]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - drivers/infiniband/hw/mlx5/main.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branches 'for-4.4/upstream-fixes', 'for-4.5/async-suspend', 'for-4.5/container...
[mirror_ubuntu-artful-kernel.git] / drivers / infiniband / hw / mlx5 / main.c
1 /*
2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32
33 #include <linux/highmem.h>
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/errno.h>
37 #include <linux/pci.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/slab.h>
40 #include <linux/io-mapping.h>
41 #include <linux/sched.h>
42 #include <rdma/ib_user_verbs.h>
43 #include <linux/mlx5/vport.h>
44 #include <rdma/ib_smi.h>
45 #include <rdma/ib_umem.h>
46 #include "user.h"
47 #include "mlx5_ib.h"
48
49 #define DRIVER_NAME "mlx5_ib"
50 #define DRIVER_VERSION "2.2-1"
51 #define DRIVER_RELDATE "Feb 2014"
52
53 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
54 MODULE_DESCRIPTION("Mellanox Connect-IB HCA IB driver");
55 MODULE_LICENSE("Dual BSD/GPL");
56 MODULE_VERSION(DRIVER_VERSION);
57
58 static int deprecated_prof_sel = 2;
59 module_param_named(prof_sel, deprecated_prof_sel, int, 0444);
60 MODULE_PARM_DESC(prof_sel, "profile selector. Deprecated here. Moved to module mlx5_core");
61
62 static char mlx5_version[] =
63 DRIVER_NAME ": Mellanox Connect-IB Infiniband driver v"
64 DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
65
66 static enum rdma_link_layer
67 mlx5_ib_port_link_layer(struct ib_device *device)
68 {
69 struct mlx5_ib_dev *dev = to_mdev(device);
70
71 switch (MLX5_CAP_GEN(dev->mdev, port_type)) {
72 case MLX5_CAP_PORT_TYPE_IB:
73 return IB_LINK_LAYER_INFINIBAND;
74 case MLX5_CAP_PORT_TYPE_ETH:
75 return IB_LINK_LAYER_ETHERNET;
76 default:
77 return IB_LINK_LAYER_UNSPECIFIED;
78 }
79 }
80
81 static int mlx5_use_mad_ifc(struct mlx5_ib_dev *dev)
82 {
83 return !dev->mdev->issi;
84 }
85
86 enum {
87 MLX5_VPORT_ACCESS_METHOD_MAD,
88 MLX5_VPORT_ACCESS_METHOD_HCA,
89 MLX5_VPORT_ACCESS_METHOD_NIC,
90 };
91
92 static int mlx5_get_vport_access_method(struct ib_device *ibdev)
93 {
94 if (mlx5_use_mad_ifc(to_mdev(ibdev)))
95 return MLX5_VPORT_ACCESS_METHOD_MAD;
96
97 if (mlx5_ib_port_link_layer(ibdev) ==
98 IB_LINK_LAYER_ETHERNET)
99 return MLX5_VPORT_ACCESS_METHOD_NIC;
100
101 return MLX5_VPORT_ACCESS_METHOD_HCA;
102 }
103
104 static int mlx5_query_system_image_guid(struct ib_device *ibdev,
105 __be64 *sys_image_guid)
106 {
107 struct mlx5_ib_dev *dev = to_mdev(ibdev);
108 struct mlx5_core_dev *mdev = dev->mdev;
109 u64 tmp;
110 int err;
111
112 switch (mlx5_get_vport_access_method(ibdev)) {
113 case MLX5_VPORT_ACCESS_METHOD_MAD:
114 return mlx5_query_mad_ifc_system_image_guid(ibdev,
115 sys_image_guid);
116
117 case MLX5_VPORT_ACCESS_METHOD_HCA:
118 err = mlx5_query_hca_vport_system_image_guid(mdev, &tmp);
119 if (!err)
120 *sys_image_guid = cpu_to_be64(tmp);
121 return err;
122
123 default:
124 return -EINVAL;
125 }
126 }
127
128 static int mlx5_query_max_pkeys(struct ib_device *ibdev,
