Merge tag 'for_linus-3.4-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/jwesse...
[linux-block.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
43 #include <rdma/ib_smi.h>
44 #include <rdma/ib_user_verbs.h>
45 #include <rdma/ib_addr.h>
46
47 #include <linux/mlx4/driver.h>
48 #include <linux/mlx4/cmd.h>
49
50 #include "mlx4_ib.h"
51 #include "user.h"
52
53 #define DRV_NAME        "mlx4_ib"
54 #define DRV_VERSION     "1.0"
55 #define DRV_RELDATE     "April 4, 2008"
56
57 MODULE_AUTHOR("Roland Dreier");
58 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
59 MODULE_LICENSE("Dual BSD/GPL");
60 MODULE_VERSION(DRV_VERSION);
61
62 static const char mlx4_ib_version[] =
63         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
64         DRV_VERSION " (" DRV_RELDATE ")\n";
65
66 struct update_gid_work {
67         struct work_struct      work;
68         union ib_gid            gids[128];
69         struct mlx4_ib_dev     *dev;
70         int                     port;
71 };
72
73 static struct workqueue_struct *wq;
74
75 static void init_query_mad(struct ib_smp *mad)
76 {
77         mad->base_version  = 1;
78         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
79         mad->class_version = 1;
80         mad->method        = IB_MGMT_METHOD_GET;
81 }
82
83 static union ib_gid zgid;
84
85 static int mlx4_ib_query_device(struct ib_device *ibdev,
86                                 struct ib_device_attr *props)
87 {
88         struct mlx4_ib_dev *dev = to_mdev(ibdev);
89         struct ib_smp *in_mad  = NULL;
90         struct ib_smp *out_mad = NULL;
91         int err = -ENOMEM;
92
93         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
94         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
95         if (!in_mad || !out_mad)
96                 goto out;
97
98         init_query_mad(in_mad);
99         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
100
101         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
102         if (err)
103                 goto out;
104
105         memset(props, 0, sizeof *props);
106
107         props->fw_ver = dev->dev->caps.fw_ver;
108         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
109                 IB_DEVICE_PORT_ACTIVE_EVENT             |
110                 IB_DEVICE_SYS_IMAGE_GUID                |
111                 IB_DEVICE_RC_RNR_NAK_GEN                |
112                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
113         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
114                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
115         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
116                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
117         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
118                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
119         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
120                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
121         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
122                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
123         if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
124                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
125         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
126                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
127         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
128             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
129             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
130                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
131         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
132                 props->device_cap_flags |= IB_DEVICE_XRC;
133
134         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
135                 0xffffff;
136         props->vendor_part_id      = be16_to_cpup((__be16 *) (out_mad->data + 30));
137         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
138         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
139
140         props->max_mr_size         = ~0ull;
141         props->page_size_cap       = dev->dev->caps.page_size_cap;
142         props->max_qp              = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
143         props->max_qp_wr           = dev->dev->caps.max_wqes;
144         props->max_sge             = min(dev->dev->caps.max_sq_sg,
145                                          dev->dev->caps.max_rq_sg);
146         props->max_cq              = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
147         props->max_cqe             = dev->dev->caps.max_cqes;
148         props->max_mr              = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
149         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
150         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
151         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
152         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
153         props->max_srq             = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
154         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
155         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
156         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
157         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
158         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
159                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
160         props->masked_atomic_cap   = IB_ATOMIC_HCA;
161         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
162         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
163         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
164         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
165                                            props->max_mcast_grp;
166         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
167
168 out:
169         kfree(in_mad);
170         kfree(out_mad);
171
172         return err;
173 }
174
175 static enum rdma_link_layer
176 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
177 {
178         struct mlx4_dev *dev = to_mdev(device)->dev;
179
