
595         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
596         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
597         if (!in_mad || !out_mad)
598                 goto out;
599
600         init_query_mad(in_mad);
601         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
602         in_mad->attr_mod = cpu_to_be32(port);
603
604         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
605                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
606
607         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
608                                 in_mad, out_mad);
609         if (err)
610                 goto out;
611
612
613         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
614         props->lmc              = out_mad->data[34] & 0x7;
615         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
616         props->sm_sl            = out_mad->data[36] & 0xf;
617         props->state            = out_mad->data[32] & 0xf;
618         props->phys_state       = out_mad->data[33] >> 4;
619         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
620         if (netw_view)
621                 props->gid_tbl_len = out_mad->data[50];
622         else
623                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
624         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
625         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
626         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
627         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
628         props->active_width     = out_mad->data[31] & 0xf;
629         props->active_speed     = out_mad->data[35] >> 4;
630         props->max_mtu          = out_mad->data[41] & 0xf;
631         props->active_mtu       = out_mad->data[36] >> 4;
632         props->subnet_timeout   = out_mad->data[51] & 0x1f;
633         props->max_vl_num       = out_mad->data[37] >> 4;
634         props->init_type_reply  = out_mad->data[41] >> 4;
635
636         /* Check if extended speeds (EDR/FDR/...) are supported */
637         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
638                 ext_active_speed = out_mad->data[62] >> 4;
639
640                 switch (ext_active_speed) {
641                 case 1:
642                         props->active_speed = IB_SPEED_FDR;
643                         break;
644                 case 2:
645                         props->active_speed = IB_SPEED_EDR;
646                         break;
647                 }
648         }
649
650         /* If reported active speed is QDR, check if is FDR-10 */
651         if (props->active_speed == IB_SPEED_QDR) {
652                 init_query_mad(in_mad);
653                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
654                 in_mad->attr_mod = cpu_to_be32(port);
655
656                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
657                                    NULL, NULL, in_mad, out_mad);
658                 if (err)
659                         goto out;
660
661                 /* Checking LinkSpeedActive for FDR-10 */
662                 if (out_mad->data[15] & 0x1)
663                         props->active_speed = IB_SPEED_FDR10;
664         }
665
666         /* Avoid wrong speed value returned by FW if the IB link is down. */
667         if (props->state == IB_PORT_DOWN)
668                  props->active_speed = IB_SPEED_SDR;
669
670 out:
671         kfree(in_mad);
672         kfree(out_mad);
673         return err;
674 }
675
676 static u8 state_to_phys_state(enum ib_port_state state)
677 {
678         return state == IB_PORT_ACTIVE ? 5 : 3;
679 }
680
681 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
682                                struct ib_port_attr *props, int netw_view)
683 {
684
685         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
686         struct mlx4_ib_iboe *iboe = &mdev->iboe;
687         struct net_device *ndev;
688         enum ib_mtu tmp;
689         struct mlx4_cmd_mailbox *mailbox;
690         int err = 0;
691         int is_bonded = mlx4_is_bonded(mdev->dev);
692
693         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
694         if (IS_ERR(mailbox))
695                 return PTR_ERR(mailbox);
696
697         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
698                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
699                            MLX4_CMD_WRAPPED);
700         if (err)
701                 goto out;
702
703         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
704                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
705                                            IB_WIDTH_4X : IB_WIDTH_1X;
706         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
707                                            IB_SPEED_FDR : IB_SPEED_QDR;
708         props->port_cap_flags   = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
709         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
710         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
711         props->pkey_tbl_len     = 1;
712         props->max_mtu          = IB_MTU_4096;
713         props->max_vl_num       = 2;
714         props->state            = IB_PORT_DOWN;
715         props->phys_state       = state_to_phys_state(props->state);
716         props->active_mtu       = IB_MTU_256;
717         spin_lock_bh(&iboe->lock);
718         ndev = iboe->netdevs[port - 1];
719         if (ndev && is_bonded) {
720                 rcu_read_lock(); /* required to get upper dev */
721                 ndev = netdev_master_upper_dev_get_rcu(ndev);
722                 rcu_read_unlock();
723         }
724         if (!ndev)
725                 goto out_unlock;
726
727         tmp = iboe_get_mtu(ndev->mtu);
728         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
729
730         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
731                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
732         props->phys_state       = state_to_phys_state(props->state);
733 out_unlock:
734         spin_unlock_bh(&iboe->lock);
735 out:
736         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
737         return err;
738 }
739
740 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
741                          struct ib_port_attr *props, int netw_view)
742 {
743         int err;
744
745         memset(props, 0, sizeof *props);
746
747         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
748                 ib_link_query_port(ibdev, port, props, netw_view) :
749                                 eth_link_query_port(ibdev, port, props, netw_view);
750
751         return err;
752 }
753
754 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
755                               struct ib_port_attr *props)
756 {
757         /* returns host view */
758         return __mlx4_ib_query_port(ibdev, port, props, 0);
759 }
760
761 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
762                         union ib_gid *gid, int netw_view)
763 {
764         struct ib_smp *in_mad  = NULL;
765         struct ib_smp *out_mad = NULL;
766         int err = -ENOMEM;
767         struct mlx4_ib_dev *dev = to_mdev(ibdev);
768         int clear = 0;
769         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
770
771         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
772         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
773         if (!in_mad || !out_mad)
774                 goto out;
775
776         init_query_mad(in_mad);
777         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
778         in_mad->attr_mod = cpu_to_be32(port);
779
780         if (mlx4_is_mfunc(dev->dev) && netw_view)
781                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
782
783         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
784         if (err)
785                 goto out;
786
787         memcpy(gid->raw, out_mad->data + 8, 8);
788
789         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
790                 if (index) {
791                         /* For any index > 0, return the null guid */
792                         err = 0;
793                         clear = 1;
794                         goto out;
795                 }
796         }
797
798         init_query_mad(in_mad);
799         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
800         in_mad->attr_mod = cpu_to_be32(index / 8);
801
802         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
803                            NULL, NULL, in_mad, out_mad);
804         if (err)
805                 goto out;
806
807         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
808
809 out:
810         if (clear)
811                 memset(gid->raw + 8, 0, 8);
812         kfree(in_mad);
813         kfree(out_mad);
814         return err;
815 }
816
817 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
818                              union ib_gid *gid)
819 {
820         int ret;
821
822         if (rdma_protocol_ib(ibdev, port))
823                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
824
825         if (!rdma_protocol_roce(ibdev, port))
826                 return -ENODEV;
827
828         if (!rdma_cap_roce_gid_table(ibdev, port))
829                 return -ENODEV;
830
831         ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
832         if (ret == -EAGAIN) {
833                 memcpy(gid, &zgid, sizeof(*gid));
834                 return 0;
835         }
836
837         return ret;
838 }
839
840 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
841 {
842         union sl2vl_tbl_to_u64 sl2vl64;
843         struct ib_smp *in_mad  = NULL;
844         struct ib_smp *out_mad = NULL;
845         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
846         int err = -ENOMEM;
847         int jj;
848
849         if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
850                 *sl2vl_tbl = 0;
851                 return 0;
852         }
853
854         in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
855         out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
856         if (!in_mad || !out_mad)
857                 goto out;
858
859         init_query_mad(in_mad);
860         in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
861         in_mad->attr_mod = 0;
862
863         if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
864                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
865
866         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
867                            in_mad, out_mad);
868         if (err)
869                 goto out;
870
871         for (jj = 0; jj < 8; jj++)
872                 sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
873         *sl2vl_tbl = sl2vl64.sl64;
874
875 out:
876         kfree(in_mad);
877         kfree(out_mad);
878         return err;
879 }
880
881 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
882 {
883         u64 sl2vl;
884         int i;
885         int err;
886
887         for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
888                 if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
889                         continue;
890                 err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
891                 if (err) {
892                         pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
893                                i, err);
894                         sl2vl = 0;
895                 }
896                 atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
897         }
898 }
899
900 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
901                          u16 *pkey, int netw_view)
902 {
903         struct ib_smp *in_mad  = NULL;
904         struct ib_smp *out_mad = NULL;
905         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
906         int err = -ENOMEM;
907
908         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
909         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
910         if (!in_mad || !out_mad)
911                 goto out;
912
913         init_query_mad(in_mad);
914         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
915         in_mad->attr_mod = cpu_to_be32(index / 32);
916
917         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
918                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
919
920         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
921                            in_mad, out_mad);
922         if (err)
923                 goto out;
924
925         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
926
927 out:
928         kfree(in_mad);
929         kfree(out_mad);
930         return err;
931 }
932
933 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
934 {
935         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
936 }
937
938 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
939                                  struct ib_device_modify *props)
940 {
941         struct mlx4_cmd_mailbox *mailbox;
942         unsigned long flags;
943
944         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
945                 return -EOPNOTSUPP;
946
947         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
948                 return 0;
949
950         if (mlx4_is_slave(to_mdev(ibdev)->dev))
951                 return -EOPNOTSUPP;
952
953         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
954         memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
955         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
956
957         /*
958          * If possible, pass node desc to FW, so it can generate
959          * a 144 trap.  If cmd fails, just ignore.
960          */
961         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
962         if (IS_ERR(mailbox))
963                 return 0;
964
965         memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
966         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
967                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
968
969         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
970
971         return 0;
972 }
973
974 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
975                             u32 cap_mask)
976 {
977         struct mlx4_cmd_mailbox *mailbox;
978         int err;
979
980         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
981         if (IS_ERR(mailbox))
982                 return PTR_ERR(mailbox);
983
984         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
985                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
986                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
987         } else {
988                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
989                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
990         }
991
992         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
993                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
994                        MLX4_CMD_WRAPPED);
995
996         mlx4_free_cmd_mailbox(dev->dev, mailbox);
997         return err;
998 }
999
1000 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1001                                struct ib_port_modify *props)
1002 {
1003         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1004         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1005         struct ib_port_attr attr;
1006         u32 cap_mask;
1007         int err;
1008
1009         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
1010          * of whether port link layer is ETH or IB. For ETH ports, qkey
1011          * violations and port capabilities are not meaningful.