129 u16 *max_pkeys)
130 {
131 struct mlx5_ib_dev *dev = to_mdev(ibdev);
132 struct mlx5_core_dev *mdev = dev->mdev;
133
134 switch (mlx5_get_vport_access_method(ibdev)) {
135 case MLX5_VPORT_ACCESS_METHOD_MAD:
136 return mlx5_query_mad_ifc_max_pkeys(ibdev, max_pkeys);
137
138 case MLX5_VPORT_ACCESS_METHOD_HCA:
139 case MLX5_VPORT_ACCESS_METHOD_NIC:
140 *max_pkeys = mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(mdev,
141 pkey_table_size));
142 return 0;
143
144 default:
145 return -EINVAL;
146 }
147 }
148
149 static int mlx5_query_vendor_id(struct ib_device *ibdev,
150 u32 *vendor_id)
151 {
152 struct mlx5_ib_dev *dev = to_mdev(ibdev);
153
154 switch (mlx5_get_vport_access_method(ibdev)) {
155 case MLX5_VPORT_ACCESS_METHOD_MAD:
156 return mlx5_query_mad_ifc_vendor_id(ibdev, vendor_id);
157
158 case MLX5_VPORT_ACCESS_METHOD_HCA:
159 case MLX5_VPORT_ACCESS_METHOD_NIC:
160 return mlx5_core_query_vendor_id(dev->mdev, vendor_id);
161
162 default:
163 return -EINVAL;
164 }
165 }
166
167 static int mlx5_query_node_guid(struct mlx5_ib_dev *dev,
168 __be64 *node_guid)
169 {
170 u64 tmp;
171 int err;
172
173 switch (mlx5_get_vport_access_method(&dev->ib_dev)) {
174 case MLX5_VPORT_ACCESS_METHOD_MAD:
175 return mlx5_query_mad_ifc_node_guid(dev, node_guid);
176
177 case MLX5_VPORT_ACCESS_METHOD_HCA:
178 err = mlx5_query_hca_vport_node_guid(dev->mdev, &tmp);
179 if (!err)
180 *node_guid = cpu_to_be64(tmp);
181 return err;
182
183 default:
184 return -EINVAL;
185 }
186 }
187
188 struct mlx5_reg_node_desc {
189 u8 desc[64];
190 };
191
192 static int mlx5_query_node_desc(struct mlx5_ib_dev *dev, char *node_desc)
193 {
194 struct mlx5_reg_node_desc in;
195
196 if (mlx5_use_mad_ifc(dev))
197 return mlx5_query_mad_ifc_node_desc(dev, node_desc);
198
199 memset(&in, 0, sizeof(in));
200
201 return mlx5_core_access_reg(dev->mdev, &in, sizeof(in), node_desc,
202 sizeof(struct mlx5_reg_node_desc),
203 MLX5_REG_NODE_DESC, 0, 0);
204 }
205
206 static int mlx5_ib_query_device(struct ib_device *ibdev,
207 struct ib_device_attr *props,
208 struct ib_udata *uhw)
209 {
210 struct mlx5_ib_dev *dev = to_mdev(ibdev);
211 struct mlx5_core_dev *mdev = dev->mdev;
212 int err = -ENOMEM;
213 int max_rq_sg;
214 int max_sq_sg;
215 u64 min_page_size = 1ull << MLX5_CAP_GEN(mdev, log_pg_sz);
216
217 if (uhw->inlen || uhw->outlen)
218 return -EINVAL;
219
220 memset(props, 0, sizeof(*props));
221 err = mlx5_query_system_image_guid(ibdev,
222 &props->sys_image_guid);
223 if (err)
224 return err;
225
226 err = mlx5_query_max_pkeys(ibdev, &props->max_pkeys);
227 if (err)
228 return err;
229
230 err = mlx5_query_vendor_id(ibdev, &props->vendor_id);
231 if (err)
232 return err;
233
234 props->fw_ver = ((u64)fw_rev_maj(dev->mdev) << 32) |
235 (fw_rev_min(dev->mdev) << 16) |
236 fw_rev_sub(dev->mdev);
237 props->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
238 IB_DEVICE_PORT_ACTIVE_EVENT |
239 IB_DEVICE_SYS_IMAGE_GUID |
240 IB_DEVICE_RC_RNR_NAK_GEN;
241
242 if (MLX5_CAP_GEN(mdev, pkv))
