IB/mlx4: Add support for XRC QPs
[linux-2.6-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 = (1 << (32 - ilog2(dev->dev->caps.num_mpts))) - 1;
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 & (1 << (port_num - 1)) ?
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                               struct ib_smp *out_mad)
187 {
188         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
189         props->lmc              = out_mad->data[34] & 0x7;
190         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
191         props->sm_sl            = out_mad->data[36] & 0xf;
192         props->state            = out_mad->data[32] & 0xf;
193         props->phys_state       = out_mad->data[33] >> 4;
194         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
195         props->gid_tbl_len      = to_mdev(ibdev)->dev->caps.gid_table_len[port];
196         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
197         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
198         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
199         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
200         props->active_width     = out_mad->data[31] & 0xf;
201         props->active_speed     = out_mad->data[35] >> 4;
202         props->max_mtu          = out_mad->data[41] & 0xf;
203         props->active_mtu       = out_mad->data[36] >> 4;
204         props->subnet_timeout   = out_mad->data[51] & 0x1f;
205         props->max_vl_num       = out_mad->data[37] >> 4;
206         props->init_type_reply  = out_mad->data[41] >> 4;
207
208         return 0;
209 }
210
211 static u8 state_to_phys_state(enum ib_port_state state)
212 {
213         return state == IB_PORT_ACTIVE ? 5 : 3;
214 }
215
216 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
217                                struct ib_port_attr *props,
218                                struct ib_smp *out_mad)
219 {
220         struct mlx4_ib_iboe *iboe = &to_mdev(ibdev)->iboe;
221         struct net_device *ndev;
222         enum ib_mtu tmp;
223
224         props->active_width     = IB_WIDTH_1X;
225         props->active_speed     = 4;
226         props->port_cap_flags   = IB_PORT_CM_SUP;
227         props->gid_tbl_len      = to_mdev(ibdev)->dev->caps.gid_table_len[port];
228         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
229         props->pkey_tbl_len     = 1;
230         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
231         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
232         props->max_mtu          = IB_MTU_2048;
233         props->subnet_timeout   = 0;
234         props->max_vl_num       = out_mad->data[37] >> 4;
235         props->init_type_reply  = 0;
236         props->state            = IB_PORT_DOWN;
237         props->phys_state       = state_to_phys_state(props->state);
238         props->active_mtu       = IB_MTU_256;
239         spin_lock(&iboe->lock);
240         ndev = iboe->netdevs[port - 1];
241         if (!ndev)
242                 goto out;
243
244         tmp = iboe_get_mtu(ndev->mtu);
245         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
246
247         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
248                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
249         props->phys_state       = state_to_phys_state(props->state);
250
251 out:
252         spin_unlock(&iboe->lock);
253         return 0;
254 }
255
256 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
257                               struct ib_port_attr *props)
258 {
259         struct ib_smp *in_mad  = NULL;
260         struct ib_smp *out_mad = NULL;
261         int err = -ENOMEM;
262
263         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
264         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
265         if (!in_mad || !out_mad)
266                 goto out;
267
268         memset(props, 0, sizeof *props);
269
270         init_query_mad(in_mad);
271         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
272         in_mad->attr_mod = cpu_to_be32(port);
273
274         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
275         if (err)
276                 goto out;
277
278         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
279                 ib_link_query_port(ibdev, port, props, out_mad) :
280                 eth_link_query_port(ibdev, port, props, out_mad);
281
282 out:
283         kfree(in_mad);
284         kfree(out_mad);
285
286         return err;
287 }
288
289 static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
290                                union ib_gid *gid)
291 {
292         struct ib_smp *in_mad  = NULL;
293         struct ib_smp *out_mad = NULL;
294         int err = -ENOMEM;
295
296         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
297         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
298         if (!in_mad || !out_mad)
299                 goto out;
300
301         init_query_mad(in_mad);
302         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
303         in_mad->attr_mod = cpu_to_be32(port);
304
305         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
306         if (err)
307                 goto out;
308
309         memcpy(gid->raw, out_mad->data + 8, 8);
310
311         init_query_mad(in_mad);
312         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
313         in_mad->attr_mod = cpu_to_be32(index / 8);
314
315         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
316         if (err)
317                 goto out;
318
319         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
320
321 out:
322         kfree(in_mad);
323         kfree(out_mad);
324         return err;
325 }
326
327 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
328                           union ib_gid *gid)
329 {
330         struct mlx4_ib_dev *dev = to_mdev(ibdev);
331
332         *gid = dev->iboe.gid_table[port - 1][index];
333
334         return 0;
335 }
336
