?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
IB/mlx4: Add support for XRC SRQs
[linux-2.6/btrfs-unstable.git] / drivers / infiniband / hw / mlx4 / main.c
blob42a538e5df368b4369146d5abfe8a703b4c52439
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
132 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
133 0xffffff;
134 props->vendor_part_id = be16_to_cpup((__be16 *) (out_mad->data + 30));
135 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
136 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
137
138 props->max_mr_size = ~0ull;
139 props->page_size_cap = dev->dev->caps.page_size_cap;
140 props->max_qp = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
141 props->max_qp_wr = dev->dev->caps.max_wqes;
142 props->max_sge = min(dev->dev->caps.max_sq_sg,
143 dev->dev->caps.max_rq_sg);
144 props->max_cq = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
145 props->max_cqe = dev->dev->caps.max_cqes;
146 props->max_mr = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
147 props->max_pd = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
148 props->max_qp_rd_atom = dev->dev->caps.max_qp_dest_rdma;
149 props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
150 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
151 props->max_srq = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
152 props->max_srq_wr = dev->dev->caps.max_srq_wqes - 1;
153 props->max_srq_sge = dev->dev->caps.max_srq_sge;
154 props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
155 props->local_ca_ack_delay = dev->dev->caps.local_ca_ack_delay;
156 props->atomic_cap = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
157 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
158 props->masked_atomic_cap = IB_ATOMIC_HCA;
159 props->max_pkeys = dev->dev->caps.pkey_table_len[1];
160 props->max_mcast_grp = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
161 props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
162 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
163 props->max_mcast_grp;
164 props->max_map_per_fmr = (1 << (32 - ilog2(dev->dev->caps.num_mpts))) - 1;
165
166 out:
167 kfree(in_mad);
168 kfree(out_mad);
169
170 return err;
171 }
172
173 static enum rdma_link_layer
174 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
175 {
176 struct mlx4_dev *dev = to_mdev(device)->dev;
177
178 return dev->caps.port_mask & (1 << (port_num - 1)) ?
179 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
180 }
181
182 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
183 struct ib_port_attr *props,
184 struct ib_smp *out_mad)
185 {
186 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
187 props->lmc = out_mad->data[34] & 0x7;
188 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
189 props->sm_sl = out_mad->data[36] & 0xf;
190 props->state = out_mad->data[32] & 0xf;
191 props->phys_state = out_mad->data[33] >> 4;
192 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
193 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
194 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
195 props->pkey_tbl_len = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
196 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
197 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
198 props->active_width = out_mad->data[31] & 0xf;
199 props->active_speed = out_mad->data[35] >> 4;
200 props->max_mtu = out_mad->data[41] & 0xf;
201 props->active_mtu = out_mad->data[36] >> 4;
202 props->subnet_timeout = out_mad->data[51] & 0x1f;
203 props->max_vl_num = out_mad->data[37] >> 4;
204 props->init_type_reply = out_mad->data[41] >> 4;
205
206 return 0;
207 }
208
209 static u8 state_to_phys_state(enum ib_port_state state)
210 {
211 return state == IB_PORT_ACTIVE ? 5 : 3;
212 }
213
214 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
215 struct ib_port_attr *props,
216 struct ib_smp *out_mad)
217 {
218 struct mlx4_ib_iboe *iboe = &to_mdev(ibdev)->iboe;
219 struct net_device *ndev;
220 enum ib_mtu tmp;
221
222 props->active_width = IB_WIDTH_1X;
223 props->active_speed = 4;
224 props->port_cap_flags = IB_PORT_CM_SUP;
225 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
226 props->max_msg_sz = to_mdev(ibdev)->dev->caps.max_msg_sz;
227 props->pkey_tbl_len = 1;
228 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