243 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
244 if (MLX5_CAP_GEN(mdev, qkv))
245 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
246 if (MLX5_CAP_GEN(mdev, apm))
247 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
248 if (MLX5_CAP_GEN(mdev, xrc))
249 props->device_cap_flags |= IB_DEVICE_XRC;
250 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
251 if (MLX5_CAP_GEN(mdev, sho)) {
252 props->device_cap_flags |= IB_DEVICE_SIGNATURE_HANDOVER;
253 /* At this stage no support for signature handover */
254 props->sig_prot_cap = IB_PROT_T10DIF_TYPE_1 |
255 IB_PROT_T10DIF_TYPE_2 |
256 IB_PROT_T10DIF_TYPE_3;
257 props->sig_guard_cap = IB_GUARD_T10DIF_CRC |
258 IB_GUARD_T10DIF_CSUM;
259 }
260 if (MLX5_CAP_GEN(mdev, block_lb_mc))
261 props->device_cap_flags |= IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
262
263 props->vendor_part_id = mdev->pdev->device;
264 props->hw_ver = mdev->pdev->revision;
265
266 props->max_mr_size = ~0ull;
267 props->page_size_cap = ~(min_page_size - 1);
268 props->max_qp = 1 << MLX5_CAP_GEN(mdev, log_max_qp);
269 props->max_qp_wr = 1 << MLX5_CAP_GEN(mdev, log_max_qp_sz);
270 max_rq_sg = MLX5_CAP_GEN(mdev, max_wqe_sz_rq) /
271 sizeof(struct mlx5_wqe_data_seg);
272 max_sq_sg = (MLX5_CAP_GEN(mdev, max_wqe_sz_sq) -
273 sizeof(struct mlx5_wqe_ctrl_seg)) /
274 sizeof(struct mlx5_wqe_data_seg);
275 props->max_sge = min(max_rq_sg, max_sq_sg);
276 props->max_sge_rd = props->max_sge;
277 props->max_cq = 1 << MLX5_CAP_GEN(mdev, log_max_cq);
278 props->max_cqe = (1 << MLX5_CAP_GEN(mdev, log_max_eq_sz)) - 1;
279 props->max_mr = 1 << MLX5_CAP_GEN(mdev, log_max_mkey);
280 props->max_pd = 1 << MLX5_CAP_GEN(mdev, log_max_pd);
281 props->max_qp_rd_atom = 1 << MLX5_CAP_GEN(mdev, log_max_ra_req_qp);
282 props->max_qp_init_rd_atom = 1 << MLX5_CAP_GEN(mdev, log_max_ra_res_qp);
283 props->max_srq = 1 << MLX5_CAP_GEN(mdev, log_max_srq);
284 props->max_srq_wr = (1 << MLX5_CAP_GEN(mdev, log_max_srq_sz)) - 1;
285 props->local_ca_ack_delay = MLX5_CAP_GEN(mdev, local_ca_ack_delay);
286 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
287 props->max_srq_sge = max_rq_sg - 1;
288 props->max_fast_reg_page_list_len = (unsigned int)-1;
289 props->atomic_cap = IB_ATOMIC_NONE;
290 props->masked_atomic_cap = IB_ATOMIC_NONE;
291 props->max_mcast_grp = 1 << MLX5_CAP_GEN(mdev, log_max_mcg);
292 props->max_mcast_qp_attach = MLX5_CAP_GEN(mdev, max_qp_mcg);
293 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
294 props->max_mcast_grp;
295 props->max_map_per_fmr = INT_MAX; /* no limit in ConnectIB */
296
297 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
298 if (MLX5_CAP_GEN(mdev, pg))
299 props->device_cap_flags |= IB_DEVICE_ON_DEMAND_PAGING;
300 props->odp_caps = dev->odp_caps;
301 #endif
302
303 return 0;
304 }
305
306 enum mlx5_ib_width {
307 MLX5_IB_WIDTH_1X = 1 << 0,
308 MLX5_IB_WIDTH_2X = 1 << 1,
309 MLX5_IB_WIDTH_4X = 1 << 2,
310 MLX5_IB_WIDTH_8X = 1 << 3,
311 MLX5_IB_WIDTH_12X = 1 << 4
312 };
313