337 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
338                              union ib_gid *gid)
339 {
340         if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
341                 return __mlx4_ib_query_gid(ibdev, port, index, gid);
342         else
343                 return iboe_query_gid(ibdev, port, index, gid);
344 }
345
346 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
347                               u16 *pkey)
348 {
349         struct ib_smp *in_mad  = NULL;
350         struct ib_smp *out_mad = NULL;
351         int err = -ENOMEM;
352
353         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
354         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
355         if (!in_mad || !out_mad)
356                 goto out;
357
358         init_query_mad(in_mad);
359         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
360         in_mad->attr_mod = cpu_to_be32(index / 32);
361
362         err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
363         if (err)
364                 goto out;
365
366         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
367
368 out:
369         kfree(in_mad);
370         kfree(out_mad);
371         return err;
372 }
373
374 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
375                                  struct ib_device_modify *props)
376 {
377         struct mlx4_cmd_mailbox *mailbox;
378
379         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
380                 return -EOPNOTSUPP;
381
382         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
383                 return 0;
384
385         spin_lock(&to_mdev(ibdev)->sm_lock);
386         memcpy(ibdev->node_desc, props->node_desc, 64);
387         spin_unlock(&to_mdev(ibdev)->sm_lock);
388
389         /*
390          * If possible, pass node desc to FW, so it can generate
391          * a 144 trap.  If cmd fails, just ignore.
392          */
393         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
394         if (IS_ERR(mailbox))
395                 return 0;
396
397         memset(mailbox->buf, 0, 256);
398         memcpy(mailbox->buf, props->node_desc, 64);
399         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
400                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A);
401
402         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
403
404         return 0;
405 }
406
407 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
408                          u32 cap_mask)
409 {
410         struct mlx4_cmd_mailbox *mailbox;
411         int err;
412         u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
413
414         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
415         if (IS_ERR(mailbox))
416                 return PTR_ERR(mailbox);
417
418         memset(mailbox->buf, 0, 256);
419
420         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
421                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
422                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
423         } else {
424                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
425                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
426         }
427
428         err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
429                        MLX4_CMD_TIME_CLASS_B);
430
431         mlx4_free_cmd_mailbox(dev->dev, mailbox);
432         return err;
433 }
434
435 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
436                                struct ib_port_modify *props)
437 {
438         struct ib_port_attr attr;
439         u32 cap_mask;
440         int err;
441
442         mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
443
444         err = mlx4_ib_query_port(ibdev, port, &attr);
445         if (err)
446                 goto out;
447
448         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
449                 ~props->clr_port_cap_mask;
450
451         err = mlx4_SET_PORT(to_mdev(ibdev), port,
452                             !!(mask & IB_PORT_RESET_QKEY_CNTR),
453                             cap_mask);
454
455 out:
456         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
457         return err;
458 }
459
460 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
461                                                   struct ib_udata *udata)
462 {
463         struct mlx4_ib_dev *dev = to_mdev(ibdev);
464         struct mlx4_ib_ucontext *context;
465         struct mlx4_ib_alloc_ucontext_resp resp;
466         int err;
467
468         if (!dev->ib_active)
469                 return ERR_PTR(-EAGAIN);
470
471         resp.qp_tab_size      = dev->dev->caps.num_qps;
472         resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
473         resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
474
475         context = kmalloc(sizeof *context, GFP_KERNEL);
476         if (!context)
477                 return ERR_PTR(-ENOMEM);
478
479         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
480         if (err) {
481                 kfree(context);
482                 return ERR_PTR(err);
483         }
484
485         INIT_LIST_HEAD(&context->db_page_list);
486         mutex_init(&context->db_page_mutex);
487
488         err = ib_copy_to_udata(udata, &resp, sizeof resp);
489         if (err) {
490                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
491                 kfree(context);
492                 return ERR_PTR(-EFAULT);
493         }
494
495         return &context->ibucontext;
496 }
497
498 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
499 {
500         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
501
502         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
503         kfree(context);
504
505         return 0;
506 }
507
508 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
509 {
510         struct mlx4_ib_dev *dev = to_mdev(context->device);
511
512         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