229 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
230 props->max_mtu = IB_MTU_2048;
231 props->subnet_timeout = 0;
232 props->max_vl_num = out_mad->data[37] >> 4;
233 props->init_type_reply = 0;
234 props->state = IB_PORT_DOWN;
235 props->phys_state = state_to_phys_state(props->state);
236 props->active_mtu = IB_MTU_256;
237 spin_lock(&iboe->lock);
238 ndev = iboe->netdevs[port - 1];
239 if (!ndev)
240 goto out;
241
242 tmp = iboe_get_mtu(ndev->mtu);
243 props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
244
245 props->state = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
246 IB_PORT_ACTIVE : IB_PORT_DOWN;
247 props->phys_state = state_to_phys_state(props->state);
248
249 out:
250 spin_unlock(&iboe->lock);
251 return 0;
252 }
253
254 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
255 struct ib_port_attr *props)
256 {
257 struct ib_smp *in_mad = NULL;
258 struct ib_smp *out_mad = NULL;
259 int err = -ENOMEM;
260
261 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
262 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
263 if (!in_mad || !out_mad)
264 goto out;
265
266 memset(props, 0, sizeof *props);
267
268 init_query_mad(in_mad);
269 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
270 in_mad->attr_mod = cpu_to_be32(port);
271
272 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
273 if (err)
274 goto out;
275
276 err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
277 ib_link_query_port(ibdev, port, props, out_mad) :
278 eth_link_query_port(ibdev, port, props, out_mad);
279
280 out:
281 kfree(in_mad);
282 kfree(out_mad);
283
284 return err;
285 }
286
287 static int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
288 union ib_gid *gid)
289 {
290 struct ib_smp *in_mad = NULL;
291 struct ib_smp *out_mad = NULL;
292 int err = -ENOMEM;
293
294 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
295 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
296 if (!in_mad || !out_mad)
297 goto out;
298
299 init_query_mad(in_mad);
300 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
301 in_mad->attr_mod = cpu_to_be32(port);
302
303 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
304 if (err)
305 goto out;
306
307 memcpy(gid->raw, out_mad->data + 8, 8);
308
309 init_query_mad(in_mad);
310 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
311 in_mad->attr_mod = cpu_to_be32(index / 8);
312
313 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
314 if (err)
315 goto out;
316
317 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
318
319 out:
320 kfree(in_mad);
321 kfree(out_mad);
322 return err;
323 }
324
325 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
326 union ib_gid *gid)
327 {
328 struct mlx4_ib_dev *dev = to_mdev(ibdev);
329
330 *gid = dev->iboe.gid_table[port - 1][index];
331
332 return 0;
333 }
334
335 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
336 union ib_gid *gid)
337 {
338 if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
339 return __mlx4_ib_query_gid(ibdev, port, index, gid);
340 else
341 return iboe_query_gid(ibdev, port, index, gid);
342 }
343
344 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
345 u16 *pkey)
346 {
347 struct ib_smp *in_mad = NULL;
348 struct ib_smp *out_mad = NULL;
349 int err = -ENOMEM;
350
351 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
352 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
353 if (!in_mad || !out_mad)
354 goto out;
355
356 init_query_mad(in_mad);
357 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
358 in_mad->attr_mod = cpu_to_be32(index / 32);
359
360 err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
361 if (err)
362 goto out;
363
364 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
365
366 out:
367 kfree(in_mad);
368 kfree(out_mad);
369 return err;
370 }
371
372 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
373 struct ib_device_modify *props)
374 {
375 struct mlx4_cmd_mailbox *mailbox;
376
377 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
378 return -EOPNOTSUPP;
379
380 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
381 return 0;
382
383 spin_lock(&to_mdev(ibdev)->sm_lock);
384 memcpy(ibdev->node_desc, props->node_desc, 64);
385 spin_unlock(&to_mdev(ibdev)->sm_lock);
386
387 /*
388 * If possible, pass node desc to FW, so it can generate
389 * a 144 trap. If cmd fails, just ignore.