1012          */
1013         if (is_eth)
1014                 return 0;
1015
1016         mutex_lock(&mdev->cap_mask_mutex);
1017
1018         err = mlx4_ib_query_port(ibdev, port, &attr);
1019         if (err)
1020                 goto out;
1021
1022         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1023                 ~props->clr_port_cap_mask;
1024
1025         err = mlx4_ib_SET_PORT(mdev, port,
1026                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
1027                                cap_mask);
1028
1029 out:
1030         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1031         return err;
1032 }
1033
1034 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1035                                                   struct ib_udata *udata)
1036 {
1037         struct mlx4_ib_dev *dev = to_mdev(ibdev);
1038         struct mlx4_ib_ucontext *context;
1039         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1040         struct mlx4_ib_alloc_ucontext_resp resp;
1041         int err;
1042
1043         if (!dev->ib_active)
1044                 return ERR_PTR(-EAGAIN);
1045
1046         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1047                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1048                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1049                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1050         } else {
1051                 resp.dev_caps         = dev->dev->caps.userspace_caps;
1052                 resp.qp_tab_size      = dev->dev->caps.num_qps;
1053                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1054                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1055                 resp.cqe_size         = dev->dev->caps.cqe_size;
1056         }
1057
1058         context = kzalloc(sizeof(*context), GFP_KERNEL);
1059         if (!context)
1060                 return ERR_PTR(-ENOMEM);
1061
1062         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1063         if (err) {
1064                 kfree(context);
1065                 return ERR_PTR(err);
1066         }
1067
1068         INIT_LIST_HEAD(&context->db_page_list);
1069         mutex_init(&context->db_page_mutex);
1070
1071         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1072                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1073         else
1074                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1075
1076         if (err) {
1077                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1078                 kfree(context);
1079                 return ERR_PTR(-EFAULT);
1080         }
1081
1082         return &context->ibucontext;
1083 }
1084
1085 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1086 {
1087         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1088
1089         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1090         kfree(context);
1091
1092         return 0;
1093 }
1094
1095 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1096 {
1097         /* vma_open is called when a new VMA is created on top of our VMA.
1098          * This is done through either mremap flow or split_vma (usually due
1099          * to mlock, madvise, munmap, etc.). We do not support a clone of the
1100          * vma, as this VMA is strongly hardware related. Therefore we set the
1101          * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1102          * calling us again and trying to do incorrect actions. We assume that
1103          * the original vma size is exactly a single page that there will be no
1104          * "splitting" operations on.
1105          */
1106         area->vm_ops = NULL;
1107 }
1108
1109 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1110 {
1111         struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1112
1113         /* It's guaranteed that all VMAs opened on a FD are closed before the
1114          * file itself is closed, therefore no sync is needed with the regular
1115          * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1116          * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1117          * The close operation is usually called under mm->mmap_sem except when
1118          * process is exiting.  The exiting case is handled explicitly as part
1119          * of mlx4_ib_disassociate_ucontext.
1120          */
1121         mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1122                                 area->vm_private_data;
1123
1124         /* set the vma context pointer to null in the mlx4_ib driver's private
1125          * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1126          */
1127         mlx4_ib_vma_priv_data->vma = NULL;
1128 }
1129
1130 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1131         .open = mlx4_ib_vma_open,
1132         .close = mlx4_ib_vma_close
1133 };
1134
1135 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1136 {
1137         int i;
1138         int ret = 0;
1139         struct vm_area_struct *vma;
1140         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1141         struct task_struct *owning_process  = NULL;
1142         struct mm_struct   *owning_mm       = NULL;
1143
1144         owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1145         if (!owning_process)
1146                 return;
1147
1148         owning_mm = get_task_mm(owning_process);
1149         if (!owning_mm) {
1150                 pr_info("no mm, disassociate ucontext is pending task termination\n");
1151                 while (1) {
1152                         /* make sure that task is dead before returning, it may
1153                          * prevent a rare case of module down in parallel to a
1154                          * call to mlx4_ib_vma_close.
1155                          */
1156                         put_task_struct(owning_process);
1157                         msleep(1);
1158                         owning_process = get_pid_task(ibcontext->tgid,
1159                                                       PIDTYPE_PID);
1160                         if (!owning_process ||
1161                             owning_process->state == TASK_DEAD) {
1162                                 pr_info("disassociate ucontext done, task was terminated\n");
1163                                 /* in case task was dead need to release the task struct */
1164                                 if (owning_process)
1165                                         put_task_struct(owning_process);
1166                                 return;
1167                         }
1168                 }
1169         }
1170
1171         /* need to protect from a race on closing the vma as part of
1172          * mlx4_ib_vma_close().
1173          */
1174         down_write(&owning_mm->mmap_sem);
1175         if (!mmget_still_valid(owning_mm))
1176                 goto skip_mm;
1177         for (i = 0; i < HW_BAR_COUNT; i++) {
1178                 vma = context->hw_bar_info[i].vma;
1179                 if (!vma)
1180                         continue;
1181
1182                 ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1183                                    context->hw_bar_info[i].vma->vm_start,
1184                                    PAGE_SIZE);
1185                 if (ret) {
1186                         pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1187                         BUG_ON(1);
1188                 }
1189
1190                 context->hw_bar_info[i].vma->vm_flags &=
1191                         ~(VM_SHARED | VM_MAYSHARE);
1192                 /* context going to be destroyed, should not access ops any more */
1193                 context->hw_bar_info[i].vma->vm_ops = NULL;
1194         }
1195 skip_mm:
1196         up_write(&owning_mm->mmap_sem);
1197         mmput(owning_mm);
1198         put_task_struct(owning_process);
1199 }
1200
1201 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1202                                  struct mlx4_ib_vma_private_data *vma_private_data)
1203 {
1204         vma_private_data->vma = vma;
1205         vma->vm_private_data = vma_private_data;
1206         vma->vm_ops =  &mlx4_ib_vm_ops;
1207 }
1208
1209 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1210 {
1211         struct mlx4_ib_dev *dev = to_mdev(context->device);
1212         struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1213
1214         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1215                 return -EINVAL;
1216
1217         if (vma->vm_pgoff == 0) {
1218                 /* We prevent double mmaping on same context */
1219                 if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1220                         return -EINVAL;
1221
1222                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1223
1224                 if (io_remap_pfn_range(vma, vma->vm_start,
1225                                        to_mucontext(context)->uar.pfn,
1226                                        PAGE_SIZE, vma->vm_page_prot))
1227                         return -EAGAIN;
1228
1229                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1230
1231         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1232                 /* We prevent double mmaping on same context */
1233                 if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1234                         return -EINVAL;
1235
1236                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1237
1238                 if (io_remap_pfn_range(vma, vma->vm_start,
1239                                        to_mucontext(context)->uar.pfn +
1240                                        dev->dev->caps.num_uars,
1241                                        PAGE_SIZE, vma->vm_page_prot))
1242                         return -EAGAIN;
1243
1244                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1245
1246         } else if (vma->vm_pgoff == 3) {
1247                 struct mlx4_clock_params params;
1248                 int ret;
1249
1250                 /* We prevent double mmaping on same context */
1251                 if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1252                         return -EINVAL;
1253
1254                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1255
1256                 if (ret)
1257                         return ret;
1258
1259                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1260                 if (io_remap_pfn_range(vma, vma->vm_start,
1261                                        (pci_resource_start(dev->dev->persist->pdev,
1262                                                            params.bar) +
1263                                         params.offset)
1264                                        >> PAGE_SHIFT,
1265                                        PAGE_SIZE, vma->vm_page_prot))
1266                         return -EAGAIN;
1267
1268                 mlx4_ib_set_vma_data(vma,
1269                                      &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1270         } else {
1271                 return -EINVAL;
1272         }
1273
1274         return 0;
1275 }
1276
1277 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1278                                       struct ib_ucontext *context,
1279                                       struct ib_udata *udata)
1280 {
1281         struct mlx4_ib_pd *pd;
1282         int err;
1283
1284         pd = kmalloc(sizeof *pd, GFP_KERNEL);
1285         if (!pd)
1286                 return ERR_PTR(-ENOMEM);
1287
1288         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1289         if (err) {
1290                 kfree(pd);
1291                 return ERR_PTR(err);
1292         }
1293
1294         if (context)
1295                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1296                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1297                         kfree(pd);
1298                         return ERR_PTR(-EFAULT);
1299                 }
1300
1301         return &pd->ibpd;
1302 }
1303
1304 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1305 {
1306         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1307         kfree(pd);
1308
1309         return 0;
1310 }
1311
1312 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1313                                           struct ib_ucontext *context,
1314                                           struct ib_udata *udata)
1315 {
1316         struct mlx4_ib_xrcd *xrcd;
1317         struct ib_cq_init_attr cq_attr = {};
1318         int err;
1319
1320         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1321                 return ERR_PTR(-ENOSYS);
1322
1323         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1324         if (!xrcd)
1325                 return ERR_PTR(-ENOMEM);
1326
1327         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1328         if (err)
1329                 goto err1;
1330
1331         xrcd->pd = ib_alloc_pd(ibdev, 0);
1332         if (IS_ERR(xrcd->pd)) {
1333                 err = PTR_ERR(xrcd->pd);
1334                 goto err2;
1335         }
1336
1337         cq_attr.cqe = 1;
1338         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1339         if (IS_ERR(xrcd->cq)) {
1340                 err = PTR_ERR(xrcd->cq);
1341                 goto err3;
1342         }
1343
1344         return &xrcd->ibxrcd;
1345
1346 err3:
1347         ib_dealloc_pd(xrcd->pd);
1348 err2:
1349         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1350 err1:
1351         kfree(xrcd);
1352         return ERR_PTR(err);
1353 }
1354
1355 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1356 {
1357         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1358         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1359         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1360         kfree(xrcd);
1361
1362         return 0;
1363 }
1364
1365 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1366 {
1367         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1368         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1369         struct mlx4_ib_gid_entry *ge;
1370
1371         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1372         if (!ge)
1373                 return -ENOMEM;
1374
1375         ge->gid = *gid;
1376         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1377                 ge->port = mqp->port;
1378                 ge->added = 1;
1379         }
1380
1381         mutex_lock(&mqp->mutex);
1382         list_add_tail(&ge->list, &mqp->gid_list);
1383         mutex_unlock(&mqp->mutex);
1384
1385         return 0;
1386 }
1387
1388 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1389                                           struct mlx4_ib_counters *ctr_table)
1390 {
1391         struct counter_index *counter, *tmp_count;
1392
1393         mutex_lock(&ctr_table->mutex);
1394         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1395                                  list) {
1396                 if (counter->allocated)
1397                         mlx4_counter_free(ibdev->dev, counter->index);
1398                 list_del(&counter->list);
1399                 kfree(counter);
1400         }
1401         mutex_unlock(&ctr_table->mutex);
1402 }
1403
1404 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1405                    union ib_gid *gid)
1406 {
1407         struct net_device *ndev;
1408         int ret = 0;
1409
1410         if (!mqp->port)
1411                 return 0;
1412
1413         spin_lock_bh(&mdev->iboe.lock);
1414         ndev = mdev->iboe.netdevs[mqp->port - 1];
1415         if (ndev)
1416                 dev_hold(ndev);
1417         spin_unlock_bh(&mdev->iboe.lock);
1418
1419         if (ndev) {
1420                 ret = 1;