314 static int translate_active_width(struct ib_device *ibdev, u8 active_width,
315 u8 *ib_width)
316 {
317 struct mlx5_ib_dev *dev = to_mdev(ibdev);
318 int err = 0;
319
320 if (active_width & MLX5_IB_WIDTH_1X) {
321 *ib_width = IB_WIDTH_1X;
322 } else if (active_width & MLX5_IB_WIDTH_2X) {
323 mlx5_ib_dbg(dev, "active_width %d is not supported by IB spec\n",
324 (int)active_width);
325 err = -EINVAL;
326 } else if (active_width & MLX5_IB_WIDTH_4X) {
327 *ib_width = IB_WIDTH_4X;
328 } else if (active_width & MLX5_IB_WIDTH_8X) {
329 *ib_width = IB_WIDTH_8X;
330 } else if (active_width & MLX5_IB_WIDTH_12X) {
331 *ib_width = IB_WIDTH_12X;
332 } else {
333 mlx5_ib_dbg(dev, "Invalid active_width %d\n",
334 (int)active_width);
335 err = -EINVAL;
336 }
337
338 return err;
339 }
340
341 static int mlx5_mtu_to_ib_mtu(int mtu)
342 {
343 switch (mtu) {
344 case 256: return 1;
345 case 512: return 2;
346 case 1024: return 3;
347 case 2048: return 4;
348 case 4096: return 5;
349 default:
350 pr_warn("invalid mtu\n");
351 return -1;
352 }
353 }
354
355 enum ib_max_vl_num {
356 __IB_MAX_VL_0 = 1,
357 __IB_MAX_VL_0_1 = 2,
358 __IB_MAX_VL_0_3 = 3,
359 __IB_MAX_VL_0_7 = 4,
360 __IB_MAX_VL_0_14 = 5,
361 };
362
363 enum mlx5_vl_hw_cap {
364 MLX5_VL_HW_0 = 1,
365 MLX5_VL_HW_0_1 = 2,
366 MLX5_VL_HW_0_2 = 3,
367 MLX5_VL_HW_0_3 = 4,
368 MLX5_VL_HW_0_4 = 5,
369 MLX5_VL_HW_0_5 = 6,
370 MLX5_VL_HW_0_6 = 7,
371 MLX5_VL_HW_0_7 = 8,
372 MLX5_VL_HW_0_14 = 15
373 };
374
375 static int translate_max_vl_num(struct ib_device *ibdev, u8 vl_hw_cap,
376 u8 *max_vl_num)
377 {
378 switch (vl_hw_cap) {
379 case MLX5_VL_HW_0:
380 *max_vl_num = __IB_MAX_VL_0;
381 break;
382 case MLX5_VL_HW_0_1:
383 *max_vl_num = __IB_MAX_VL_0_1;
384 break;
385 case MLX5_VL_HW_0_3:
386 *max_vl_num = __IB_MAX_VL_0_3;
387 break;
388 case MLX5_VL_HW_0_7:
389 *max_vl_num = __IB_MAX_VL_0_7;
390 break;
391 case MLX5_VL_HW_0_14:
392 *max_vl_num = __IB_MAX_VL_0_14;
393 break;
394
395 default:
396 return -EINVAL;
397 }
398
399 return 0;
400 }
401
402 static int mlx5_query_hca_port(struct ib_device *ibdev, u8 port,
403 struct ib_port_attr *props)
404 {
405 struct mlx5_ib_dev *dev = to_mdev(ibdev);
406 struct mlx5_core_dev *mdev = dev->mdev;
407 struct mlx5_hca_vport_context *rep;
408 int max_mtu;
409 int oper_mtu;
410 int err;
411 u8 ib_link_width_oper;
412 u8 vl_hw_cap;
413
414 rep = kzalloc(sizeof(*rep), GFP_KERNEL);
415 if (!rep) {
416 err = -ENOMEM;
417 goto out;
418 }
419
420 memset(props, 0, sizeof(*props));
421
422 err = mlx5_query_hca_vport_context(mdev, 0, port, 0, rep);
423 if (err)
424 goto out;
425
426 props->lid = rep->lid;
427 props->lmc = rep->lmc;
428 props->sm_lid = rep->sm_lid;
429 props->sm_sl = rep->sm_sl;
430 props->state = rep->vport_state;
431 props->phys_state = rep->port_physical_state;
432 props->port_cap_flags = rep->cap_mask1;
433 props->gid_tbl_len = mlx5_get_gid_table_len(MLX5_CAP_GEN(mdev, gid_table_size));
434 props->max_msg_sz = 1 << MLX5_CAP_GEN(mdev, log_max_msg);
435 props->pkey_tbl_len = mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(mdev, pkey_table_size));