180         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
181                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
182 }
183
184 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
185                               struct ib_port_attr *props)
186 {
187         struct ib_smp *in_mad  = NULL;
188         struct ib_smp *out_mad = NULL;
189         int ext_active_speed;
190         int err = -ENOMEM;
191
192         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
193         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
194         if (!in_mad || !out_mad)
195                 goto out;
196
197         init_query_mad(in_mad);
198         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
199         in_mad->attr_mod = cpu_to_be32(port);
200
201         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL,
202                                 in_mad, out_mad);
203         if (err)
204                 goto out;
205
206
207         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
208         props->lmc              = out_mad->data[34] & 0x7;
209         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
210         props->sm_sl            = out_mad->data[36] & 0xf;
211         props->state            = out_mad->data[32] & 0xf;
212         props->phys_state       = out_mad->data[33] >> 4;
213         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
214         props->gid_tbl_len      = to_mdev(ibdev)->dev->caps.gid_table_len[port];
215         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
216         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
217         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
218         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
219         props->active_width     = out_mad->data[31] & 0xf;
220         props->active_speed     = out_mad->data[35] >> 4;
221         props->max_mtu          = out_mad->data[41] & 0xf;
222         props->active_mtu       = out_mad->data[36] >> 4;
223         props->subnet_timeout   = out_mad->data[51] & 0x1f;
224         props->max_vl_num       = out_mad->data[37] >> 4;
225         props->init_type_reply  = out_mad->data[41] >> 4;
226
227         /* Check if extended speeds (EDR/FDR/...) are supported */
228         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
229                 ext_active_speed = out_mad->data[62] >> 4;
230
231                 switch (ext_active_speed) {
232                 case 1:
233                         props->active_speed = IB_SPEED_FDR;
234                         break;
235                 case 2:
236                         props->active_speed = IB_SPEED_EDR;
237                         break;
238                 }
239         }
240
241         /* If reported active speed is QDR, check if is FDR-10 */
242         if (props->active_speed == IB_SPEED_QDR) {
243                 init_query_mad(in_mad);
244                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
245                 in_mad->attr_mod = cpu_to_be32(port);
246
247                 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port,
248                                    NULL, NULL, in_mad, out_mad);
249                 if (err)
250                         return err;
251
252                 /* Checking LinkSpeedActive for FDR-10 */
253                 if (out_mad->data[15] & 0x1)
254                         props->active_speed = IB_SPEED_FDR10;
255         }
256 out:
257         kfree(in_mad);
258         kfree(out_mad);
259         return err;
260 }
261
262 static u8 state_to_phys_state(enum ib_port_state state)
263 {
264         return state == IB_PORT_ACTIVE ? 5 : 3;
265 }
266
267 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
268                                struct ib_port_attr *props)
269 {
270
271         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
272         struct mlx4_ib_iboe *iboe = &mdev->iboe;
273         struct net_device *ndev;
274         enum ib_mtu tmp;
275         struct mlx4_cmd_mailbox *mailbox;
276         int err = 0;
277
278         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
279         if (IS_ERR(mailbox))
280                 return PTR_ERR(mailbox);
281
282         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
283                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
284                            MLX4_CMD_WRAPPED);
285         if (err)
286                 goto out;
287
288         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ?
289                                                 IB_WIDTH_4X : IB_WIDTH_1X;
290         props->active_speed     = IB_SPEED_QDR;
291         props->port_cap_flags   = IB_PORT_CM_SUP;
292         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
293         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
294         props->pkey_tbl_len     = 1;
295         props->max_mtu          = IB_MTU_4096;
296         props->max_vl_num       = 2;
297         props->state            = IB_PORT_DOWN;
298         props->phys_state       = state_to_phys_state(props->state);
299         props->active_mtu       = IB_MTU_256;
300         spin_lock(&iboe->lock);
301         ndev = iboe->netdevs[port - 1];
302         if (!ndev)
303                 goto out_unlock;
304
305         tmp = iboe_get_mtu(ndev->mtu);
306         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
307
308         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
309                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
310         props->phys_state       = state_to_phys_state(props->state);
311 out_unlock:
312         spin_unlock(&iboe->lock);
313 out:
314         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
315         return err;
316 }
317
318 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
319                               struct ib_port_attr *props)
320 {
321         int err;
322
323         memset(props, 0, sizeof *props);
324
325         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
326                 ib_link_query_port(ibdev, port, props) :
327                                 eth_link_query_port(ibdev, port, props);
328
329         return err;
330 }
331
332 static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
333                                union ib_gid *gid)