513                 return -EINVAL;
514
515         if (vma->vm_pgoff == 0) {
516                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
517
518                 if (io_remap_pfn_range(vma, vma->vm_start,
519                                        to_mucontext(context)->uar.pfn,
520                                        PAGE_SIZE, vma->vm_page_prot))
521                         return -EAGAIN;
522         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
523                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
524
525                 if (io_remap_pfn_range(vma, vma->vm_start,
526                                        to_mucontext(context)->uar.pfn +
527                                        dev->dev->caps.num_uars,
528                                        PAGE_SIZE, vma->vm_page_prot))
529                         return -EAGAIN;
530         } else
531                 return -EINVAL;
532
533         return 0;
534 }
535
536 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
537                                       struct ib_ucontext *context,
538                                       struct ib_udata *udata)
539 {
540         struct mlx4_ib_pd *pd;
541         int err;
542
543         pd = kmalloc(sizeof *pd, GFP_KERNEL);
544         if (!pd)
545                 return ERR_PTR(-ENOMEM);
546
547         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
548         if (err) {
549                 kfree(pd);
550                 return ERR_PTR(err);
551         }
552
553         if (context)
554                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
555                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
556                         kfree(pd);
557                         return ERR_PTR(-EFAULT);
558                 }
559
560         return &pd->ibpd;
561 }
562
563 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
564 {
565         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
566         kfree(pd);
567
568         return 0;
569 }
570
571 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
572                                           struct ib_ucontext *context,
573                                           struct ib_udata *udata)
574 {
575         struct mlx4_ib_xrcd *xrcd;
576         int err;
577
578         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
579                 return ERR_PTR(-ENOSYS);
580
581         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
582         if (!xrcd)
583                 return ERR_PTR(-ENOMEM);
584
585         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
586         if (err)
587                 goto err1;
588
589         xrcd->pd = ib_alloc_pd(ibdev);
590         if (IS_ERR(xrcd->pd)) {
591                 err = PTR_ERR(xrcd->pd);
592                 goto err2;
593         }
594
595         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
596         if (IS_ERR(xrcd->cq)) {
597                 err = PTR_ERR(xrcd->cq);
598                 goto err3;
599         }
600
601         return &xrcd->ibxrcd;
602
603 err3:
604         ib_dealloc_pd(xrcd->pd);
605 err2:
606         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
607 err1:
608         kfree(xrcd);
609         return ERR_PTR(err);
610 }
611
612 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
613 {
614         ib_destroy_cq(to_mxrcd(xrcd)->cq);
615         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
616         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
617         kfree(xrcd);
618
619         return 0;
620 }
621
622 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
623 {
624         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
625         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
626         struct mlx4_ib_gid_entry *ge;
627
628         ge = kzalloc(sizeof *ge, GFP_KERNEL);
629         if (!ge)
630                 return -ENOMEM;
631
632         ge->gid = *gid;
633         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
634                 ge->port = mqp->port;
635                 ge->added = 1;
636         }
637
638         mutex_lock(&mqp->mutex);
639         list_add_tail(&ge->list, &mqp->gid_list);
640         mutex_unlock(&mqp->mutex);
641
642         return 0;
643 }
644
645 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
646                    union ib_gid *gid)
647 {
648         u8 mac[6];
649         struct net_device *ndev;
650         int ret = 0;
651
652         if (!mqp->port)
653                 return 0;
654
655         spin_lock(&mdev->iboe.lock);
656         ndev = mdev->iboe.netdevs[mqp->port - 1];
657         if (ndev)
658                 dev_hold(ndev);
659         spin_unlock(&mdev->iboe.lock);
660
661         if (ndev) {
662                 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
663                 rtnl_lock();
664                 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
665                 ret = 1;
666                 rtnl_unlock();
667                 dev_put(ndev);
668         }
669
670         return ret;
671 }
672
673 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
674 {
675         int err;
676         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
677         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
678
679         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
680                                     !!(mqp->flags & MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
681                                     MLX4_PROT_IB_IPV6);
682         if (err)
683                 return err;
684
685         err = add_gid_entry(ibqp, gid);
686         if (err)
687                 goto err_add;
688
689         return 0;
690
691 err_add:
692         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw, MLX4_PROT_IB_IPV6);
693         return err;
694 }
695
696 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
697 {
698         struct mlx4_ib_gid_entry *ge;