436 props->bad_pkey_cntr = rep->pkey_violation_counter;
437 props->qkey_viol_cntr = rep->qkey_violation_counter;
438 props->subnet_timeout = rep->subnet_timeout;
439 props->init_type_reply = rep->init_type_reply;
440
441 err = mlx5_query_port_link_width_oper(mdev, &ib_link_width_oper, port);
442 if (err)
443 goto out;
444
445 err = translate_active_width(ibdev, ib_link_width_oper,
446 &props->active_width);
447 if (err)
448 goto out;
449 err = mlx5_query_port_proto_oper(mdev, &props->active_speed, MLX5_PTYS_IB,
450 port);
451 if (err)
452 goto out;
453
454 mlx5_query_port_max_mtu(mdev, &max_mtu, port);
455
456 props->max_mtu = mlx5_mtu_to_ib_mtu(max_mtu);
457
458 mlx5_query_port_oper_mtu(mdev, &oper_mtu, port);
459
460 props->active_mtu = mlx5_mtu_to_ib_mtu(oper_mtu);
461
462 err = mlx5_query_port_vl_hw_cap(mdev, &vl_hw_cap, port);
463 if (err)
464 goto out;
465
466 err = translate_max_vl_num(ibdev, vl_hw_cap,
467 &props->max_vl_num);
468 out:
469 kfree(rep);
470 return err;
471 }
472
473 int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
474 struct ib_port_attr *props)
475 {
476 switch (mlx5_get_vport_access_method(ibdev)) {
477 case MLX5_VPORT_ACCESS_METHOD_MAD:
478 return mlx5_query_mad_ifc_port(ibdev, port, props);
479
480 case MLX5_VPORT_ACCESS_METHOD_HCA:
481 return mlx5_query_hca_port(ibdev, port, props);
482
483 default:
484 return -EINVAL;
485 }
486 }
487
488 static int mlx5_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
489 union ib_gid *gid)
490 {
491 struct mlx5_ib_dev *dev = to_mdev(ibdev);
492 struct mlx5_core_dev *mdev = dev->mdev;
493
494 switch (mlx5_get_vport_access_method(ibdev)) {
495 case MLX5_VPORT_ACCESS_METHOD_MAD:
496 return mlx5_query_mad_ifc_gids(ibdev, port, index, gid);
497
498 case MLX5_VPORT_ACCESS_METHOD_HCA:
499 return mlx5_query_hca_vport_gid(mdev, 0, port, 0, index, gid);
500
501 default:
502 return -EINVAL;
503 }
504
505 }
506
507 static int mlx5_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
508 u16 *pkey)
509 {
510 struct mlx5_ib_dev *dev = to_mdev(ibdev);
511 struct mlx5_core_dev *mdev = dev->mdev;
512
513 switch (mlx5_get_vport_access_method(ibdev)) {
514 case MLX5_VPORT_ACCESS_METHOD_MAD:
515 return mlx5_query_mad_ifc_pkey(ibdev, port, index, pkey);
516
517 case MLX5_VPORT_ACCESS_METHOD_HCA:
518 case MLX5_VPORT_ACCESS_METHOD_NIC:
519 return mlx5_query_hca_vport_pkey(mdev, 0, port, 0, index,
520 pkey);
521 default:
522 return -EINVAL;
523 }
524 }
525
526 static int mlx5_ib_modify_device(struct ib_device *ibdev, int mask,
527 struct ib_device_modify *props)
528 {
529 struct mlx5_ib_dev *dev = to_mdev(ibdev);
530 struct mlx5_reg_node_desc in;
531 struct mlx5_reg_node_desc out;
532 int err;
533
534 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
535 return -EOPNOTSUPP;
536
537 if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
538 return 0;
539
540 /*
541 * If possible, pass node desc to FW, so it can generate
542 * a 144 trap. If cmd fails, just ignore.