334 {
335         struct ib_smp *in_mad  = NULL;
336         struct ib_smp *out_mad = NULL;
337         int err = -ENOMEM;
338
339         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
340         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
341         if (!in_mad || !out_mad)
342                 goto out;
343
344         init_query_mad(in_mad);
345         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
346         in_mad->attr_mod = cpu_to_be32(port);
347
348         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
349         if (err)
350                 goto out;
351
352         memcpy(gid->raw, out_mad->data + 8, 8);
353
354         init_query_mad(in_mad);
355         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
356         in_mad->attr_mod = cpu_to_be32(index / 8);
357
358         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
359         if (err)
360                 goto out;
361
362         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
363
364 out:
365         kfree(in_mad);
366         kfree(out_mad);
367         return err;
368 }
369
370 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
371                           union ib_gid *gid)
372 {
373         struct mlx4_ib_dev *dev = to_mdev(ibdev);
374
375         *gid = dev->iboe.gid_table[port - 1][index];
376
377         return 0;
378 }
379
380 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
381                              union ib_gid *gid)
382 {
383         if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
384                 return __mlx4_ib_query_gid(ibdev, port, index, gid);
385         else
386                 return iboe_query_gid(ibdev, port, index, gid);
387 }
388
389 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
390                               u16 *pkey)
391 {
392         struct ib_smp *in_mad  = NULL;
393         struct ib_smp *out_mad = NULL;
394         int err = -ENOMEM;
395
396         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
397         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
398         if (!in_mad || !out_mad)
399                 goto out;
400
401         init_query_mad(in_mad);
402         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
403         in_mad->attr_mod = cpu_to_be32(index / 32);
404
405         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
406         if (err)
407                 goto out;
408
409         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
410
411 out:
412         kfree(in_mad);
413         kfree(out_mad);
414         return err;
415 }
416
417 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
418                                  struct ib_device_modify *props)
419 {
420         struct mlx4_cmd_mailbox *mailbox;
421
422         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
423                 return -EOPNOTSUPP;
424
425         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
426                 return 0;
427
428         spin_lock(&to_mdev(ibdev)->sm_lock);
429         memcpy(ibdev->node_desc, props->node_desc, 64);
430         spin_unlock(&to_mdev(ibdev)->sm_lock);
431
432         /*
433          * If possible, pass node desc to FW, so it can generate
434          * a 144 trap.  If cmd fails, just ignore.
435          */
436         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
437         if (IS_ERR(mailbox))
438                 return 0;
439
440         memset(mailbox->buf, 0, 256);
441         memcpy(mailbox->buf, props->node_desc, 64);
442         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
443                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
444
445         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
446
447         return 0;
448 }
449
450 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
451                          u32 cap_mask)
452 {
453         struct mlx4_cmd_mailbox *mailbox;
454         int err;
455         u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
456
457         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
458         if (IS_ERR(mailbox))
459                 return PTR_ERR(mailbox);
460
461         memset(mailbox->buf, 0, 256);
462
463         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
464                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
465                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
466         } else {
467                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
468                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
469         }
470
471         err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
472                        MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
473
474         mlx4_free_cmd_mailbox(dev->dev, mailbox);
475         return err;
476 }
477
478 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
479                                struct ib_port_modify *props)
480 {
481         struct ib_port_attr attr;
482         u32 cap_mask;
483         int err;
484
485         mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
486
487         err = mlx4_ib_query_port(ibdev, port, &attr);
488         if (err)
489                 goto out;
490
491         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
492                 ~props->clr_port_cap_mask;
493
494         err = mlx4_SET_PORT(to_mdev(ibdev), port,
495                             !!(mask & IB_PORT_RESET_QKEY_CNTR),
496                             cap_mask);
497
498 out:
499         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
500         return err;
501 }
502
503 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
504                                                   struct ib_udata *udata)
505 {
506         struct mlx4_ib_dev *dev = to_mdev(ibdev);
507         struct mlx4_ib_ucontext *context;
508         struct mlx4_ib_alloc_ucontext_resp resp;
509         int err;
510
511         if (!dev->ib_active)
512                 return ERR_PTR(-EAGAIN);
513
514         resp.qp_tab_size      = dev->dev->caps.num_qps;
515         resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
516         resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
517
518         context = kmalloc(sizeof *context, GFP_KERNEL);