1421                 dev_put(ndev);
1422         }
1423
1424         return ret;
1425 }
1426
1427 struct mlx4_ib_steering {
1428         struct list_head list;
1429         struct mlx4_flow_reg_id reg_id;
1430         union ib_gid gid;
1431 };
1432
1433 #define LAST_ETH_FIELD vlan_tag
1434 #define LAST_IB_FIELD sl
1435 #define LAST_IPV4_FIELD dst_ip
1436 #define LAST_TCP_UDP_FIELD src_port
1437
1438 /* Field is the last supported field */
1439 #define FIELDS_NOT_SUPPORTED(filter, field)\
1440         memchr_inv((void *)&filter.field  +\
1441                    sizeof(filter.field), 0,\
1442                    sizeof(filter) -\
1443                    offsetof(typeof(filter), field) -\
1444                    sizeof(filter.field))
1445
1446 static int parse_flow_attr(struct mlx4_dev *dev,
1447                            u32 qp_num,
1448                            union ib_flow_spec *ib_spec,
1449                            struct _rule_hw *mlx4_spec)
1450 {
1451         enum mlx4_net_trans_rule_id type;
1452
1453         switch (ib_spec->type) {
1454         case IB_FLOW_SPEC_ETH:
1455                 if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1456                         return -ENOTSUPP;
1457
1458                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1459                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1460                        ETH_ALEN);
1461                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1462                        ETH_ALEN);
1463                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1464                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1465                 break;
1466         case IB_FLOW_SPEC_IB:
1467                 if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1468                         return -ENOTSUPP;
1469
1470                 type = MLX4_NET_TRANS_RULE_ID_IB;
1471                 mlx4_spec->ib.l3_qpn =
1472                         cpu_to_be32(qp_num);
1473                 mlx4_spec->ib.qpn_mask =
1474                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1475                 break;
1476
1477
1478         case IB_FLOW_SPEC_IPV4:
1479                 if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1480                         return -ENOTSUPP;
1481
1482                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1483                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1484                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1485                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1486                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1487                 break;
1488
1489         case IB_FLOW_SPEC_TCP:
1490         case IB_FLOW_SPEC_UDP:
1491                 if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1492                         return -ENOTSUPP;
1493
1494                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1495                                         MLX4_NET_TRANS_RULE_ID_TCP :
1496                                         MLX4_NET_TRANS_RULE_ID_UDP;
1497                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1498                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1499                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1500                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1501                 break;
1502
1503         default:
1504                 return -EINVAL;
1505         }
1506         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1507             mlx4_hw_rule_sz(dev, type) < 0)
1508                 return -EINVAL;
1509         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1510         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1511         return mlx4_hw_rule_sz(dev, type);
1512 }
1513
1514 struct default_rules {
1515         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1516         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1517         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1518         __u8  link_layer;
1519 };
1520 static const struct default_rules default_table[] = {
1521         {
1522                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1523                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1524                 .rules_create_list = {IB_FLOW_SPEC_IB},
1525                 .link_layer = IB_LINK_LAYER_INFINIBAND
1526         }
1527 };
1528
1529 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1530                                          struct ib_flow_attr *flow_attr)
1531 {
1532         int i, j, k;
1533         void *ib_flow;
1534         const struct default_rules *pdefault_rules = default_table;
1535         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1536
1537         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1538                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1539                 memset(&field_types, 0, sizeof(field_types));
1540
1541                 if (link_layer != pdefault_rules->link_layer)
1542                         continue;
1543
1544                 ib_flow = flow_attr + 1;
1545                 /* we assume the specs are sorted */
1546                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1547                      j < flow_attr->num_of_specs; k++) {
1548                         union ib_flow_spec *current_flow =
1549                                 (union ib_flow_spec *)ib_flow;
1550
1551                         /* same layer but different type */
1552                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1553                              (pdefault_rules->mandatory_fields[k] &
1554                               IB_FLOW_SPEC_LAYER_MASK)) &&
1555                             (current_flow->type !=
1556                              pdefault_rules->mandatory_fields[k]))
1557                                 goto out;
1558
1559                         /* same layer, try match next one */
1560                         if (current_flow->type ==
1561                             pdefault_rules->mandatory_fields[k]) {
1562                                 j++;
1563                                 ib_flow +=
1564                                         ((union ib_flow_spec *)ib_flow)->size;
1565                         }
1566                 }
1567
1568                 ib_flow = flow_attr + 1;
1569                 for (j = 0; j < flow_attr->num_of_specs;
1570                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1571                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1572                                 /* same layer and same type */
1573                                 if (((union ib_flow_spec *)ib_flow)->type ==
1574                                     pdefault_rules->mandatory_not_fields[k])
1575                                         goto out;
1576
1577                 return i;
1578         }
1579 out:
1580         return -1;
1581 }
1582
1583 static int __mlx4_ib_create_default_rules(
1584                 struct mlx4_ib_dev *mdev,
1585                 struct ib_qp *qp,
1586                 const struct default_rules *pdefault_rules,
1587                 struct _rule_hw *mlx4_spec) {
1588         int size = 0;
1589         int i;
1590
1591         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1592                 union ib_flow_spec ib_spec = {};
1593                 int ret;
1594
1595                 switch (pdefault_rules->rules_create_list[i]) {
1596                 case 0:
1597                         /* no rule */
1598                         continue;
1599                 case IB_FLOW_SPEC_IB:
1600                         ib_spec.type = IB_FLOW_SPEC_IB;
1601                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1602
1603                         break;
1604                 default:
1605                         /* invalid rule */
1606                         return -EINVAL;
1607                 }
1608                 /* We must put empty rule, qpn is being ignored */
1609                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1610                                       mlx4_spec);
1611                 if (ret < 0) {
1612                         pr_info("invalid parsing\n");
1613                         return -EINVAL;
1614                 }
1615
1616                 mlx4_spec = (void *)mlx4_spec + ret;
1617                 size += ret;
1618         }
1619         return size;
1620 }
1621
1622 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1623                           int domain,
1624                           enum mlx4_net_trans_promisc_mode flow_type,
1625                           u64 *reg_id)
1626 {
1627         int ret, i;
1628         int size = 0;
1629         void *ib_flow;
1630         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1631         struct mlx4_cmd_mailbox *mailbox;
1632         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1633         int default_flow;
1634
1635         static const u16 __mlx4_domain[] = {
1636                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1637                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1638                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1639                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1640         };
1641
1642         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1643                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1644                 return -EINVAL;
1645         }
1646
1647         if (domain >= IB_FLOW_DOMAIN_NUM) {
1648                 pr_err("Invalid domain value %d\n", domain);
1649                 return -EINVAL;
1650         }
1651
1652         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1653                 return -EINVAL;
1654
1655         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1656         if (IS_ERR(mailbox))
1657                 return PTR_ERR(mailbox);
1658         ctrl = mailbox->buf;
1659
1660         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1661                                  flow_attr->priority);
1662         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1663         ctrl->port = flow_attr->port;
1664         ctrl->qpn = cpu_to_be32(qp->qp_num);
1665
1666         ib_flow = flow_attr + 1;
1667         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1668         /* Add default flows */
1669         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1670         if (default_flow >= 0) {
1671                 ret = __mlx4_ib_create_default_rules(
1672                                 mdev, qp, default_table + default_flow,
1673                                 mailbox->buf + size);
1674                 if (ret < 0) {
1675                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1676                         return -EINVAL;
1677                 }
1678                 size += ret;
1679         }
1680         for (i = 0; i < flow_attr->num_of_specs; i++) {
1681                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1682                                       mailbox->buf + size);
1683                 if (ret < 0) {
1684                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1685                         return -EINVAL;
1686                 }
1687                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1688                 size += ret;
1689         }
1690
1691         if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1692             flow_attr->num_of_specs == 1) {
1693                 struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1694                 enum ib_flow_spec_type header_spec =
1695                         ((union ib_flow_spec *)(flow_attr + 1))->type;
1696
1697                 if (header_spec == IB_FLOW_SPEC_ETH)
1698                         mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1699         }
1700
1701         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1702                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1703                            MLX4_CMD_NATIVE);
1704         if (ret == -ENOMEM)
1705                 pr_err("mcg table is full. Fail to register network rule.\n");
1706         else if (ret == -ENXIO)
1707                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1708         else if (ret)
1709                 pr_err("Invalid argument. Fail to register network rule.\n");
1710
1711         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1712         return ret;
1713 }
1714
1715 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1716 {
1717         int err;
1718         err = mlx4_cmd(dev, reg_id, 0, 0,
1719                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1720                        MLX4_CMD_NATIVE);
1721         if (err)
1722                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1723                        reg_id);
1724         return err;
1725 }
1726
1727 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1728                                     u64 *reg_id)
1729 {
1730         void *ib_flow;
1731         union ib_flow_spec *ib_spec;
1732         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1733         int err = 0;
1734
1735         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1736             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1737                 return 0; /* do nothing */
1738
1739         ib_flow = flow_attr + 1;
1740         ib_spec = (union ib_flow_spec *)ib_flow;
1741
1742         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1743                 return 0; /* do nothing */
1744
1745         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1746                                     flow_attr->port, qp->qp_num,
1747                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1748                                     reg_id);
1749         return err;
1750 }
1751
1752 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1753                                       struct ib_flow_attr *flow_attr,
1754                                       enum mlx4_net_trans_promisc_mode *type)
1755 {
1756         int err = 0;
1757
1758         if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1759             (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1760             (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1761                 return -EOPNOTSUPP;
1762         }
1763
1764         if (flow_attr->num_of_specs == 0) {
1765                 type[0] = MLX4_FS_MC_SNIFFER;
1766                 type[1] = MLX4_FS_UC_SNIFFER;
1767         } else {
1768                 union ib_flow_spec *ib_spec;
1769
1770                 ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1771                 if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1772                         return -EINVAL;
1773
1774                 /* if all is zero than MC and UC */
1775                 if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1776                         type[0] = MLX4_FS_MC_SNIFFER;
1777                         type[1] = MLX4_FS_UC_SNIFFER;
1778                 } else {
1779                         u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1780                                             ib_spec->eth.mask.dst_mac[1],
1781                                             ib_spec->eth.mask.dst_mac[2],
1782                                             ib_spec->eth.mask.dst_mac[3],
1783                                             ib_spec->eth.mask.dst_mac[4],
1784                                             ib_spec->eth.mask.dst_mac[5]};
1785
1786                         /* Above xor was only on MC bit, non empty mask is valid
1787                          * only if this bit is set and rest are zero.