390 */
391 mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
392 if (IS_ERR(mailbox))
393 return 0;
394
395 memset(mailbox->buf, 0, 256);
396 memcpy(mailbox->buf, props->node_desc, 64);
397 mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
398 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A);
399
400 mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
401
402 return 0;
403 }
404
405 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
406 u32 cap_mask)
407 {
408 struct mlx4_cmd_mailbox *mailbox;
409 int err;
410 u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
411
412 mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
413 if (IS_ERR(mailbox))
414 return PTR_ERR(mailbox);
415
416 memset(mailbox->buf, 0, 256);
417
418 if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
419 *(u8 *) mailbox->buf = !!reset_qkey_viols << 6;
420 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
421 } else {
422 ((u8 *) mailbox->buf)[3] = !!reset_qkey_viols;
423 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
424 }
425
426 err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
427 MLX4_CMD_TIME_CLASS_B);
428
429 mlx4_free_cmd_mailbox(dev->dev, mailbox);
430 return err;
431 }
432
433 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
434 struct ib_port_modify *props)
435 {
436 struct ib_port_attr attr;
437 u32 cap_mask;
438 int err;
439
440 mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
441
442 err = mlx4_ib_query_port(ibdev, port, &attr);
443 if (err)
444 goto out;
445
446 cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
447 ~props->clr_port_cap_mask;
448
449 err = mlx4_SET_PORT(to_mdev(ibdev), port,
450 !!(mask & IB_PORT_RESET_QKEY_CNTR),
451 cap_mask);
452
453 out:
454 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
455 return err;
456 }
457
458 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
459 struct ib_udata *udata)
460 {
461 struct mlx4_ib_dev *dev = to_mdev(ibdev);
462 struct mlx4_ib_ucontext *context;
463 struct mlx4_ib_alloc_ucontext_resp resp;
464 int err;
465
466 if (!dev->ib_active)
467 return ERR_PTR(-EAGAIN);
468
469 resp.qp_tab_size = dev->dev->caps.num_qps;
470 resp.bf_reg_size = dev->dev->caps.bf_reg_size;
471 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
472
473 context = kmalloc(sizeof *context, GFP_KERNEL);
474 if (!context)
475 return ERR_PTR(-ENOMEM);
476
477 err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
478 if (err) {
479 kfree(context);
480 return ERR_PTR(err);
481 }
482
483 INIT_LIST_HEAD(&context->db_page_list);
484 mutex_init(&context->db_page_mutex);
485
486 err = ib_copy_to_udata(udata, &resp, sizeof resp);
487 if (err) {
488 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
489 kfree(context);
490 return ERR_PTR(-EFAULT);
491 }
492
493 return &context->ibucontext;
494 }
495
496 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
497 {
498 struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
499
500 mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
501 kfree(context);
502
503 return 0;
504 }
505
506 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
507 {
508 struct mlx4_ib_dev *dev = to_mdev(context->device);
509
510 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
511 return -EINVAL;
512
513 if (vma->vm_pgoff == 0) {
514 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
515
516 if (io_remap_pfn_range(vma, vma->vm_start,
517 to_mucontext(context)->uar.pfn,
518 PAGE_SIZE, vma->vm_page_prot))
519 return -EAGAIN;
520 } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
521 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
522
523 if (io_remap_pfn_range(vma, vma->vm_start,
524 to_mucontext(context)->uar.pfn +
525 dev->dev->caps.num_uars,
526 PAGE_SIZE, vma->vm_page_prot))
527 return -EAGAIN;
528 } else
529 return -EINVAL;
530
531 return 0;
532 }
533
534 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
535 struct ib_ucontext *context,
536 struct ib_udata *udata)
537 {
538 struct mlx4_ib_pd *pd;
539 int err;
540
541 pd = kmalloc(sizeof *pd, GFP_KERNEL);
542 if (!pd)
543 return ERR_PTR(-ENOMEM);
544
545 err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
546 if (err) {
547 kfree(pd);
548 return ERR_PTR(err);
549 }
550
551 if (context)