543 */
544 memcpy(&in, props->node_desc, 64);
545 err = mlx5_core_access_reg(dev->mdev, &in, sizeof(in), &out,
546 sizeof(out), MLX5_REG_NODE_DESC, 0, 1);
547 if (err)
548 return err;
549
550 memcpy(ibdev->node_desc, props->node_desc, 64);
551
552 return err;
553 }
554
555 static int mlx5_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
556 struct ib_port_modify *props)
557 {
558 struct mlx5_ib_dev *dev = to_mdev(ibdev);
559 struct ib_port_attr attr;
560 u32 tmp;
561 int err;
562
563 mutex_lock(&dev->cap_mask_mutex);
564
565 err = mlx5_ib_query_port(ibdev, port, &attr);
566 if (err)
567 goto out;
568
569 tmp = (attr.port_cap_flags | props->set_port_cap_mask) &
570 ~props->clr_port_cap_mask;
571
572 err = mlx5_set_port_caps(dev->mdev, port, tmp);
573
574 out:
575 mutex_unlock(&dev->cap_mask_mutex);
576 return err;
577 }
578
579 static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev,
580 struct ib_udata *udata)
581 {
582 struct mlx5_ib_dev *dev = to_mdev(ibdev);
583 struct mlx5_ib_alloc_ucontext_req_v2 req;
584 struct mlx5_ib_alloc_ucontext_resp resp;
585 struct mlx5_ib_ucontext *context;
586 struct mlx5_uuar_info *uuari;
587 struct mlx5_uar *uars;
588 int gross_uuars;
589 int num_uars;
590 int ver;
591 int uuarn;
592 int err;
593 int i;
594 size_t reqlen;
595
596 if (!dev->ib_active)
597 return ERR_PTR(-EAGAIN);
598
599 memset(&req, 0, sizeof(req));
600 reqlen = udata->inlen - sizeof(struct ib_uverbs_cmd_hdr);
601 if (reqlen == sizeof(struct mlx5_ib_alloc_ucontext_req))
602 ver = 0;
603 else if (reqlen == sizeof(struct mlx5_ib_alloc_ucontext_req_v2))
604 ver = 2;
605 else
606 return ERR_PTR(-EINVAL);
607
608 err = ib_copy_from_udata(&req, udata, reqlen);
609 if (err)
610 return ERR_PTR(err);
611
612 if (req.flags || req.reserved)
613 return ERR_PTR(-EINVAL);
614
615 if (req.total_num_uuars > MLX5_MAX_UUARS)
616 return ERR_PTR(-ENOMEM);
617
618 if (req.total_num_uuars == 0)
619 return ERR_PTR(-EINVAL);
620
621 req.total_num_uuars = ALIGN(req.total_num_uuars,
622 MLX5_NON_FP_BF_REGS_PER_PAGE);
623 if (req.num_low_latency_uuars > req.total_num_uuars - 1)
624 return ERR_PTR(-EINVAL);
625
626 num_uars = req.total_num_uuars / MLX5_NON_FP_BF_REGS_PER_PAGE;
627 gross_uuars = num_uars * MLX5_BF_REGS_PER_PAGE;
628 resp.qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp);
629 resp.bf_reg_size = 1 << MLX5_CAP_GEN(dev->mdev, log_bf_reg_size);
630 resp.cache_line_size = L1_CACHE_BYTES;
631 resp.max_sq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_sq);
632 resp.max_rq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_rq);
633 resp.max_send_wqebb = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp_sz);
634 resp.max_recv_wr = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp_sz);
635 resp.max_srq_recv_wr = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz);
636
637 context = kzalloc(sizeof(*context), GFP_KERNEL);
638 if (!context)
639 return ERR_PTR(-ENOMEM);
640
641 uuari = &context->uuari;
642 mutex_init(&uuari->lock);
643 uars = kcalloc(num_uars, sizeof(*uars), GFP_KERNEL);
644 if (!uars) {
645 err = -ENOMEM;
646 goto out_ctx;
647 }
648
649 uuari->bitmap = kcalloc(BITS_TO_LONGS(gross_uuars),
650 sizeof(*uuari->bitmap),
651 GFP_KERNEL);
652 if (!uuari->bitmap) {
653 err = -ENOMEM;
654 goto out_uar_ctx;
655 }
656 /*
657 * clear all fast path uuars
658 */
659 for (i = 0; i < gross_uuars; i++) {
660 uuarn = i & 3;
661 if (uuarn == 2 || uuarn == 3)
662 set_bit(i, uuari->bitmap);
663 }
664
665 uuari->count = kcalloc(gross_uuars, sizeof(*uuari->count), GFP_KERNEL);
666 if (!uuari->count) {
667 err = -ENOMEM;
668 goto out_bitmap;
669 }
670
671 for (i = 0; i < num_uars; i++) {
672 err = mlx5_cmd_alloc_uar(dev->mdev, &uars[i].index);
673 if (err)
674 goto out_count;
675 }
676