519         if (!context)
520                 return ERR_PTR(-ENOMEM);
521
522         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
523         if (err) {
524                 kfree(context);
525                 return ERR_PTR(err);
526         }
527
528         INIT_LIST_HEAD(&context->db_page_list);
529         mutex_init(&context->db_page_mutex);
530
531         err = ib_copy_to_udata(udata, &resp, sizeof resp);
532         if (err) {
533                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
534                 kfree(context);
535                 return ERR_PTR(-EFAULT);
536         }
537
538         return &context->ibucontext;
539 }
540
541 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
542 {
543         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
544
545         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
546         kfree(context);
547
548         return 0;
549 }
550
551 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
552 {
553         struct mlx4_ib_dev *dev = to_mdev(context->device);
554
555         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
556                 return -EINVAL;
557
558         if (vma->vm_pgoff == 0) {
559                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
560
561                 if (io_remap_pfn_range(vma, vma->vm_start,
562                                        to_mucontext(context)->uar.pfn,
563                                        PAGE_SIZE, vma->vm_page_prot))
564                         return -EAGAIN;
565         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
566                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
567
568                 if (io_remap_pfn_range(vma, vma->vm_start,
569                                        to_mucontext(context)->uar.pfn +
570                                        dev->dev->caps.num_uars,
571                                        PAGE_SIZE, vma->vm_page_prot))
572                         return -EAGAIN;
573         } else
574                 return -EINVAL;
575
576         return 0;
577 }
578
579 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
580                                       struct ib_ucontext *context,
581                                       struct ib_udata *udata)
582 {
583         struct mlx4_ib_pd *pd;
584         int err;
585
586         pd = kmalloc(sizeof *pd, GFP_KERNEL);
587         if (!pd)
588                 return ERR_PTR(-ENOMEM);
589
590         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
591         if (err) {
592                 kfree(pd);
593                 return ERR_PTR(err);
594         }
595
596         if (context)
597                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
598                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
599                         kfree(pd);
600                         return ERR_PTR(-EFAULT);
601                 }
602
603         return &pd->ibpd;
604 }
605
606 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
607 {
608         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
609         kfree(pd);
610
611         return 0;
612 }
613
614 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
615                                           struct ib_ucontext *context,
616                                           struct ib_udata *udata)
617 {
618         struct mlx4_ib_xrcd *xrcd;
619         int err;
620
621         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
622                 return ERR_PTR(-ENOSYS);
623
624         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
625         if (!xrcd)
626                 return ERR_PTR(-ENOMEM);
627
628         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
629         if (err)
630                 goto err1;
631
632         xrcd->pd = ib_alloc_pd(ibdev);
633         if (IS_ERR(xrcd->pd)) {
634                 err = PTR_ERR(xrcd->pd);
635                 goto err2;
636         }
637
638         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
639         if (IS_ERR(xrcd->cq)) {
640                 err = PTR_ERR(xrcd->cq);
641                 goto err3;
642         }
643
644         return &xrcd->ibxrcd;
645
646 err3:
647         ib_dealloc_pd(xrcd->pd);
648 err2:
649         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
650 err1:
651         kfree(xrcd);
652         return ERR_PTR(err);
653 }
654
655 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
656 {
657         ib_destroy_cq(to_mxrcd(xrcd)->cq);
658         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
659         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
660         kfree(xrcd);
661
662         return 0;
663 }
664
665 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
666 {
667         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
668         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
669         struct mlx4_ib_gid_entry *ge;
670
671         ge = kzalloc(sizeof *ge, GFP_KERNEL);
672         if (!ge)
673                 return -ENOMEM;
674
675         ge->gid = *gid;
676         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
677                 ge->port = mqp->port;
678                 ge->added = 1;
679         }
680
681         mutex_lock(&mqp->mutex);
682         list_add_tail(&ge->list, &mqp->gid_list);
683         mutex_unlock(&mqp->mutex);
684
685         return 0;
686 }
687
688 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
689                    union ib_gid *gid)
690 {
691         u8 mac[6];
692         struct net_device *ndev;
693         int ret = 0;
694
695         if (!mqp->port)
696                 return 0;
697
698         spin_lock(&mdev->iboe.lock);
699         ndev = mdev->iboe.netdevs[mqp->port - 1];
700         if (ndev)
701                 dev_hold(ndev);
702         spin_unlock(&mdev->iboe.lock);
703
704         if (ndev) {
705                 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
706                 rtnl_lock();
707                 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
708                 ret = 1;
709                 rtnl_unlock();
710                 dev_put(ndev);
711         }
712
713         return ret;
714 }
715
716 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
717 {
718         int err;