552 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
553 mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
554 kfree(pd);
555 return ERR_PTR(-EFAULT);
556 }
557
558 return &pd->ibpd;
559 }
560
561 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
562 {
563 mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
564 kfree(pd);
565
566 return 0;
567 }
568
569 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
570 struct ib_ucontext *context,
571 struct ib_udata *udata)
572 {
573 struct mlx4_ib_xrcd *xrcd;
574 int err;
575
576 if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
577 return ERR_PTR(-ENOSYS);
578
579 xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
580 if (!xrcd)
581 return ERR_PTR(-ENOMEM);
582
583 err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
584 if (err)
585 goto err1;
586
587 xrcd->pd = ib_alloc_pd(ibdev);
588 if (IS_ERR(xrcd->pd)) {
589 err = PTR_ERR(xrcd->pd);
590 goto err2;
591 }
592
593 xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
594 if (IS_ERR(xrcd->cq)) {
595 err = PTR_ERR(xrcd->cq);
596 goto err3;
597 }
598
599 return &xrcd->ibxrcd;
600
601 err3:
602 ib_dealloc_pd(xrcd->pd);
603 err2:
604 mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
605 err1:
606 kfree(xrcd);
607 return ERR_PTR(err);
608 }
609
610 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
611 {
612 ib_destroy_cq(to_mxrcd(xrcd)->cq);
613 ib_dealloc_pd(to_mxrcd(xrcd)->pd);
614 mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
615 kfree(xrcd);
616
617 return 0;
618 }
619
620 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
621 {
622 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
623 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
624 struct mlx4_ib_gid_entry *ge;
625
626 ge = kzalloc(sizeof *ge, GFP_KERNEL);
627 if (!ge)
628 return -ENOMEM;
629
630 ge->gid = *gid;
631 if (mlx4_ib_add_mc(mdev, mqp, gid)) {
632 ge->port = mqp->port;
633 ge->added = 1;
634 }
635
636 mutex_lock(&mqp->mutex);
637 list_add_tail(&ge->list, &mqp->gid_list);
638 mutex_unlock(&mqp->mutex);
639
640 return 0;
641 }
642
643 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
644 union ib_gid *gid)
645 {
646 u8 mac[6];
647 struct net_device *ndev;
648 int ret = 0;
649
650 if (!mqp->port)
651 return 0;
652
653 spin_lock(&mdev->iboe.lock);
654 ndev = mdev->iboe.netdevs[mqp->port - 1];
655 if (ndev)
656 dev_hold(ndev);
657 spin_unlock(&mdev->iboe.lock);
658
659 if (ndev) {
660 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
661 rtnl_lock();
662 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
663 ret = 1;
664 rtnl_unlock();
665 dev_put(ndev);
666 }
667
668 return ret;
669 }
670
671 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
672 {
673 int err;
674 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
675 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
676
677 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
678 !!(mqp->flags & MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
679 MLX4_PROT_IB_IPV6);
680 if (err)
681 return err;
682
683 err = add_gid_entry(ibqp, gid);
684 if (err)
685 goto err_add;
686
687 return 0;
688
689 err_add:
690 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw, MLX4_PROT_IB_IPV6);
691 return err;
692 }
693
694 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
695 {
696 struct mlx4_ib_gid_entry *ge;
697 struct mlx4_ib_gid_entry *tmp;
698 struct mlx4_ib_gid_entry *ret = NULL;
699
700 list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
701 if (!memcmp(raw, ge->gid.raw, 16)) {
702 ret = ge;
703 break;
704 }
705 }
706
707 return ret;
708 }
709
710 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
711 {
712 int err;
713 struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
714 struct mlx4_ib_qp *mqp = to_mqp(ibqp);
715 u8 mac[6];
716 struct net_device *ndev;
717 struct mlx4_ib_gid_entry *ge;
718
719 err = mlx4_multicast_detach(mdev->dev,
720 &mqp->mqp, gid->raw, MLX4_PROT_IB_IPV6);
721 if (err)
722 return err;
723
724 mutex_lock(&mqp->mutex);
725 ge = find_gid_entry(mqp, gid->raw);
726 if (ge) {
727 spin_lock(&mdev->iboe.lock);
728 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
729 if (ndev)
730 dev_hold(ndev);
731 spin_unlock(&mdev->iboe.lock);
732 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