699         struct mlx4_ib_gid_entry *tmp;
700         struct mlx4_ib_gid_entry *ret = NULL;
701
702         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
703                 if (!memcmp(raw, ge->gid.raw, 16)) {
704                         ret = ge;
705                         break;
706                 }
707         }
708
709         return ret;
710 }
711
712 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
713 {
714         int err;
715         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
716         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
717         u8 mac[6];
718         struct net_device *ndev;
719         struct mlx4_ib_gid_entry *ge;
720
721         err = mlx4_multicast_detach(mdev->dev,
722                                     &mqp->mqp, gid->raw, MLX4_PROT_IB_IPV6);
723         if (err)
724                 return err;
725
726         mutex_lock(&mqp->mutex);
727         ge = find_gid_entry(mqp, gid->raw);
728         if (ge) {
729                 spin_lock(&mdev->iboe.lock);
730                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
731                 if (ndev)
732                         dev_hold(ndev);
733                 spin_unlock(&mdev->iboe.lock);
734                 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
735                 if (ndev) {
736                         rtnl_lock();
737                         dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
738                         rtnl_unlock();
739                         dev_put(ndev);
740                 }
741                 list_del(&ge->list);
742                 kfree(ge);
743         } else
744                 printk(KERN_WARNING "could not find mgid entry\n");
745
746         mutex_unlock(&mqp->mutex);
747
748         return 0;
749 }
750
751 static int init_node_data(struct mlx4_ib_dev *dev)
752 {
753         struct ib_smp *in_mad  = NULL;
754         struct ib_smp *out_mad = NULL;
755         int err = -ENOMEM;
756
757         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
758         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
759         if (!in_mad || !out_mad)
760                 goto out;
761
762         init_query_mad(in_mad);
763         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
764
765         err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
766         if (err)
767                 goto out;
768
769         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
770
771         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
772
773         err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
774         if (err)
775                 goto out;
776
777         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
778
779 out:
780         kfree(in_mad);
781         kfree(out_mad);
782         return err;
783 }
784
785 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
786                         char *buf)
787 {
788         struct mlx4_ib_dev *dev =
789                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
790         return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
791 }
792
793 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
794                            char *buf)
795 {
796         struct mlx4_ib_dev *dev =
797                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
798         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
799                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
800                        (int) dev->dev->caps.fw_ver & 0xffff);
801 }
802
803 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
804                         char *buf)
805 {
806         struct mlx4_ib_dev *dev =
807                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
808         return sprintf(buf, "%x\n", dev->dev->rev_id);
809 }
810
811 static ssize_t show_board(struct device *device, struct device_attribute *attr,
812                           char *buf)
813 {
814         struct mlx4_ib_dev *dev =
815                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
816         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
817                        dev->dev->board_id);
818 }
819
820 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
821 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
822 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
823 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
824
825 static struct device_attribute *mlx4_class_attributes[] = {
826         &dev_attr_hw_rev,
827         &dev_attr_fw_ver,
828         &dev_attr_hca_type,
829         &dev_attr_board_id
830 };
831
832 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
833 {
834         memcpy(eui, dev->dev_addr, 3);
835         memcpy(eui + 5, dev->dev_addr + 3, 3);
836         if (vlan_id < 0x1000) {
837                 eui[3] = vlan_id >> 8;
838                 eui[4] = vlan_id & 0xff;
839         } else {
840                 eui[3] = 0xff;
841                 eui[4] = 0xfe;
842         }
843         eui[0] ^= 2;
844 }
845
846 static void update_gids_task(struct work_struct *work)
847 {
848         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
849         struct mlx4_cmd_mailbox *mailbox;
850         union ib_gid *gids;
851         int err;
852         struct mlx4_dev *dev = gw->dev->dev;
853         struct ib_event event;
854
855         mailbox = mlx4_alloc_cmd_mailbox(dev);
856         if (IS_ERR(mailbox)) {
857                 printk(KERN_WARNING "update gid table failed %ld\n", PTR_ERR(mailbox));
858                 return;
859         }
860
861         gids = mailbox->buf;
862         memcpy(gids, gw->gids, sizeof gw->gids);
863
864         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
865                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B);
866         if (err)
867                 printk(KERN_WARNING "set port command failed\n");
868         else {
869                 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