719         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
720         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
721
722         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
723                                     !!(mqp->flags & MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
724                                     MLX4_PROT_IB_IPV6);
725         if (err)
726                 return err;
727
728         err = add_gid_entry(ibqp, gid);
729         if (err)
730                 goto err_add;
731
732         return 0;
733
734 err_add:
735         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw, MLX4_PROT_IB_IPV6);
736         return err;
737 }
738
739 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
740 {
741         struct mlx4_ib_gid_entry *ge;
742         struct mlx4_ib_gid_entry *tmp;
743         struct mlx4_ib_gid_entry *ret = NULL;
744
745         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
746                 if (!memcmp(raw, ge->gid.raw, 16)) {
747                         ret = ge;
748                         break;
749                 }
750         }
751
752         return ret;
753 }
754
755 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
756 {
757         int err;
758         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
759         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
760         u8 mac[6];
761         struct net_device *ndev;
762         struct mlx4_ib_gid_entry *ge;
763
764         err = mlx4_multicast_detach(mdev->dev,
765                                     &mqp->mqp, gid->raw, MLX4_PROT_IB_IPV6);
766         if (err)
767                 return err;
768
769         mutex_lock(&mqp->mutex);
770         ge = find_gid_entry(mqp, gid->raw);
771         if (ge) {
772                 spin_lock(&mdev->iboe.lock);
773                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
774                 if (ndev)
775                         dev_hold(ndev);
776                 spin_unlock(&mdev->iboe.lock);
777                 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
778                 if (ndev) {
779                         rtnl_lock();
780                         dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
781                         rtnl_unlock();
782                         dev_put(ndev);
783                 }
784                 list_del(&ge->list);
785                 kfree(ge);
786         } else
787                 printk(KERN_WARNING "could not find mgid entry\n");
788
789         mutex_unlock(&mqp->mutex);
790
791         return 0;
792 }
793
794 static int init_node_data(struct mlx4_ib_dev *dev)
795 {
796         struct ib_smp *in_mad  = NULL;
797         struct ib_smp *out_mad = NULL;
798         int err = -ENOMEM;
799
800         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
801         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
802         if (!in_mad || !out_mad)
803                 goto out;
804
805         init_query_mad(in_mad);
806         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
807
808         err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
809         if (err)
810                 goto out;
811
812         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
813
814         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
815
816         err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
817         if (err)
818                 goto out;
819
820         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
821
822 out:
823         kfree(in_mad);
824         kfree(out_mad);
825         return err;
826 }
827
828 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
829                         char *buf)
830 {
831         struct mlx4_ib_dev *dev =
832                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
833         return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
834 }
835
836 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
837                            char *buf)
838 {
839         struct mlx4_ib_dev *dev =
840                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
841         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
842                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
843                        (int) dev->dev->caps.fw_ver & 0xffff);
844 }
845
846 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
847                         char *buf)
848 {
849         struct mlx4_ib_dev *dev =
850                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
851         return sprintf(buf, "%x\n", dev->dev->rev_id);
852 }
853
854 static ssize_t show_board(struct device *device, struct device_attribute *attr,
855                           char *buf)
856 {
857         struct mlx4_ib_dev *dev =
858                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
859         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
860                        dev->dev->board_id);
861 }
862
863 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
864 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
865 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
866 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
867
868 static struct device_attribute *mlx4_class_attributes[] = {
869         &dev_attr_hw_rev,
870         &dev_attr_fw_ver,
871         &dev_attr_hca_type,
872         &dev_attr_board_id
873 };
874
875 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
876 {
877         memcpy(eui, dev->dev_addr, 3);
878         memcpy(eui + 5, dev->dev_addr + 3, 3);
879         if (vlan_id < 0x1000) {
880                 eui[3] = vlan_id >> 8;
881                 eui[4] = vlan_id & 0xff;
882         } else {
883                 eui[3] = 0xff;
884                 eui[4] = 0xfe;
885         }
886         eui[0] ^= 2;
887 }
888
889 static void update_gids_task(struct work_struct *work)
890 {
891         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
892         struct mlx4_cmd_mailbox *mailbox;
893         union ib_gid *gids;
894         int err;
895         struct mlx4_dev *dev = gw->dev->dev;
896         struct ib_event event;
897
898         mailbox = mlx4_alloc_cmd_mailbox(dev);
899         if (IS_ERR(mailbox)) {
900                 printk(KERN_WARNING "update gid table failed %ld\n", PTR_ERR(mailbox));