733 if (ndev) {
734 rtnl_lock();
735 dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
736 rtnl_unlock();
737 dev_put(ndev);
738 }
739 list_del(&ge->list);
740 kfree(ge);
741 } else
742 printk(KERN_WARNING "could not find mgid entry\n");
743
744 mutex_unlock(&mqp->mutex);
745
746 return 0;
747 }
748
749 static int init_node_data(struct mlx4_ib_dev *dev)
750 {
751 struct ib_smp *in_mad = NULL;
752 struct ib_smp *out_mad = NULL;
753 int err = -ENOMEM;
754
755 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
756 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
757 if (!in_mad || !out_mad)
758 goto out;
759
760 init_query_mad(in_mad);
761 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
762
763 err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
764 if (err)
765 goto out;
766
767 memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
768
769 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
770
771 err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
772 if (err)
773 goto out;
774
775 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
776
777 out:
778 kfree(in_mad);
779 kfree(out_mad);
780 return err;
781 }
782
783 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
784 char *buf)
785 {
786 struct mlx4_ib_dev *dev =
787 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
788 return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
789 }
790
791 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
792 char *buf)
793 {
794 struct mlx4_ib_dev *dev =
795 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
796 return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
797 (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
798 (int) dev->dev->caps.fw_ver & 0xffff);
799 }
800
801 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
802 char *buf)
803 {
804 struct mlx4_ib_dev *dev =
805 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
806 return sprintf(buf, "%x\n", dev->dev->rev_id);
807 }
808
809 static ssize_t show_board(struct device *device, struct device_attribute *attr,
810 char *buf)
811 {
812 struct mlx4_ib_dev *dev =
813 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
814 return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
815 dev->dev->board_id);
816 }
817
818 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
819 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
820 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
821 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
822
823 static struct device_attribute *mlx4_class_attributes[] = {
824 &dev_attr_hw_rev,
825 &dev_attr_fw_ver,
826 &dev_attr_hca_type,
827 &dev_attr_board_id
828 };
829
830 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
831 {
832 memcpy(eui, dev->dev_addr, 3);
833 memcpy(eui + 5, dev->dev_addr + 3, 3);
834 if (vlan_id < 0x1000) {
835 eui[3] = vlan_id >> 8;
836 eui[4] = vlan_id & 0xff;
837 } else {
838 eui[3] = 0xff;
839 eui[4] = 0xfe;
840 }
841 eui[0] ^= 2;
842 }
843
844 static void update_gids_task(struct work_struct *work)
845 {
846 struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
847 struct mlx4_cmd_mailbox *mailbox;
848 union ib_gid *gids;
849 int err;
850 struct mlx4_dev *dev = gw->dev->dev;
851 struct ib_event event;
852
853 mailbox = mlx4_alloc_cmd_mailbox(dev);
854 if (IS_ERR(mailbox)) {
855 printk(KERN_WARNING "update gid table failed %ld\n", PTR_ERR(mailbox));
856 return;
857 }
858
859 gids = mailbox->buf;
860 memcpy(gids, gw->gids, sizeof gw->gids);
861
862 err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
863 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B);
864 if (err)
865 printk(KERN_WARNING "set port command failed\n");
866 else {
867 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
868 event.device = &gw->dev->ib_dev;
869 event.element.port_num = gw->port;
870 event.event = IB_EVENT_GID_CHANGE;
871 ib_dispatch_event(&event);
872 }
873
874 mlx4_free_cmd_mailbox(dev, mailbox);
875 kfree(gw);
876 }
877
878 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
879 {
880 struct net_device *ndev = dev->iboe.netdevs[port - 1];
881 struct update_gid_work *work;
882 struct net_device *tmp;
883 int i;
884 u8 *hits;
885 int ret;
886 union ib_gid gid;
887 int free;
888 int found;