870                 event.device = &gw->dev->ib_dev;
871                 event.element.port_num = gw->port;
872                 event.event    = IB_EVENT_GID_CHANGE;
873                 ib_dispatch_event(&event);
874         }
875
876         mlx4_free_cmd_mailbox(dev, mailbox);
877         kfree(gw);
878 }
879
880 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
881 {
882         struct net_device *ndev = dev->iboe.netdevs[port - 1];
883         struct update_gid_work *work;
884         struct net_device *tmp;
885         int i;
886         u8 *hits;
887         int ret;
888         union ib_gid gid;
889         int free;
890         int found;
891         int need_update = 0;
892         u16 vid;
893
894         work = kzalloc(sizeof *work, GFP_ATOMIC);
895         if (!work)
896                 return -ENOMEM;
897
898         hits = kzalloc(128, GFP_ATOMIC);
899         if (!hits) {
900                 ret = -ENOMEM;
901                 goto out;
902         }
903
904         rcu_read_lock();
905         for_each_netdev_rcu(&init_net, tmp) {
906                 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
907                         gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
908                         vid = rdma_vlan_dev_vlan_id(tmp);
909                         mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
910                         found = 0;
911                         free = -1;
912                         for (i = 0; i < 128; ++i) {
913                                 if (free < 0 &&
914                                     !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
915                                         free = i;
916                                 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
917                                         hits[i] = 1;
918                                         found = 1;
919                                         break;
920                                 }
921                         }
922
923                         if (!found) {
924                                 if (tmp == ndev &&
925                                     (memcmp(&dev->iboe.gid_table[port - 1][0],
926                                             &gid, sizeof gid) ||
927                                      !memcmp(&dev->iboe.gid_table[port - 1][0],
928                                              &zgid, sizeof gid))) {
929                                         dev->iboe.gid_table[port - 1][0] = gid;
930                                         ++need_update;
931                                         hits[0] = 1;
932                                 } else if (free >= 0) {
933                                         dev->iboe.gid_table[port - 1][free] = gid;
934                                         hits[free] = 1;
935                                         ++need_update;
936                                 }
937                         }
938                 }
939         }
940         rcu_read_unlock();
941
942         for (i = 0; i < 128; ++i)
943                 if (!hits[i]) {
944                         if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
945                                 ++need_update;
946                         dev->iboe.gid_table[port - 1][i] = zgid;
947                 }
948
949         if (need_update) {
950                 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
951                 INIT_WORK(&work->work, update_gids_task);
952                 work->port = port;
953                 work->dev = dev;
954                 queue_work(wq, &work->work);
955         } else
956                 kfree(work);
957
958         kfree(hits);
959         return 0;
960
961 out:
962         kfree(work);
963         return ret;
964 }
965
966 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
967 {
968         switch (event) {
969         case NETDEV_UP:
970         case NETDEV_CHANGEADDR:
971                 update_ipv6_gids(dev, port, 0);
972                 break;
973
974         case NETDEV_DOWN:
975                 update_ipv6_gids(dev, port, 1);
976                 dev->iboe.netdevs[port - 1] = NULL;
977         }
978 }
979
980 static void netdev_added(struct mlx4_ib_dev *dev, int port)
981 {
982         update_ipv6_gids(dev, port, 0);
983 }
984
985 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
986 {
987         update_ipv6_gids(dev, port, 1);
988 }
989
990 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
991                                 void *ptr)
992 {
993         struct net_device *dev = ptr;
994         struct mlx4_ib_dev *ibdev;
995         struct net_device *oldnd;
996         struct mlx4_ib_iboe *iboe;
997         int port;
998
999         if (!net_eq(dev_net(dev), &init_net))
1000                 return NOTIFY_DONE;
1001
1002         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1003         iboe = &ibdev->iboe;
1004
1005         spin_lock(&iboe->lock);
1006         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1007                 oldnd = iboe->netdevs[port - 1];
1008                 iboe->netdevs[port - 1] =
1009                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1010                 if (oldnd != iboe->netdevs[port - 1]) {
1011                         if (iboe->netdevs[port - 1])
1012                                 netdev_added(ibdev, port);
1013                         else
1014                                 netdev_removed(ibdev, port);
1015                 }
1016         }
1017
1018         if (dev == iboe->netdevs[0] ||
1019             (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1020                 handle_en_event(ibdev, 1, event);
1021         else if (dev == iboe->netdevs[1]
1022                  || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1023                 handle_en_event(ibdev, 2, event);
1024
1025         spin_unlock(&iboe->lock);
1026
1027         return NOTIFY_DONE;
1028 }
1029
1030 static void *mlx4_ib_add(struct mlx4_dev *dev)