901                 return;
902         }
903
904         gids = mailbox->buf;
905         memcpy(gids, gw->gids, sizeof gw->gids);
906
907         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
908                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
909                        MLX4_CMD_NATIVE);
910         if (err)
911                 printk(KERN_WARNING "set port command failed\n");
912         else {
913                 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
914                 event.device = &gw->dev->ib_dev;
915                 event.element.port_num = gw->port;
916                 event.event    = IB_EVENT_GID_CHANGE;
917                 ib_dispatch_event(&event);
918         }
919
920         mlx4_free_cmd_mailbox(dev, mailbox);
921         kfree(gw);
922 }
923
924 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
925 {
926         struct net_device *ndev = dev->iboe.netdevs[port - 1];
927         struct update_gid_work *work;
928         struct net_device *tmp;
929         int i;
930         u8 *hits;
931         int ret;
932         union ib_gid gid;
933         int free;
934         int found;
935         int need_update = 0;
936         u16 vid;
937
938         work = kzalloc(sizeof *work, GFP_ATOMIC);
939         if (!work)
940                 return -ENOMEM;
941
942         hits = kzalloc(128, GFP_ATOMIC);
943         if (!hits) {
944                 ret = -ENOMEM;
945                 goto out;
946         }
947
948         rcu_read_lock();
949         for_each_netdev_rcu(&init_net, tmp) {
950                 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
951                         gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
952                         vid = rdma_vlan_dev_vlan_id(tmp);
953                         mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
954                         found = 0;
955                         free = -1;
956                         for (i = 0; i < 128; ++i) {
957                                 if (free < 0 &&
958                                     !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
959                                         free = i;
960                                 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
961                                         hits[i] = 1;
962                                         found = 1;
963                                         break;
964                                 }
965                         }
966
967                         if (!found) {
968                                 if (tmp == ndev &&
969                                     (memcmp(&dev->iboe.gid_table[port - 1][0],
970                                             &gid, sizeof gid) ||
971                                      !memcmp(&dev->iboe.gid_table[port - 1][0],
972                                              &zgid, sizeof gid))) {
973                                         dev->iboe.gid_table[port - 1][0] = gid;
974                                         ++need_update;
975                                         hits[0] = 1;
976                                 } else if (free >= 0) {
977                                         dev->iboe.gid_table[port - 1][free] = gid;
978                                         hits[free] = 1;
979                                         ++need_update;
980                                 }
981                         }
982                 }
983         }
984         rcu_read_unlock();
985
986         for (i = 0; i < 128; ++i)
987                 if (!hits[i]) {
988                         if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
989                                 ++need_update;
990                         dev->iboe.gid_table[port - 1][i] = zgid;
991                 }
992
993         if (need_update) {
994                 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
995                 INIT_WORK(&work->work, update_gids_task);
996                 work->port = port;
997                 work->dev = dev;
998                 queue_work(wq, &work->work);
999         } else
1000                 kfree(work);
1001
1002         kfree(hits);
1003         return 0;
1004
1005 out:
1006         kfree(work);
1007         return ret;
1008 }
1009
1010 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1011 {
1012         switch (event) {
1013         case NETDEV_UP:
1014         case NETDEV_CHANGEADDR:
1015                 update_ipv6_gids(dev, port, 0);
1016                 break;
1017
1018         case NETDEV_DOWN:
1019                 update_ipv6_gids(dev, port, 1);
1020                 dev->iboe.netdevs[port - 1] = NULL;
1021         }
1022 }
1023
1024 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1025 {
1026         update_ipv6_gids(dev, port, 0);
1027 }
1028
1029 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1030 {
1031         update_ipv6_gids(dev, port, 1);
1032 }
1033
1034 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1035                                 void *ptr)
1036 {
1037         struct net_device *dev = ptr;
1038         struct mlx4_ib_dev *ibdev;
1039         struct net_device *oldnd;
1040         struct mlx4_ib_iboe *iboe;
1041         int port;
1042
1043         if (!net_eq(dev_net(dev), &init_net))
1044                 return NOTIFY_DONE;
1045
1046         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1047         iboe = &ibdev->iboe;
1048
1049         spin_lock(&iboe->lock);
1050         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1051                 oldnd = iboe->netdevs[port - 1];
1052                 iboe->netdevs[port - 1] =
1053                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1054                 if (oldnd != iboe->netdevs[port - 1]) {
1055                         if (iboe->netdevs[port - 1])
1056                                 netdev_added(ibdev, port);
1057                         else
1058                                 netdev_removed(ibdev, port);
1059                 }
1060         }
1061
1062         if (dev == iboe->netdevs[0] ||
1063             (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1064                 handle_en_event(ibdev, 1, event);
1065         else if (dev == iboe->netdevs[1]