889 int need_update = 0;
890 u16 vid;
891
892 work = kzalloc(sizeof *work, GFP_ATOMIC);
893 if (!work)
894 return -ENOMEM;
895
896 hits = kzalloc(128, GFP_ATOMIC);
897 if (!hits) {
898 ret = -ENOMEM;
899 goto out;
900 }
901
902 rcu_read_lock();
903 for_each_netdev_rcu(&init_net, tmp) {
904 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
905 gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
906 vid = rdma_vlan_dev_vlan_id(tmp);
907 mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
908 found = 0;
909 free = -1;
910 for (i = 0; i < 128; ++i) {
911 if (free < 0 &&
912 !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
913 free = i;
914 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
915 hits[i] = 1;
916 found = 1;
917 break;
918 }
919 }
920
921 if (!found) {
922 if (tmp == ndev &&
923 (memcmp(&dev->iboe.gid_table[port - 1][0],
924 &gid, sizeof gid) ||
925 !memcmp(&dev->iboe.gid_table[port - 1][0],
926 &zgid, sizeof gid))) {
927 dev->iboe.gid_table[port - 1][0] = gid;
928 ++need_update;
929 hits[0] = 1;
930 } else if (free >= 0) {
931 dev->iboe.gid_table[port - 1][free] = gid;
932 hits[free] = 1;
933 ++need_update;
934 }
935 }
936 }
937 }
938 rcu_read_unlock();
939
940 for (i = 0; i < 128; ++i)
941 if (!hits[i]) {
942 if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
943 ++need_update;
944 dev->iboe.gid_table[port - 1][i] = zgid;
945 }
946
947 if (need_update) {
948 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
949 INIT_WORK(&work->work, update_gids_task);
950 work->port = port;
951 work->dev = dev;
952 queue_work(wq, &work->work);
953 } else
954 kfree(work);
955
956 kfree(hits);
957 return 0;
958
959 out:
960 kfree(work);
961 return ret;
962 }
963
964 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
965 {
966 switch (event) {
967 case NETDEV_UP:
968 case NETDEV_CHANGEADDR:
969 update_ipv6_gids(dev, port, 0);
970 break;
971
972 case NETDEV_DOWN:
973 update_ipv6_gids(dev, port, 1);
974 dev->iboe.netdevs[port - 1] = NULL;
975 }
976 }
977
978 static void netdev_added(struct mlx4_ib_dev *dev, int port)
979 {
980 update_ipv6_gids(dev, port, 0);
981 }
982
983 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
984 {
985 update_ipv6_gids(dev, port, 1);
986 }
987
988 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
989 void *ptr)
990 {
991 struct net_device *dev = ptr;
992 struct mlx4_ib_dev *ibdev;
993 struct net_device *oldnd;
994 struct mlx4_ib_iboe *iboe;
995 int port;
996
997 if (!net_eq(dev_net(dev), &init_net))
998 return NOTIFY_DONE;
999
1000 ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1001 iboe = &ibdev->iboe;
1002
1003 spin_lock(&iboe->lock);
1004 mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1005 oldnd = iboe->netdevs[port - 1];
1006 iboe->netdevs[port - 1] =
1007 mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1008 if (oldnd != iboe->netdevs[port - 1]) {
1009 if (iboe->netdevs[port - 1])
1010 netdev_added(ibdev, port);
1011 else
1012 netdev_removed(ibdev, port);
1013 }
1014 }
1015
1016 if (dev == iboe->netdevs[0] ||
1017 (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1018 handle_en_event(ibdev, 1, event);
1019 else if (dev == iboe->netdevs[1]
1020 || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1021 handle_en_event(ibdev, 2, event);
1022
1023 spin_unlock(&iboe->lock);
1024
1025 return NOTIFY_DONE;
1026 }
1027
1028 static void *mlx4_ib_add(struct mlx4_dev *dev)
1029 {
1030 struct mlx4_ib_dev *ibdev;
1031 int num_ports = 0;
1032 int i;
1033 int err;
1034 struct mlx4_ib_iboe *iboe;
1035
1036 printk_once(KERN_INFO "%s", mlx4_ib_version);
1037
1038 mlx4_foreach_ib_transport_port(i, dev)
1039 num_ports++;
1040
1041 /* No point in registering a device with no ports... */
1042 if (num_ports == 0)
1043 return NULL;
1044
1045 ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1046 if (!ibdev) {
1047 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1048 return NULL;
1049 }
1050
1051 iboe = &ibdev->iboe;
1052
1053 if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1054 goto err_dealloc;
1055
1056 if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1057 goto err_pd;
1058
1059 ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1060 PAGE_SIZE);
1061 if (!ibdev->uar_map)
1062 goto err_uar;