1788                          */
1789                         if (!is_zero_ether_addr(&mac[0]))
1790                                 return -EINVAL;
1791
1792                         if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1793                                 type[0] = MLX4_FS_MC_SNIFFER;
1794                         else
1795                                 type[0] = MLX4_FS_UC_SNIFFER;
1796                 }
1797         }
1798
1799         return err;
1800 }
1801
1802 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1803                                     struct ib_flow_attr *flow_attr,
1804                                     int domain)
1805 {
1806         int err = 0, i = 0, j = 0;
1807         struct mlx4_ib_flow *mflow;
1808         enum mlx4_net_trans_promisc_mode type[2];
1809         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1810         int is_bonded = mlx4_is_bonded(dev);
1811
1812         if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1813                 return ERR_PTR(-EINVAL);
1814
1815         if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1816             (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1817                 return ERR_PTR(-EOPNOTSUPP);
1818
1819         memset(type, 0, sizeof(type));
1820
1821         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1822         if (!mflow) {
1823                 err = -ENOMEM;
1824                 goto err_free;
1825         }
1826
1827         switch (flow_attr->type) {
1828         case IB_FLOW_ATTR_NORMAL:
1829                 /* If dont trap flag (continue match) is set, under specific
1830                  * condition traffic be replicated to given qp,
1831                  * without stealing it
1832                  */
1833                 if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1834                         err = mlx4_ib_add_dont_trap_rule(dev,
1835                                                          flow_attr,
1836                                                          type);
1837                         if (err)
1838                                 goto err_free;
1839                 } else {
1840                         type[0] = MLX4_FS_REGULAR;
1841                 }
1842                 break;
1843
1844         case IB_FLOW_ATTR_ALL_DEFAULT:
1845                 type[0] = MLX4_FS_ALL_DEFAULT;
1846                 break;
1847
1848         case IB_FLOW_ATTR_MC_DEFAULT:
1849                 type[0] = MLX4_FS_MC_DEFAULT;
1850                 break;
1851
1852         case IB_FLOW_ATTR_SNIFFER:
1853                 type[0] = MLX4_FS_MIRROR_RX_PORT;
1854                 type[1] = MLX4_FS_MIRROR_SX_PORT;
1855                 break;
1856
1857         default:
1858                 err = -EINVAL;
1859                 goto err_free;
1860         }
1861
1862         while (i < ARRAY_SIZE(type) && type[i]) {
1863                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1864                                             &mflow->reg_id[i].id);
1865                 if (err)
1866                         goto err_create_flow;
1867                 if (is_bonded) {
1868                         /* Application always sees one port so the mirror rule
1869                          * must be on port #2
1870                          */
1871                         flow_attr->port = 2;
1872                         err = __mlx4_ib_create_flow(qp, flow_attr,
1873                                                     domain, type[j],
1874                                                     &mflow->reg_id[j].mirror);
1875                         flow_attr->port = 1;
1876                         if (err)
1877                                 goto err_create_flow;
1878                         j++;
1879                 }
1880
1881                 i++;
1882         }
1883
1884         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1885                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1886                                                &mflow->reg_id[i].id);
1887                 if (err)
1888                         goto err_create_flow;
1889
1890                 if (is_bonded) {
1891                         flow_attr->port = 2;
1892                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1893                                                        &mflow->reg_id[j].mirror);
1894                         flow_attr->port = 1;
1895                         if (err)
1896                                 goto err_create_flow;
1897                         j++;
1898                 }
1899                 /* function to create mirror rule */
1900                 i++;
1901         }
1902
1903         return &mflow->ibflow;
1904
1905 err_create_flow:
1906         while (i) {
1907                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1908                                              mflow->reg_id[i].id);
1909                 i--;
1910         }
1911
1912         while (j) {
1913                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1914                                              mflow->reg_id[j].mirror);
1915                 j--;
1916         }
1917 err_free:
1918         kfree(mflow);
1919         return ERR_PTR(err);
1920 }
1921
1922 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1923 {
1924         int err, ret = 0;
1925         int i = 0;
1926         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1927         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1928
1929         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1930                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1931                 if (err)
1932                         ret = err;
1933                 if (mflow->reg_id[i].mirror) {
1934                         err = __mlx4_ib_destroy_flow(mdev->dev,
1935                                                      mflow->reg_id[i].mirror);
1936                         if (err)
1937                                 ret = err;
1938                 }
1939                 i++;
1940         }
1941
1942         kfree(mflow);
1943         return ret;
1944 }
1945
1946 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1947 {
1948         int err;
1949         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1950         struct mlx4_dev *dev = mdev->dev;
1951         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1952         struct mlx4_ib_steering *ib_steering = NULL;
1953         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1954         struct mlx4_flow_reg_id reg_id;
1955
1956         if (mdev->dev->caps.steering_mode ==
1957             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1958                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1959                 if (!ib_steering)
1960                         return -ENOMEM;
1961         }
1962
1963         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1964                                     !!(mqp->flags &
1965                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1966                                     prot, &reg_id.id);
1967         if (err) {
1968                 pr_err("multicast attach op failed, err %d\n", err);
1969                 goto err_malloc;
1970         }
1971
1972         reg_id.mirror = 0;
1973         if (mlx4_is_bonded(dev)) {
1974                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1975                                             (mqp->port == 1) ? 2 : 1,
1976                                             !!(mqp->flags &
1977                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1978                                             prot, &reg_id.mirror);
1979                 if (err)
1980                         goto err_add;
1981         }
1982
1983         err = add_gid_entry(ibqp, gid);
1984         if (err)
1985                 goto err_add;
1986
1987         if (ib_steering) {
1988                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1989                 ib_steering->reg_id = reg_id;
1990                 mutex_lock(&mqp->mutex);
1991                 list_add(&ib_steering->list, &mqp->steering_rules);
1992                 mutex_unlock(&mqp->mutex);
1993         }
1994         return 0;
1995
1996 err_add:
1997         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1998                               prot, reg_id.id);
1999         if (reg_id.mirror)
2000                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2001                                       prot, reg_id.mirror);
2002 err_malloc:
2003         kfree(ib_steering);
2004
2005         return err;
2006 }
2007
2008 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2009 {
2010         struct mlx4_ib_gid_entry *ge;
2011         struct mlx4_ib_gid_entry *tmp;
2012         struct mlx4_ib_gid_entry *ret = NULL;
2013
2014         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2015                 if (!memcmp(raw, ge->gid.raw, 16)) {
2016                         ret = ge;
2017                         break;
2018                 }
2019         }
2020
2021         return ret;
2022 }
2023
2024 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2025 {
2026         int err;
2027         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2028         struct mlx4_dev *dev = mdev->dev;
2029         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2030         struct net_device *ndev;
2031         struct mlx4_ib_gid_entry *ge;
2032         struct mlx4_flow_reg_id reg_id = {0, 0};
2033         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2034
2035         if (mdev->dev->caps.steering_mode ==
2036             MLX4_STEERING_MODE_DEVICE_MANAGED) {
2037                 struct mlx4_ib_steering *ib_steering;
2038
2039                 mutex_lock(&mqp->mutex);
2040                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2041                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2042                                 list_del(&ib_steering->list);
2043                                 break;
2044                         }
2045                 }
2046                 mutex_unlock(&mqp->mutex);
2047                 if (&ib_steering->list == &mqp->steering_rules) {
2048                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2049                         return -EINVAL;
2050                 }
2051                 reg_id = ib_steering->reg_id;
2052                 kfree(ib_steering);
2053         }
2054
2055         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2056                                     prot, reg_id.id);
2057         if (err)
2058                 return err;
2059
2060         if (mlx4_is_bonded(dev)) {
2061                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2062                                             prot, reg_id.mirror);
2063                 if (err)
2064                         return err;
2065         }
2066
2067         mutex_lock(&mqp->mutex);
2068         ge = find_gid_entry(mqp, gid->raw);
2069         if (ge) {
2070                 spin_lock_bh(&mdev->iboe.lock);
2071                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2072                 if (ndev)
2073                         dev_hold(ndev);
2074                 spin_unlock_bh(&mdev->iboe.lock);
2075                 if (ndev)
2076                         dev_put(ndev);
2077                 list_del(&ge->list);
2078                 kfree(ge);
2079         } else
2080                 pr_warn("could not find mgid entry\n");
2081
2082         mutex_unlock(&mqp->mutex);
2083
2084         return 0;
2085 }
2086
2087 static int init_node_data(struct mlx4_ib_dev *dev)
2088 {
2089         struct ib_smp *in_mad  = NULL;
2090         struct ib_smp *out_mad = NULL;
2091         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2092         int err = -ENOMEM;
2093
2094         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2095         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2096         if (!in_mad || !out_mad)
2097                 goto out;
2098
2099         init_query_mad(in_mad);
2100         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2101         if (mlx4_is_master(dev->dev))
2102                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2103
2104         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2105         if (err)
2106                 goto out;
2107
2108         memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2109
2110         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2111
2112         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2113         if (err)
2114                 goto out;
2115
2116         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2117         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2118
2119 out:
2120         kfree(in_mad);
2121         kfree(out_mad);
2122         return err;
2123 }
2124
2125 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2126                         char *buf)
2127 {
2128         struct mlx4_ib_dev *dev =
2129                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2130         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2131 }
2132
2133 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2134                         char *buf)
2135 {