1066                  || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1067                 handle_en_event(ibdev, 2, event);
1068
1069         spin_unlock(&iboe->lock);
1070
1071         return NOTIFY_DONE;
1072 }
1073
1074 static void *mlx4_ib_add(struct mlx4_dev *dev)
1075 {
1076         struct mlx4_ib_dev *ibdev;
1077         int num_ports = 0;
1078         int i;
1079         int err;
1080         struct mlx4_ib_iboe *iboe;
1081
1082         printk_once(KERN_INFO "%s", mlx4_ib_version);
1083
1084         if (mlx4_is_mfunc(dev)) {
1085                 printk(KERN_WARNING "IB not yet supported in SRIOV\n");
1086                 return NULL;
1087         }
1088
1089         mlx4_foreach_ib_transport_port(i, dev)
1090                 num_ports++;
1091
1092         /* No point in registering a device with no ports... */
1093         if (num_ports == 0)
1094                 return NULL;
1095
1096         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1097         if (!ibdev) {
1098                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1099                 return NULL;
1100         }
1101
1102         iboe = &ibdev->iboe;
1103
1104         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1105                 goto err_dealloc;
1106
1107         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1108                 goto err_pd;
1109
1110         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1111                                  PAGE_SIZE);
1112         if (!ibdev->uar_map)
1113                 goto err_uar;
1114         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1115
1116         ibdev->dev = dev;
1117
1118         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1119         ibdev->ib_dev.owner             = THIS_MODULE;
1120         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
1121         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
1122         ibdev->num_ports                = num_ports;
1123         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1124         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
1125         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
1126
1127         ibdev->ib_dev.uverbs_abi_ver    = MLX4_IB_UVERBS_ABI_VERSION;
1128         ibdev->ib_dev.uverbs_cmd_mask   =
1129                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
1130                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
1131                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
1132                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
1133                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
1134                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
1135                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
1136                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1137                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
1138                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
1139                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
1140                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
1141                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
1142                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
1143                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
1144                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
1145                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
1146                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
1147                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
1148                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
1149                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
1150                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
1151                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1152
1153         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
1154         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
1155         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
1156         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
1157         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
1158         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
1159         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
1160         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
1161         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
1162         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
1163         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
1164         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
1165         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
1166         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
1167         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
1168         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
1169         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
1170         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
1171         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
1172         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
1173         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
1174         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
1175         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
1176         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
1177         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
1178         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
1179         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
1180         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
1181         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
1182         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
1183         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
1184         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
1185         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
1186         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
1187         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