1063 MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1064
1065 ibdev->dev = dev;
1066
1067 strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1068 ibdev->ib_dev.owner = THIS_MODULE;
1069 ibdev->ib_dev.node_type = RDMA_NODE_IB_CA;
1070 ibdev->ib_dev.local_dma_lkey = dev->caps.reserved_lkey;
1071 ibdev->num_ports = num_ports;
1072 ibdev->ib_dev.phys_port_cnt = ibdev->num_ports;
1073 ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1074 ibdev->ib_dev.dma_device = &dev->pdev->dev;
1075
1076 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1077 ibdev->ib_dev.uverbs_cmd_mask =
1078 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1079 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1080 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1081 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1082 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1083 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1084 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1085 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1086 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1087 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1088 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1089 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1090 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1091 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1092 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1093 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1094 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1095 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1096 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1097 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1098 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1099 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ);
1100
1101 ibdev->ib_dev.query_device = mlx4_ib_query_device;
1102 ibdev->ib_dev.query_port = mlx4_ib_query_port;
1103 ibdev->ib_dev.get_link_layer = mlx4_ib_port_link_layer;
1104 ibdev->ib_dev.query_gid = mlx4_ib_query_gid;
1105 ibdev->ib_dev.query_pkey = mlx4_ib_query_pkey;
1106 ibdev->ib_dev.modify_device = mlx4_ib_modify_device;
1107 ibdev->ib_dev.modify_port = mlx4_ib_modify_port;
1108 ibdev->ib_dev.alloc_ucontext = mlx4_ib_alloc_ucontext;
1109 ibdev->ib_dev.dealloc_ucontext = mlx4_ib_dealloc_ucontext;
1110 ibdev->ib_dev.mmap = mlx4_ib_mmap;
1111 ibdev->ib_dev.alloc_pd = mlx4_ib_alloc_pd;
1112 ibdev->ib_dev.dealloc_pd = mlx4_ib_dealloc_pd;
1113 ibdev->ib_dev.create_ah = mlx4_ib_create_ah;
1114 ibdev->ib_dev.query_ah = mlx4_ib_query_ah;
1115 ibdev->ib_dev.destroy_ah = mlx4_ib_destroy_ah;
1116 ibdev->ib_dev.create_srq = mlx4_ib_create_srq;
1117 ibdev->ib_dev.modify_srq = mlx4_ib_modify_srq;
1118 ibdev->ib_dev.query_srq = mlx4_ib_query_srq;
1119 ibdev->ib_dev.destroy_srq = mlx4_ib_destroy_srq;
1120 ibdev->ib_dev.post_srq_recv = mlx4_ib_post_srq_recv;
1121 ibdev->ib_dev.create_qp = mlx4_ib_create_qp;
1122 ibdev->ib_dev.modify_qp = mlx4_ib_modify_qp;
1123 ibdev->ib_dev.query_qp = mlx4_ib_query_qp;
1124 ibdev->ib_dev.destroy_qp = mlx4_ib_destroy_qp;
1125 ibdev->ib_dev.post_send = mlx4_ib_post_send;
1126 ibdev->ib_dev.post_recv = mlx4_ib_post_recv;
1127 ibdev->ib_dev.create_cq = mlx4_ib_create_cq;
1128 ibdev->ib_dev.modify_cq = mlx4_ib_modify_cq;
1129 ibdev->ib_dev.resize_cq = mlx4_ib_resize_cq;
1130 ibdev->ib_dev.destroy_cq = mlx4_ib_destroy_cq;
1131 ibdev->ib_dev.poll_cq = mlx4_ib_poll_cq;
1132 ibdev->ib_dev.req_notify_cq = mlx4_ib_arm_cq;
1133 ibdev->ib_dev.get_dma_mr = mlx4_ib_get_dma_mr;
1134 ibdev->ib_dev.reg_user_mr = mlx4_ib_reg_user_mr;
1135 ibdev->ib_dev.dereg_mr = mlx4_ib_dereg_mr;
1136 ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1137 ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1138 ibdev->ib_dev.free_fast_reg_page_list = mlx4_ib_free_fast_reg_page_list;
1139 ibdev->ib_dev.attach_mcast = mlx4_ib_mcg_attach;
1140 ibdev->ib_dev.detach_mcast = mlx4_ib_mcg_detach;
1141 ibdev->ib_dev.process_mad = mlx4_ib_process_mad;
1142
1143 ibdev->ib_dev.alloc_fmr = mlx4_ib_fmr_alloc;
1144 ibdev->ib_dev.map_phys_fmr = mlx4_ib_map_phys_fmr;
1145 ibdev->ib_dev.unmap_fmr = mlx4_ib_unmap_fmr;
1146 ibdev->ib_dev.dealloc_fmr = mlx4_ib_fmr_dealloc;
1147