2136         struct mlx4_ib_dev *dev =
2137                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2138         return sprintf(buf, "%x\n", dev->dev->rev_id);
2139 }
2140
2141 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2142                           char *buf)
2143 {
2144         struct mlx4_ib_dev *dev =
2145                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2146         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2147                        dev->dev->board_id);
2148 }
2149
2150 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2151 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2152 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2153
2154 static struct device_attribute *mlx4_class_attributes[] = {
2155         &dev_attr_hw_rev,
2156         &dev_attr_hca_type,
2157         &dev_attr_board_id
2158 };
2159
2160 struct diag_counter {
2161         const char *name;
2162         u32 offset;
2163 };
2164
2165 #define DIAG_COUNTER(_name, _offset)                    \
2166         { .name = #_name, .offset = _offset }
2167
2168 static const struct diag_counter diag_basic[] = {
2169         DIAG_COUNTER(rq_num_lle, 0x00),
2170         DIAG_COUNTER(sq_num_lle, 0x04),
2171         DIAG_COUNTER(rq_num_lqpoe, 0x08),
2172         DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2173         DIAG_COUNTER(rq_num_lpe, 0x18),
2174         DIAG_COUNTER(sq_num_lpe, 0x1C),
2175         DIAG_COUNTER(rq_num_wrfe, 0x20),
2176         DIAG_COUNTER(sq_num_wrfe, 0x24),
2177         DIAG_COUNTER(sq_num_mwbe, 0x2C),
2178         DIAG_COUNTER(sq_num_bre, 0x34),
2179         DIAG_COUNTER(sq_num_rire, 0x44),
2180         DIAG_COUNTER(rq_num_rire, 0x48),
2181         DIAG_COUNTER(sq_num_rae, 0x4C),
2182         DIAG_COUNTER(rq_num_rae, 0x50),
2183         DIAG_COUNTER(sq_num_roe, 0x54),
2184         DIAG_COUNTER(sq_num_tree, 0x5C),
2185         DIAG_COUNTER(sq_num_rree, 0x64),
2186         DIAG_COUNTER(rq_num_rnr, 0x68),
2187         DIAG_COUNTER(sq_num_rnr, 0x6C),
2188         DIAG_COUNTER(rq_num_oos, 0x100),
2189         DIAG_COUNTER(sq_num_oos, 0x104),
2190 };
2191
2192 static const struct diag_counter diag_ext[] = {
2193         DIAG_COUNTER(rq_num_dup, 0x130),
2194         DIAG_COUNTER(sq_num_to, 0x134),
2195 };
2196
2197 static const struct diag_counter diag_device_only[] = {
2198         DIAG_COUNTER(num_cqovf, 0x1A0),
2199         DIAG_COUNTER(rq_num_udsdprd, 0x118),
2200 };
2201
2202 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2203                                                     u8 port_num)
2204 {
2205         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2206         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2207
2208         if (!diag[!!port_num].name)
2209                 return NULL;
2210
2211         return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2212                                           diag[!!port_num].num_counters,
2213                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
2214 }
2215
2216 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2217                                 struct rdma_hw_stats *stats,
2218                                 u8 port, int index)
2219 {
2220         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2221         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2222         u32 hw_value[ARRAY_SIZE(diag_device_only) +
2223                 ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2224         int ret;
2225         int i;
2226
2227         ret = mlx4_query_diag_counters(dev->dev,
2228                                        MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2229                                        diag[!!port].offset, hw_value,
2230                                        diag[!!port].num_counters, port);
2231
2232         if (ret)
2233                 return ret;
2234
2235         for (i = 0; i < diag[!!port].num_counters; i++)
2236                 stats->value[i] = hw_value[i];
2237
2238         return diag[!!port].num_counters;
2239 }
2240
2241 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2242                                          const char ***name,
2243                                          u32 **offset,
2244                                          u32 *num,
2245                                          bool port)
2246 {
2247         u32 num_counters;
2248
2249         num_counters = ARRAY_SIZE(diag_basic);
2250
2251         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2252                 num_counters += ARRAY_SIZE(diag_ext);
2253
2254         if (!port)
2255                 num_counters += ARRAY_SIZE(diag_device_only);
2256
2257         *name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2258         if (!*name)
2259                 return -ENOMEM;
2260
2261         *offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2262         if (!*offset)
2263                 goto err_name;
2264
2265         *num = num_counters;
2266
2267         return 0;
2268
2269 err_name:
2270         kfree(*name);
2271         return -ENOMEM;
2272 }
2273
2274 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2275                                        const char **name,
2276                                        u32 *offset,
2277                                        bool port)
2278 {
2279         int i;
2280         int j;
2281
2282         for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2283                 name[i] = diag_basic[i].name;
2284                 offset[i] = diag_basic[i].offset;
2285         }
2286
2287         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2288                 for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2289                         name[j] = diag_ext[i].name;
2290                         offset[j] = diag_ext[i].offset;
2291                 }
2292         }
2293
2294         if (!port) {
2295                 for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2296                         name[j] = diag_device_only[i].name;
2297                         offset[j] = diag_device_only[i].offset;
2298                 }
2299         }
2300 }
2301
2302 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2303 {
2304         struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2305         int i;
2306         int ret;
2307         bool per_port = !!(ibdev->dev->caps.flags2 &
2308                 MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2309
2310         if (mlx4_is_slave(ibdev->dev))
2311                 return 0;
2312
2313         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2314                 /* i == 1 means we are building port counters */
2315                 if (i && !per_port)
2316                         continue;
2317
2318                 ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2319                                                     &diag[i].offset,
2320                                                     &diag[i].num_counters, i);
2321                 if (ret)
2322                         goto err_alloc;
2323
2324                 mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2325                                            diag[i].offset, i);
2326         }
2327
2328         ibdev->ib_dev.get_hw_stats      = mlx4_ib_get_hw_stats;
2329         ibdev->ib_dev.alloc_hw_stats    = mlx4_ib_alloc_hw_stats;
2330
2331         return 0;
2332
2333 err_alloc:
2334         if (i) {
2335                 kfree(diag[i - 1].name);
2336                 kfree(diag[i - 1].offset);
2337         }
2338
2339         return ret;
2340 }
2341
2342 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2343 {
2344         int i;
2345
2346         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2347                 kfree(ibdev->diag_counters[i].offset);
2348                 kfree(ibdev->diag_counters[i].name);
2349         }
2350 }
2351
2352 #define MLX4_IB_INVALID_MAC     ((u64)-1)
2353 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2354                                struct net_device *dev,
2355                                int port)
2356 {
2357         u64 new_smac = 0;
2358         u64 release_mac = MLX4_IB_INVALID_MAC;
2359         struct mlx4_ib_qp *qp;
2360
2361         read_lock(&dev_base_lock);
2362         new_smac = mlx4_mac_to_u64(dev->dev_addr);
2363         read_unlock(&dev_base_lock);
2364
2365         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2366
2367         /* no need for update QP1 and mac registration in non-SRIOV */
2368         if (!mlx4_is_mfunc(ibdev->dev))
2369                 return;
2370
2371         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2372         qp = ibdev->qp1_proxy[port - 1];
2373         if (qp) {
2374                 int new_smac_index;
2375                 u64 old_smac;
2376                 struct mlx4_update_qp_params update_params;
2377
2378                 mutex_lock(&qp->mutex);
2379                 old_smac = qp->pri.smac;
2380                 if (new_smac == old_smac)
2381                         goto unlock;
2382
2383                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2384
2385                 if (new_smac_index < 0)
2386                         goto unlock;
2387
2388                 update_params.smac_index = new_smac_index;
2389                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2390                                    &update_params)) {
2391                         release_mac = new_smac;
2392                         goto unlock;
2393                 }
2394                 /* if old port was zero, no mac was yet registered for this QP */
2395                 if (qp->pri.smac_port)
2396                         release_mac = old_smac;
2397                 qp->pri.smac = new_smac;
2398                 qp->pri.smac_port = port;
2399                 qp->pri.smac_index = new_smac_index;
2400         }
2401
2402 unlock:
2403         if (release_mac != MLX4_IB_INVALID_MAC)
2404                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
2405         if (qp)
2406                 mutex_unlock(&qp->mutex);
2407         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2408 }
2409
2410 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2411                                  struct net_device *dev,
2412                                  unsigned long event)
2413
2414 {
2415         struct mlx4_ib_iboe *iboe;
2416         int update_qps_port = -1;
2417         int port;
2418
2419         ASSERT_RTNL();
2420
2421         iboe = &ibdev->iboe;
2422
2423         spin_lock_bh(&iboe->lock);
2424         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2425
2426                 iboe->netdevs[port - 1] =
2427                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2428
2429                 if (dev == iboe->netdevs[port - 1] &&
2430                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2431                      event == NETDEV_UP || event == NETDEV_CHANGE))
2432                         update_qps_port = port;
2433
2434         }
2435         spin_unlock_bh(&iboe->lock);
2436
2437         if (update_qps_port > 0)
2438                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2439 }
2440
2441 static int mlx4_ib_netdev_event(struct notifier_block *this,
2442                                 unsigned long event, void *ptr)
2443 {
2444         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2445         struct mlx4_ib_dev *ibdev;
2446
2447         if (!net_eq(dev_net(dev), &init_net))
2448                 return NOTIFY_DONE;
2449
2450         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2451         mlx4_ib_scan_netdevs(ibdev, dev, event);
2452
2453         return NOTIFY_DONE;
2454 }
2455
2456 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2457 {
2458         int port;
2459         int slave;
2460         int i;
2461
2462         if (mlx4_is_master(ibdev->dev)) {
2463                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2464                      ++slave) {
2465                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2466                                 for (i = 0;
2467                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2468                                      ++i) {
2469                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2470                                         /* master has the identity virt2phys pkey mapping */
2471                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2472                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2473                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2474                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2475                                 }
2476                         }
2477                 }
2478                 /* initialize pkey cache */
2479                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2480                         for (i = 0;
2481                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2482                              ++i)
2483                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2484                                         (i) ? 0 : 0xFFFF;
2485                 }
2486         }
2487 }
2488
2489 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2490 {