1188         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1189         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1190         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1191         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
1192         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
1193         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
1194
1195         ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
1196         ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
1197         ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
1198         ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
1199
1200         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1201                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1202                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1203                 ibdev->ib_dev.uverbs_cmd_mask |=
1204                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1205                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1206         }
1207
1208         spin_lock_init(&iboe->lock);
1209
1210         if (init_node_data(ibdev))
1211                 goto err_map;
1212
1213         for (i = 0; i < ibdev->num_ports; ++i) {
1214                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1215                                                 IB_LINK_LAYER_ETHERNET) {
1216                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1217                         if (err)
1218                                 ibdev->counters[i] = -1;
1219                 } else
1220                                 ibdev->counters[i] = -1;
1221         }
1222
1223         spin_lock_init(&ibdev->sm_lock);
1224         mutex_init(&ibdev->cap_mask_mutex);
1225
1226         if (ib_register_device(&ibdev->ib_dev, NULL))
1227                 goto err_counter;
1228
1229         if (mlx4_ib_mad_init(ibdev))
1230                 goto err_reg;
1231
1232         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1233                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1234                 err = register_netdevice_notifier(&iboe->nb);
1235                 if (err)
1236                         goto err_reg;
1237         }
1238
1239         for (i = 0; i < ARRAY_SIZE(mlx4_class_attributes); ++i) {
1240                 if (device_create_file(&ibdev->ib_dev.dev,
1241                                        mlx4_class_attributes[i]))
1242                         goto err_notif;
1243         }
1244
1245         ibdev->ib_active = true;
1246
1247         return ibdev;
1248
1249 err_notif:
1250         if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1251                 printk(KERN_WARNING "failure unregistering notifier\n");
1252         flush_workqueue(wq);
1253
1254 err_reg:
1255         ib_unregister_device(&ibdev->ib_dev);
1256
1257 err_counter:
1258         for (; i; --i)
1259                 if (ibdev->counters[i - 1] != -1)
1260                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1261
1262 err_map:
1263         iounmap(ibdev->uar_map);
1264
1265 err_uar:
1266         mlx4_uar_free(dev, &ibdev->priv_uar);
1267
1268 err_pd:
1269         mlx4_pd_free(dev, ibdev->priv_pdn);
1270
1271 err_dealloc:
1272         ib_dealloc_device(&ibdev->ib_dev);
1273
1274         return NULL;
1275 }
1276
1277 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1278 {
1279         struct mlx4_ib_dev *ibdev = ibdev_ptr;
1280         int p;
1281
1282         mlx4_ib_mad_cleanup(ibdev);
1283         ib_unregister_device(&ibdev->ib_dev);
1284         if (ibdev->iboe.nb.notifier_call) {
1285                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1286                         printk(KERN_WARNING "failure unregistering notifier\n");
1287                 ibdev->iboe.nb.notifier_call = NULL;
1288         }
1289         iounmap(ibdev->uar_map);
1290         for (p = 0; p < ibdev->num_ports; ++p)
1291                 if (ibdev->counters[p] != -1)
1292                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1293         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1294                 mlx4_CLOSE_PORT(dev, p);
1295
1296         mlx4_uar_free(dev, &ibdev->priv_uar);
1297         mlx4_pd_free(dev, ibdev->priv_pdn);
1298         ib_dealloc_device(&ibdev->ib_dev);
1299 }
1300
1301 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1302                           enum mlx4_dev_event event, int port)
1303 {
1304         struct ib_event ibev;
1305         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1306
1307         if (port > ibdev->num_ports)
1308                 return;
1309
1310         switch (event) {
1311         case MLX4_DEV_EVENT_PORT_UP:
1312                 ibev.event = IB_EVENT_PORT_ACTIVE;
1313                 break;
1314
1315         case MLX4_DEV_EVENT_PORT_DOWN:
1316                 ibev.event = IB_EVENT_PORT_ERR;
1317                 break;
1318
1319         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1320                 ibdev->ib_active = false;
1321                 ibev.event = IB_EVENT_DEVICE_FATAL;
1322                 break;
1323
1324         default:
1325                 return;
1326         }
1327
1328         ibev.device           = ibdev_ptr;
1329         ibev.element.port_num = port;
1330
1331         ib_dispatch_event(&ibev);
1332 }
1333
1334 static struct mlx4_interface mlx4_ib_interface = {
1335         .add            = mlx4_ib_add,
1336         .remove         = mlx4_ib_remove,
1337         .event          = mlx4_ib_event,
1338         .protocol       = MLX4_PROT_IB_IPV6
1339 };
1340
1341 static int __init mlx4_ib_init(void)
1342 {
1343         int err;
1344
1345         wq = create_singlethread_workqueue("mlx4_ib");
1346         if (!wq)
1347                 return -ENOMEM;
1348
1349         err = mlx4_register_interface(&mlx4_ib_interface);
1350         if (err) {
1351                 destroy_workqueue(wq);
1352                 return err;
1353         }
1354
1355         return 0;
1356 }
1357
1358 static void __exit mlx4_ib_cleanup(void)
1359 {
1360         mlx4_unregister_interface(&mlx4_ib_interface);
1361         destroy_workqueue(wq);
1362 }
1363
1364 module_init(mlx4_ib_init);
1365 module_exit(mlx4_ib_cleanup);