1148 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1149 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1150 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1151 ibdev->ib_dev.uverbs_cmd_mask |=
1152 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1153 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1154 }
1155
1156 spin_lock_init(&iboe->lock);
1157
1158 if (init_node_data(ibdev))
1159 goto err_map;
1160
1161 for (i = 0; i < ibdev->num_ports; ++i) {
1162 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1163 IB_LINK_LAYER_ETHERNET) {
1164 err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1165 if (err)
1166 ibdev->counters[i] = -1;
1167 } else
1168 ibdev->counters[i] = -1;
1169 }
1170
1171 spin_lock_init(&ibdev->sm_lock);
1172 mutex_init(&ibdev->cap_mask_mutex);
1173
1174 if (ib_register_device(&ibdev->ib_dev, NULL))
1175 goto err_counter;
1176
1177 if (mlx4_ib_mad_init(ibdev))
1178 goto err_reg;
1179
1180 if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1181 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1182 err = register_netdevice_notifier(&iboe->nb);
1183 if (err)
1184 goto err_reg;
1185 }
1186
1187 for (i = 0; i < ARRAY_SIZE(mlx4_class_attributes); ++i) {
1188 if (device_create_file(&ibdev->ib_dev.dev,
1189 mlx4_class_attributes[i]))
1190 goto err_notif;
1191 }
1192
1193 ibdev->ib_active = true;
1194
1195 return ibdev;
1196
1197 err_notif:
1198 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1199 printk(KERN_WARNING "failure unregistering notifier\n");
1200 flush_workqueue(wq);
1201
1202 err_reg:
1203 ib_unregister_device(&ibdev->ib_dev);
1204
1205 err_counter:
1206 for (; i; --i)
1207 mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1208
1209 err_map:
1210 iounmap(ibdev->uar_map);
1211
1212 err_uar:
1213 mlx4_uar_free(dev, &ibdev->priv_uar);
1214
1215 err_pd:
1216 mlx4_pd_free(dev, ibdev->priv_pdn);
1217
1218 err_dealloc:
1219 ib_dealloc_device(&ibdev->ib_dev);
1220
1221 return NULL;
1222 }
1223
1224 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1225 {
1226 struct mlx4_ib_dev *ibdev = ibdev_ptr;
1227 int p;
1228
1229 mlx4_ib_mad_cleanup(ibdev);
1230 ib_unregister_device(&ibdev->ib_dev);
1231 if (ibdev->iboe.nb.notifier_call) {
1232 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1233 printk(KERN_WARNING "failure unregistering notifier\n");
1234 ibdev->iboe.nb.notifier_call = NULL;
1235 }
1236 iounmap(ibdev->uar_map);
1237 for (p = 0; p < ibdev->num_ports; ++p)
1238 mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1239 mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1240 mlx4_CLOSE_PORT(dev, p);
1241
1242 mlx4_uar_free(dev, &ibdev->priv_uar);
1243 mlx4_pd_free(dev, ibdev->priv_pdn);
1244 ib_dealloc_device(&ibdev->ib_dev);
1245 }
1246
1247 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1248 enum mlx4_dev_event event, int port)
1249 {
1250 struct ib_event ibev;
1251 struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1252
1253 if (port > ibdev->num_ports)
1254 return;
1255
1256 switch (event) {
1257 case MLX4_DEV_EVENT_PORT_UP:
1258 ibev.event = IB_EVENT_PORT_ACTIVE;
1259 break;
1260
1261 case MLX4_DEV_EVENT_PORT_DOWN:
1262 ibev.event = IB_EVENT_PORT_ERR;
1263 break;
1264
1265 case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1266 ibdev->ib_active = false;
1267 ibev.event = IB_EVENT_DEVICE_FATAL;
1268 break;
1269
1270 default:
1271 return;
1272 }
1273
1274 ibev.device = ibdev_ptr;
1275 ibev.element.port_num = port;
1276
1277 ib_dispatch_event(&ibev);
1278 }
1279
1280 static struct mlx4_interface mlx4_ib_interface = {
1281 .add = mlx4_ib_add,
1282 .remove = mlx4_ib_remove,
1283 .event = mlx4_ib_event,
1284 .protocol = MLX4_PROT_IB_IPV6
1285 };
1286
1287 static int __init mlx4_ib_init(void)
1288 {
1289 int err;
1290
1291 wq = create_singlethread_workqueue("mlx4_ib");
1292 if (!wq)
1293 return -ENOMEM;
1294
1295 err = mlx4_register_interface(&mlx4_ib_interface);
1296 if (err) {
1297 destroy_workqueue(wq);
1298 return err;
1299 }
1300
1301 return 0;
1302 }
1303
1304 static void __exit mlx4_ib_cleanup(void)
1305 {
1306 mlx4_unregister_interface(&mlx4_ib_interface);
1307 destroy_workqueue(wq);
1308 }
1309
1310 module_init(mlx4_ib_init);
1311 module_exit(mlx4_ib_cleanup);
(deprecated) Btrfs devel tree