2491         int i, j, eq = 0, total_eqs = 0;
2492
2493         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2494                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2495         if (!ibdev->eq_table)
2496                 return;
2497
2498         for (i = 1; i <= dev->caps.num_ports; i++) {
2499                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2500                      j++, total_eqs++) {
2501                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2502                                 continue;
2503                         ibdev->eq_table[eq] = total_eqs;
2504                         if (!mlx4_assign_eq(dev, i,
2505                                             &ibdev->eq_table[eq]))
2506                                 eq++;
2507                         else
2508                                 ibdev->eq_table[eq] = -1;
2509                 }
2510         }
2511
2512         for (i = eq; i < dev->caps.num_comp_vectors;
2513              ibdev->eq_table[i++] = -1)
2514                 ;
2515
2516         /* Advertise the new number of EQs to clients */
2517         ibdev->ib_dev.num_comp_vectors = eq;
2518 }
2519
2520 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2521 {
2522         int i;
2523         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2524
2525         /* no eqs were allocated */
2526         if (!ibdev->eq_table)
2527                 return;
2528
2529         /* Reset the advertised EQ number */
2530         ibdev->ib_dev.num_comp_vectors = 0;
2531
2532         for (i = 0; i < total_eqs; i++)
2533                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2534
2535         kfree(ibdev->eq_table);
2536         ibdev->eq_table = NULL;
2537 }
2538
2539 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2540                                struct ib_port_immutable *immutable)
2541 {
2542         struct ib_port_attr attr;
2543         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2544         int err;
2545
2546         err = mlx4_ib_query_port(ibdev, port_num, &attr);
2547         if (err)
2548                 return err;
2549
2550         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2551         immutable->gid_tbl_len = attr.gid_tbl_len;
2552
2553         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2554                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2555         } else {
2556                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2557                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2558                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2559                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2560                                 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2561         }
2562
2563         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2564
2565         return 0;
2566 }
2567
2568 static void get_fw_ver_str(struct ib_device *device, char *str,
2569                            size_t str_len)
2570 {
2571         struct mlx4_ib_dev *dev =
2572                 container_of(device, struct mlx4_ib_dev, ib_dev);
2573         snprintf(str, str_len, "%d.%d.%d",
2574                  (int) (dev->dev->caps.fw_ver >> 32),
2575                  (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2576                  (int) dev->dev->caps.fw_ver & 0xffff);
2577 }
2578
2579 static void *mlx4_ib_add(struct mlx4_dev *dev)
2580 {
2581         struct mlx4_ib_dev *ibdev;
2582         int num_ports = 0;
2583         int i, j;
2584         int err;
2585         struct mlx4_ib_iboe *iboe;
2586         int ib_num_ports = 0;
2587         int num_req_counters;
2588         int allocated;
2589         u32 counter_index;
2590         struct counter_index *new_counter_index = NULL;
2591
2592         pr_info_once("%s", mlx4_ib_version);
2593
2594         num_ports = 0;
2595         mlx4_foreach_ib_transport_port(i, dev)
2596                 num_ports++;
2597
2598         /* No point in registering a device with no ports... */
2599         if (num_ports == 0)
2600                 return NULL;
2601
2602         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2603         if (!ibdev) {
2604                 dev_err(&dev->persist->pdev->dev,
2605                         "Device struct alloc failed\n");
2606                 return NULL;
2607         }
2608
2609         iboe = &ibdev->iboe;
2610
2611         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2612                 goto err_dealloc;
2613
2614         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2615                 goto err_pd;
2616
2617         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2618                                  PAGE_SIZE);
2619         if (!ibdev->uar_map)
2620                 goto err_uar;
2621         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2622
2623         ibdev->dev = dev;
2624         ibdev->bond_next_port   = 0;
2625
2626         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2627         ibdev->ib_dev.owner             = THIS_MODULE;
2628         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2629         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2630         ibdev->num_ports                = num_ports;
2631         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2632                                                 1 : ibdev->num_ports;
2633         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2634         ibdev->ib_dev.dma_device        = &dev->persist->pdev->dev;
2635         ibdev->ib_dev.get_netdev        = mlx4_ib_get_netdev;
2636         ibdev->ib_dev.add_gid           = mlx4_ib_add_gid;
2637         ibdev->ib_dev.del_gid           = mlx4_ib_del_gid;
2638
2639         if (dev->caps.userspace_caps)
2640                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2641         else
2642                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2643
2644         ibdev->ib_dev.uverbs_cmd_mask   =
2645                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2646                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2647                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2648                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2649                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2650                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2651                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2652                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2653                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2654                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2655                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2656                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2657                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2658                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2659                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2660                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2661                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2662                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2663                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2664                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2665                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2666                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2667                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2668                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2669
2670         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2671         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2672         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2673         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2674         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2675         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2676         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2677         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2678         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2679         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2680         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2681         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2682         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2683         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2684         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2685         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2686         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2687         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2688         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2689         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2690         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2691         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2692         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2693         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2694         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2695         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2696         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2697         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2698         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2699         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2700         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2701         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2702         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2703         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2704         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2705         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2706         ibdev->ib_dev.alloc_mr          = mlx4_ib_alloc_mr;
2707         ibdev->ib_dev.map_mr_sg         = mlx4_ib_map_mr_sg;
2708         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2709         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2710         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2711         ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2712         ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2713         ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2714
2715         if (!mlx4_is_slave(ibdev->dev)) {
2716                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2717                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2718                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2719                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2720         }
2721
2722         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2723             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2724                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2725                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2726
2727                 ibdev->ib_dev.uverbs_cmd_mask |=
2728                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2729                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2730         }
2731
2732         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2733                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2734                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2735                 ibdev->ib_dev.uverbs_cmd_mask |=
2736                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2737                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2738         }
2739
2740         if (check_flow_steering_support(dev)) {
2741                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2742                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2743                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2744
2745                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2746                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2747                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2748         }
2749
2750         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2751                 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2752                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2753                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2754
2755         mlx4_ib_alloc_eqs(dev, ibdev);
2756
2757         spin_lock_init(&iboe->lock);
2758
2759         if (init_node_data(ibdev))
2760                 goto err_map;
2761         mlx4_init_sl2vl_tbl(ibdev);
2762
2763         for (i = 0; i < ibdev->num_ports; ++i) {
2764                 mutex_init(&ibdev->counters_table[i].mutex);
2765                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2766         }
2767
2768         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2769         for (i = 0; i < num_req_counters; ++i) {
2770                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2771                 allocated = 0;