1031 {
1032         struct mlx4_ib_dev *ibdev;
1033         int num_ports = 0;
1034         int i;
1035         int err;
1036         struct mlx4_ib_iboe *iboe;
1037
1038         printk_once(KERN_INFO "%s", mlx4_ib_version);
1039
1040         mlx4_foreach_ib_transport_port(i, dev)
1041                 num_ports++;
1042
1043         /* No point in registering a device with no ports... */
1044         if (num_ports == 0)
1045                 return NULL;
1046
1047         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1048         if (!ibdev) {
1049                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1050                 return NULL;
1051         }
1052
1053         iboe = &ibdev->iboe;
1054
1055         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1056                 goto err_dealloc;
1057
1058         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1059                 goto err_pd;
1060
1061         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1062                                  PAGE_SIZE);
1063         if (!ibdev->uar_map)
1064                 goto err_uar;
1065         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1066
1067         ibdev->dev = dev;
1068
1069         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1070         ibdev->ib_dev.owner             = THIS_MODULE;
1071         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
1072         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
1073         ibdev->num_ports                = num_ports;
1074         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1075         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
1076         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
1077
1078         ibdev->ib_dev.uverbs_abi_ver    = MLX4_IB_UVERBS_ABI_VERSION;
1079         ibdev->ib_dev.uverbs_cmd_mask   =
1080                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
1081                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
1082                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
1083                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
1084                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
1085                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
1086                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
1087                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1088                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
1089                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
1090                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
1091                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
1092                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
1093                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
1094                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
1095                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
1096                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
1097                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
1098                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
1099                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
1100                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
1101                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ);
1102
1103         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
1104         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
1105         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
1106         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
1107         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
1108         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
1109         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
1110         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
1111         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
1112         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
1113         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
1114         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
1115         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
1116         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
1117         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
1118         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
1119         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
1120         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
1121         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
1122         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
1123         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
1124         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
1125         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
1126         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
1127         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
1128         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
1129         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
1130         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
1131         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
1132         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
1133         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
1134         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
1135         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
1136         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
1137         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
1138         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1139         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1140         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1141         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