2772                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2773                                                 IB_LINK_LAYER_ETHERNET) {
2774                         err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2775                         /* if failed to allocate a new counter, use default */
2776                         if (err)
2777                                 counter_index =
2778                                         mlx4_get_default_counter_index(dev,
2779                                                                        i + 1);
2780                         else
2781                                 allocated = 1;
2782                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2783                         counter_index = mlx4_get_default_counter_index(dev,
2784                                                                        i + 1);
2785                 }
2786                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2787                                             GFP_KERNEL);
2788                 if (!new_counter_index) {
2789                         if (allocated)
2790                                 mlx4_counter_free(ibdev->dev, counter_index);
2791                         goto err_counter;
2792                 }
2793                 new_counter_index->index = counter_index;
2794                 new_counter_index->allocated = allocated;
2795                 list_add_tail(&new_counter_index->list,
2796                               &ibdev->counters_table[i].counters_list);
2797                 ibdev->counters_table[i].default_counter = counter_index;
2798                 pr_info("counter index %d for port %d allocated %d\n",
2799                         counter_index, i + 1, allocated);
2800         }
2801         if (mlx4_is_bonded(dev))
2802                 for (i = 1; i < ibdev->num_ports ; ++i) {
2803                         new_counter_index =
2804                                         kmalloc(sizeof(struct counter_index),
2805                                                 GFP_KERNEL);
2806                         if (!new_counter_index)
2807                                 goto err_counter;
2808                         new_counter_index->index = counter_index;
2809                         new_counter_index->allocated = 0;
2810                         list_add_tail(&new_counter_index->list,
2811                                       &ibdev->counters_table[i].counters_list);
2812                         ibdev->counters_table[i].default_counter =
2813                                                                 counter_index;
2814                 }
2815
2816         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2817                 ib_num_ports++;
2818
2819         spin_lock_init(&ibdev->sm_lock);
2820         mutex_init(&ibdev->cap_mask_mutex);
2821         INIT_LIST_HEAD(&ibdev->qp_list);
2822         spin_lock_init(&ibdev->reset_flow_resource_lock);
2823
2824         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2825             ib_num_ports) {
2826                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2827                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2828                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2829                                             &ibdev->steer_qpn_base, 0);
2830                 if (err)
2831                         goto err_counter;
2832
2833                 ibdev->ib_uc_qpns_bitmap =
2834                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2835                                 sizeof(long),
2836                                 GFP_KERNEL);
2837                 if (!ibdev->ib_uc_qpns_bitmap) {
2838                         dev_err(&dev->persist->pdev->dev,
2839                                 "bit map alloc failed\n");
2840                         goto err_steer_qp_release;
2841                 }
2842
2843                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2844                         bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2845                                     ibdev->steer_qpn_count);
2846                         err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2847                                         dev, ibdev->steer_qpn_base,
2848                                         ibdev->steer_qpn_base +
2849                                         ibdev->steer_qpn_count - 1);
2850                         if (err)
2851                                 goto err_steer_free_bitmap;
2852                 } else {
2853                         bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2854                                     ibdev->steer_qpn_count);
2855                 }
2856         }
2857
2858         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2859                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2860
2861         if (mlx4_ib_alloc_diag_counters(ibdev))
2862                 goto err_steer_free_bitmap;
2863
2864         if (ib_register_device(&ibdev->ib_dev, NULL))
2865                 goto err_diag_counters;
2866
2867         if (mlx4_ib_mad_init(ibdev))
2868                 goto err_reg;
2869
2870         if (mlx4_ib_init_sriov(ibdev))
2871                 goto err_mad;
2872
2873         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2874             dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2875                 if (!iboe->nb.notifier_call) {
2876                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2877                         err = register_netdevice_notifier(&iboe->nb);
2878                         if (err) {
2879                                 iboe->nb.notifier_call = NULL;
2880                                 goto err_notif;
2881                         }
2882                 }
2883                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2884                         err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2885                         if (err) {
2886                                 goto err_notif;
2887                         }
2888                 }
2889         }
2890
2891         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2892                 if (device_create_file(&ibdev->ib_dev.dev,
2893                                        mlx4_class_attributes[j]))
2894                         goto err_notif;
2895         }
2896
2897         ibdev->ib_active = true;
2898         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2899                 devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2900                                          &ibdev->ib_dev);
2901
2902         if (mlx4_is_mfunc(ibdev->dev))
2903                 init_pkeys(ibdev);
2904
2905         /* create paravirt contexts for any VFs which are active */
2906         if (mlx4_is_master(ibdev->dev)) {
2907                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2908                         if (j == mlx4_master_func_num(ibdev->dev))
2909                                 continue;
2910                         if (mlx4_is_slave_active(ibdev->dev, j))
2911                                 do_slave_init(ibdev, j, 1);
2912                 }
2913         }
2914         return ibdev;
2915
2916 err_notif:
2917         if (ibdev->iboe.nb.notifier_call) {
2918                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2919                         pr_warn("failure unregistering notifier\n");
2920                 ibdev->iboe.nb.notifier_call = NULL;
2921         }
2922         flush_workqueue(wq);
2923
2924         mlx4_ib_close_sriov(ibdev);
2925
2926 err_mad:
2927         mlx4_ib_mad_cleanup(ibdev);
2928
2929 err_reg:
2930         ib_unregister_device(&ibdev->ib_dev);
2931
2932 err_diag_counters:
2933         mlx4_ib_diag_cleanup(ibdev);
2934
2935 err_steer_free_bitmap:
2936         kfree(ibdev->ib_uc_qpns_bitmap);
2937
2938 err_steer_qp_release:
2939         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2940                               ibdev->steer_qpn_count);
2941 err_counter:
2942         for (i = 0; i < ibdev->num_ports; ++i)
2943                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2944
2945 err_map:
2946         mlx4_ib_free_eqs(dev, ibdev);
2947         iounmap(ibdev->uar_map);
2948
2949 err_uar:
2950         mlx4_uar_free(dev, &ibdev->priv_uar);
2951
2952 err_pd:
2953         mlx4_pd_free(dev, ibdev->priv_pdn);
2954
2955 err_dealloc:
2956         ib_dealloc_device(&ibdev->ib_dev);
2957
2958         return NULL;
2959 }
2960
2961 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2962 {
2963         int offset;
2964
2965         WARN_ON(!dev->ib_uc_qpns_bitmap);
2966
2967         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2968                                          dev->steer_qpn_count,
2969                                          get_count_order(count));
2970         if (offset < 0)
2971                 return offset;
2972
2973         *qpn = dev->steer_qpn_base + offset;
2974         return 0;
2975 }
2976
2977 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2978 {
2979         if (!qpn ||
2980             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2981                 return;
2982
2983         BUG_ON(qpn < dev->steer_qpn_base);
2984
2985         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2986                               qpn - dev->steer_qpn_base,
2987                               get_count_order(count));
2988 }
2989
2990 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2991                          int is_attach)
2992 {
2993         int err;
2994         size_t flow_size;
2995         struct ib_flow_attr *flow = NULL;
2996         struct ib_flow_spec_ib *ib_spec;
2997
2998         if (is_attach) {
2999                 flow_size = sizeof(struct ib_flow_attr) +
3000                             sizeof(struct ib_flow_spec_ib);
3001                 flow = kzalloc(flow_size, GFP_KERNEL);
3002                 if (!flow)
3003                         return -ENOMEM;
3004                 flow->port = mqp->port;
3005                 flow->num_of_specs = 1;
3006                 flow->size = flow_size;
3007                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3008                 ib_spec->type = IB_FLOW_SPEC_IB;
3009                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
3010                 /* Add an empty rule for IB L2 */
3011                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3012
3013                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3014                                             IB_FLOW_DOMAIN_NIC,
3015                                             MLX4_FS_REGULAR,
3016                                             &mqp->reg_id);
3017         } else {
3018                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3019         }
3020         kfree(flow);
3021         return err;
3022 }
3023
3024 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3025 {
3026         struct mlx4_ib_dev *ibdev = ibdev_ptr;
3027         int p;
3028         int i;
3029
3030         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3031                 devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3032         ibdev->ib_active = false;
3033         flush_workqueue(wq);
3034
3035         if (ibdev->iboe.nb.notifier_call) {
3036                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3037                         pr_warn("failure unregistering notifier\n");
3038                 ibdev->iboe.nb.notifier_call = NULL;
3039         }
3040
3041         mlx4_ib_close_sriov(ibdev);
3042         mlx4_ib_mad_cleanup(ibdev);
3043         ib_unregister_device(&ibdev->ib_dev);
3044         mlx4_ib_diag_cleanup(ibdev);
3045
3046         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3047                               ibdev->steer_qpn_count);
3048         kfree(ibdev->ib_uc_qpns_bitmap);
3049
3050         iounmap(ibdev->uar_map);
3051         for (p = 0; p < ibdev->num_ports; ++p)
3052                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3053
3054         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3055                 mlx4_CLOSE_PORT(dev, p);
3056
3057         mlx4_ib_free_eqs(dev, ibdev);
3058
3059         mlx4_uar_free(dev, &ibdev->priv_uar);
3060         mlx4_pd_free(dev, ibdev->priv_pdn);
3061         ib_dealloc_device(&ibdev->ib_dev);
3062 }
3063
3064 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3065 {
3066         struct mlx4_ib_demux_work **dm = NULL;
3067         struct mlx4_dev *dev = ibdev->dev;
3068         int i;
3069         unsigned long flags;
3070         struct mlx4_active_ports actv_ports;
3071         unsigned int ports;
3072         unsigned int first_port;
3073
3074         if (!mlx4_is_master(dev))
3075                 return;
3076
3077         actv_ports = mlx4_get_active_ports(dev, slave);
3078         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3079         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3080
3081         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3082         if (!dm) {
3083                 pr_err("failed to allocate memory for tunneling qp update\n");
3084                 return;
3085         }
3086
3087         for (i = 0; i < ports; i++) {
3088                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3089                 if (!dm[i]) {
3090                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
3091                         while (--i >= 0)
3092                                 kfree(dm[i]);
3093                         goto out;
3094                 }
3095                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3096                 dm[i]->port = first_port + i + 1;
3097                 dm[i]->slave = slave;
3098                 dm[i]->do_init = do_init;
3099                 dm[i]->dev = ibdev;
3100         }
3101         /* initialize or tear down tunnel QPs for the slave */
3102         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3103         if (!ibdev->sriov.is_going_down) {
3104                 for (i = 0; i < ports; i++)
3105                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3106                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3107         } else {
3108                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3109                 for (i = 0; i < ports; i++)
3110                         kfree(dm[i]);
3111         }
3112 out:
3113         kfree(dm);
3114         return;
3115 }
3116
3117 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3118 {
3119         struct mlx4_ib_qp *mqp;
3120         unsigned long flags_qp;
3121         unsigned long flags_cq;
3122         struct mlx4_ib_cq *send_mcq, *recv_mcq;
3123         struct list_head    cq_notify_list;
3124         struct mlx4_cq *mcq;
3125         unsigned long flags;
3126
3127         pr_warn("mlx4_ib_handle_catas_error was started\n");
3128         INIT_LIST_HEAD(&cq_notify_list);
3129
3130         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3131         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3132
3133         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3134                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3135                 if (mqp->sq.tail != mqp->sq.head) {
3136                         send_mcq = to_mcq(mqp->ibqp.send_cq);
3137                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
3138                         if (send_mcq->mcq.comp &&
3139                             mqp->ibqp.send_cq->comp_handler) {
3140                                 if (!send_mcq->mcq.reset_notify_added) {
3141                                         send_mcq->mcq.reset_notify_added = 1;
3142                                         list_add_tail(&send_mcq->mcq.reset_notify,
3143                                                       &cq_notify_list);
3144                                 }
3145                         }
3146                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3147                 }
3148                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3149                 /* Now, handle the QP's receive queue */
3150                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3151                 /* no handling is needed for SRQ */
3152                 if (!mqp->ibqp.srq) {
3153                         if (mqp->rq.tail != mqp->rq.head) {
3154                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3155                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3156                                 if (recv_mcq->mcq.comp &&
3157                                     mqp->ibqp.recv_cq->comp_handler) {
3158                                         if (!recv_mcq->mcq.reset_notify_added) {
3159                                                 recv_mcq->mcq.reset_notify_added = 1;
3160                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
3161                                                               &cq_notify_list);
3162                                         }
3163                                 }
3164                                 spin_unlock_irqrestore(&recv_mcq->lock,
3165                                                        flags_cq);
3166                         }
3167                 }
3168                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3169         }
3170
3171         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3172                 mcq->comp(mcq);
3173         }
3174         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3175         pr_warn("mlx4_ib_handle_catas_error ended\n");
3176 }
3177
3178 static void handle_bonded_port_state_event(struct work_struct *work)
3179 {
3180         struct ib_event_work *ew =
3181                 container_of(work, struct ib_event_work, work);
3182         struct mlx4_ib_dev *ibdev = ew->ib_dev;
3183         enum ib_port_state bonded_port_state = IB_PORT_NOP;
3184         int i;
3185         struct ib_event ibev;
3186
3187         kfree(ew);
3188         spin_lock_bh(&ibdev->iboe.lock);
3189         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3190                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3191                 enum ib_port_state curr_port_state;
3192
3193                 if (!curr_netdev)
3194                         continue;
3195
3196                 curr_port_state =
3197                         (netif_running(curr_netdev) &&
3198                          netif_carrier_ok(curr_netdev)) ?
3199                         IB_PORT_ACTIVE : IB_PORT_DOWN;
3200
3201                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3202                         curr_port_state : IB_PORT_ACTIVE;
3203         }
3204         spin_unlock_bh(&ibdev->iboe.lock);
3205
3206         ibev.device = &ibdev->ib_dev;
3207         ibev.element.port_num = 1;
3208         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3209                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3210
3211         ib_dispatch_event(&ibev);
3212 }
3213
3214 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3215 {
3216         u64 sl2vl;
3217         int err;
3218
3219         err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3220         if (err) {
3221                 pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3222                        port, err);
3223                 sl2vl = 0;
3224         }
3225         atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3226 }
3227
3228 static void ib_sl2vl_update_work(struct work_struct *work)
3229 {
3230         struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3231         struct mlx4_ib_dev *mdev = ew->ib_dev;
3232         int port = ew->port;
3233
3234         mlx4_ib_sl2vl_update(mdev, port);
3235
3236         kfree(ew);
3237 }
3238
3239 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3240                                      int port)
3241 {
3242         struct ib_event_work *ew;
3243
3244         ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3245         if (ew) {
3246                 INIT_WORK(&ew->work, ib_sl2vl_update_work);
3247                 ew->port = port;
3248                 ew->ib_dev = ibdev;
3249                 queue_work(wq, &ew->work);
3250         } else {
3251                 pr_err("failed to allocate memory for sl2vl update work\n");
3252         }
3253 }
3254
3255 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3256                           enum mlx4_dev_event event, unsigned long param)
3257 {
3258         struct ib_event ibev;
3259         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3260         struct mlx4_eqe *eqe = NULL;
3261         struct ib_event_work *ew;
3262         int p = 0;
3263
3264         if (mlx4_is_bonded(dev) &&
3265             ((event == MLX4_DEV_EVENT_PORT_UP) ||
3266             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3267                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3268                 if (!ew)
3269                         return;
3270                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
3271                 ew->ib_dev = ibdev;
3272                 queue_work(wq, &ew->work);
3273                 return;
3274         }
3275
3276         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3277                 eqe = (struct mlx4_eqe *)param;
3278         else
3279                 p = (int) param;
3280
3281         switch (event) {
3282         case MLX4_DEV_EVENT_PORT_UP:
3283                 if (p > ibdev->num_ports)
3284                         return;
3285                 if (!mlx4_is_slave(dev) &&
3286                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3287                         IB_LINK_LAYER_INFINIBAND) {
3288                         if (mlx4_is_master(dev))
3289                                 mlx4_ib_invalidate_all_guid_record(ibdev, p);
3290                         if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3291                             !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3292                                 mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3293                 }
3294                 ibev.event = IB_EVENT_PORT_ACTIVE;
3295                 break;
3296
3297         case MLX4_DEV_EVENT_PORT_DOWN:
3298                 if (p > ibdev->num_ports)
3299                         return;
3300                 ibev.event = IB_EVENT_PORT_ERR;
3301                 break;
3302
3303         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3304                 ibdev->ib_active = false;
3305                 ibev.event = IB_EVENT_DEVICE_FATAL;
3306                 mlx4_ib_handle_catas_error(ibdev);
3307                 break;
3308
3309         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3310                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3311                 if (!ew) {
3312                         pr_err("failed to allocate memory for events work\n");
3313                         break;
3314                 }
3315
3316                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3317                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3318                 ew->ib_dev = ibdev;
3319                 /* need to queue only for port owner, which uses GEN_EQE */
3320                 if (mlx4_is_master(dev))
3321                         queue_work(wq, &ew->work);
3322                 else
3323                         handle_port_mgmt_change_event(&ew->work);
3324                 return;
3325
3326         case MLX4_DEV_EVENT_SLAVE_INIT:
3327                 /* here, p is the slave id */
3328                 do_slave_init(ibdev, p, 1);
3329                 if (mlx4_is_master(dev)) {
3330                         int i;
3331
3332                         for (i = 1; i <= ibdev->num_ports; i++) {
3333                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3334                                         == IB_LINK_LAYER_INFINIBAND)
3335                                         mlx4_ib_slave_alias_guid_event(ibdev,
3336                                                                        p, i,
3337                                                                        1);
3338                         }
3339                 }
3340                 return;
3341
3342         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3343                 if (mlx4_is_master(dev)) {
3344                         int i;
3345
3346                         for (i = 1; i <= ibdev->num_ports; i++) {
3347                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3348                                         == IB_LINK_LAYER_INFINIBAND)
3349                                         mlx4_ib_slave_alias_guid_event(ibdev,
3350                                                                        p, i,
3351                                                                        0);
3352                         }
3353                 }
3354                 /* here, p is the slave id */
3355                 do_slave_init(ibdev, p, 0);
3356                 return;
3357
3358         default:
3359                 return;
3360         }
3361
3362         ibev.device           = ibdev_ptr;
3363         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3364
3365         ib_dispatch_event(&ibev);
3366 }
3367
3368 static struct mlx4_interface mlx4_ib_interface = {
3369         .add            = mlx4_ib_add,
3370         .remove         = mlx4_ib_remove,
3371         .event          = mlx4_ib_event,
3372         .protocol       = MLX4_PROT_IB_IPV6,
3373         .flags          = MLX4_INTFF_BONDING
3374 };
3375
3376 static int __init mlx4_ib_init(void)
3377 {
3378         int err;
3379
3380         wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3381         if (!wq)
3382                 return -ENOMEM;
3383
3384         err = mlx4_ib_mcg_init();
3385         if (err)
3386                 goto clean_wq;
3387
3388         err = mlx4_register_interface(&mlx4_ib_interface);
3389         if (err)
3390                 goto clean_mcg;
3391
3392         return 0;
3393
3394 clean_mcg:
3395         mlx4_ib_mcg_destroy();
3396
3397 clean_wq:
3398         destroy_workqueue(wq);
3399         return err;
3400 }
3401
3402 static void __exit mlx4_ib_cleanup(void)
3403 {
3404         mlx4_unregister_interface(&mlx4_ib_interface);
3405         mlx4_ib_mcg_destroy();
3406         destroy_workqueue(wq);
3407 }
3408
3409 module_init(mlx4_ib_init);
3410 module_exit(mlx4_ib_cleanup);