1142         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
1143         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
1144
1145         ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
1146         ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
1147         ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
1148         ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
1149
1150         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1151                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1152                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1153                 ibdev->ib_dev.uverbs_cmd_mask |=
1154                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1155                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1156         }
1157
1158         spin_lock_init(&iboe->lock);
1159
1160         if (init_node_data(ibdev))
1161                 goto err_map;
1162
1163         for (i = 0; i < ibdev->num_ports; ++i) {
1164                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1165                                                 IB_LINK_LAYER_ETHERNET) {
1166                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1167                         if (err)
1168                                 ibdev->counters[i] = -1;
1169                 } else
1170                                 ibdev->counters[i] = -1;
1171         }
1172
1173         spin_lock_init(&ibdev->sm_lock);
1174         mutex_init(&ibdev->cap_mask_mutex);
1175
1176         if (ib_register_device(&ibdev->ib_dev, NULL))
1177                 goto err_counter;
1178
1179         if (mlx4_ib_mad_init(ibdev))
1180                 goto err_reg;
1181
1182         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1183                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1184                 err = register_netdevice_notifier(&iboe->nb);
1185                 if (err)
1186                         goto err_reg;
1187         }
1188
1189         for (i = 0; i < ARRAY_SIZE(mlx4_class_attributes); ++i) {
1190                 if (device_create_file(&ibdev->ib_dev.dev,
1191                                        mlx4_class_attributes[i]))
1192                         goto err_notif;
1193         }
1194
1195         ibdev->ib_active = true;
1196
1197         return ibdev;
1198
1199 err_notif:
1200         if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1201                 printk(KERN_WARNING "failure unregistering notifier\n");
1202         flush_workqueue(wq);
1203
1204 err_reg:
1205         ib_unregister_device(&ibdev->ib_dev);
1206
1207 err_counter:
1208         for (; i; --i)
1209                 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1210
1211 err_map:
1212         iounmap(ibdev->uar_map);
1213
1214 err_uar:
1215         mlx4_uar_free(dev, &ibdev->priv_uar);
1216
1217 err_pd:
1218         mlx4_pd_free(dev, ibdev->priv_pdn);
1219
1220 err_dealloc:
1221         ib_dealloc_device(&ibdev->ib_dev);
1222
1223         return NULL;
1224 }
1225
1226 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1227 {
1228         struct mlx4_ib_dev *ibdev = ibdev_ptr;
1229         int p;
1230
1231         mlx4_ib_mad_cleanup(ibdev);
1232         ib_unregister_device(&ibdev->ib_dev);
1233         if (ibdev->iboe.nb.notifier_call) {
1234                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1235                         printk(KERN_WARNING "failure unregistering notifier\n");
1236                 ibdev->iboe.nb.notifier_call = NULL;
1237         }
1238         iounmap(ibdev->uar_map);
1239         for (p = 0; p < ibdev->num_ports; ++p)
1240                 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1241         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1242                 mlx4_CLOSE_PORT(dev, p);
1243
1244         mlx4_uar_free(dev, &ibdev->priv_uar);
1245         mlx4_pd_free(dev, ibdev->priv_pdn);
1246         ib_dealloc_device(&ibdev->ib_dev);
1247 }
1248
1249 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1250                           enum mlx4_dev_event event, int port)
1251 {
1252         struct ib_event ibev;
1253         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1254
1255         if (port > ibdev->num_ports)
1256                 return;
1257
1258         switch (event) {
1259         case MLX4_DEV_EVENT_PORT_UP:
1260                 ibev.event = IB_EVENT_PORT_ACTIVE;
1261                 break;
1262
1263         case MLX4_DEV_EVENT_PORT_DOWN:
1264                 ibev.event = IB_EVENT_PORT_ERR;
1265                 break;
1266
1267         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1268                 ibdev->ib_active = false;
1269                 ibev.event = IB_EVENT_DEVICE_FATAL;
1270                 break;
1271
1272         default:
1273                 return;
1274         }
1275
1276         ibev.device           = ibdev_ptr;
1277         ibev.element.port_num = port;
1278
1279         ib_dispatch_event(&ibev);
1280 }
1281
1282 static struct mlx4_interface mlx4_ib_interface = {
1283         .add            = mlx4_ib_add,
1284         .remove         = mlx4_ib_remove,
1285         .event          = mlx4_ib_event,
1286         .protocol       = MLX4_PROT_IB_IPV6
1287 };
1288
1289 static int __init mlx4_ib_init(void)
1290 {
1291         int err;
1292
1293         wq = create_singlethread_workqueue("mlx4_ib");
1294         if (!wq)
1295                 return -ENOMEM;
1296
1297         err = mlx4_register_interface(&mlx4_ib_interface);
1298         if (err) {
1299                 destroy_workqueue(wq);
1300                 return err;
1301         }
1302
1303         return 0;
1304 }
1305
1306 static void __exit mlx4_ib_cleanup(void)
1307 {
1308         mlx4_unregister_interface(&mlx4_ib_interface);
1309         destroy_workqueue(wq);
1310 }
1311
1312 module_init(mlx4_ib_init);
1313 module_exit(mlx4_ib_cleanup);