677 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
678 context->ibucontext.invalidate_range = &mlx5_ib_invalidate_range;
679 #endif
680
681 INIT_LIST_HEAD(&context->db_page_list);
682 mutex_init(&context->db_page_mutex);
683
684 resp.tot_uuars = req.total_num_uuars;
685 resp.num_ports = MLX5_CAP_GEN(dev->mdev, num_ports);
686 err = ib_copy_to_udata(udata, &resp,
687 sizeof(resp) - sizeof(resp.reserved));
688 if (err)
689 goto out_uars;
690
691 uuari->ver = ver;
692 uuari->num_low_latency_uuars = req.num_low_latency_uuars;
693 uuari->uars = uars;
694 uuari->num_uars = num_uars;
695 return &context->ibucontext;
696
697 out_uars:
698 for (i--; i >= 0; i--)
699 mlx5_cmd_free_uar(dev->mdev, uars[i].index);
700 out_count:
701 kfree(uuari->count);
702
703 out_bitmap:
704 kfree(uuari->bitmap);
705
706 out_uar_ctx:
707 kfree(uars);
708
709 out_ctx:
710 kfree(context);
711 return ERR_PTR(err);
712 }
713
714 static int mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
715 {
716 struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
717 struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
718 struct mlx5_uuar_info *uuari = &context->uuari;
719 int i;
720
721 for (i = 0; i < uuari->num_uars; i++) {
722 if (mlx5_cmd_free_uar(dev->mdev, uuari->uars[i].index))
723 mlx5_ib_warn(dev, "failed to free UAR 0x%x\n", uuari->uars[i].index);
724 }
725
726 kfree(uuari->count);
727 kfree(uuari->bitmap);
728 kfree(uuari->uars);
729 kfree(context);
730
731 return 0;
732 }
733
734 static phys_addr_t uar_index2pfn(struct mlx5_ib_dev *dev, int index)
735 {
736 return (pci_resource_start(dev->mdev->pdev, 0) >> PAGE_SHIFT) + index;
737 }
738
739 static int get_command(unsigned long offset)
740 {
741 return (offset >> MLX5_IB_MMAP_CMD_SHIFT) & MLX5_IB_MMAP_CMD_MASK;
742 }
743
744 static int get_arg(unsigned long offset)
745 {
746 return offset & ((1 << MLX5_IB_MMAP_CMD_SHIFT) - 1);
747 }
748
749 static int get_index(unsigned long offset)
750 {
751 return get_arg(offset);
752 }
753
754 static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
755 {
756 struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
757 struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
758 struct mlx5_uuar_info *uuari = &context->uuari;
759 unsigned long command;
760 unsigned long idx;
761 phys_addr_t pfn;
762
763 command = get_command(vma->vm_pgoff);
764 switch (command) {
765 case MLX5_IB_MMAP_REGULAR_PAGE:
766 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
767 return -EINVAL;
768
769 idx = get_index(vma->vm_pgoff);
770 if (idx >= uuari->num_uars)
771 return -EINVAL;
772
773 pfn = uar_index2pfn(dev, uuari->uars[idx].index);
774 mlx5_ib_dbg(dev, "uar idx 0x%lx, pfn 0x%llx\n", idx,
775 (unsigned long long)pfn);
776
777 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
778 if (io_remap_pfn_range(vma, vma->vm_start, pfn,
779 PAGE_SIZE, vma->vm_page_prot))
780 return -EAGAIN;
781
782 mlx5_ib_dbg(dev, "mapped WC at 0x%lx, PA 0x%llx\n",
783 vma->vm_start,
784 (unsigned long long)pfn << PAGE_SHIFT);
785 break;
786
787 case MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES:
788 return -ENOSYS;
789
790 default:
791 return -EINVAL;
792 }
793
794 return 0;
795 }
796
797 static struct ib_pd *mlx5_ib_alloc_pd(struct ib_device *ibdev,
798 struct ib_ucontext *context,
799 struct ib_udata *udata)
800 {
801 struct mlx5_ib_alloc_pd_resp resp;
802 struct mlx5_ib_pd *pd;
803 int err;
804
805 pd = kmalloc(sizeof(*pd), GFP_KERNEL);
806 if (!pd)
807 return ERR_PTR(-ENOMEM);
808
809 err = mlx5_core_alloc_pd(to_mdev(ibdev)->mdev, &pd->pdn);
810 if (err) {
811 kfree(pd);
812 return ERR_PTR(err);
813 }
814
815 if (context) {
816 resp.pdn = pd->pdn;
817 if (ib_copy_to_udata(udata, &resp, sizeof(resp))) {
818 mlx5_core_dealloc_pd(to_mdev(ibdev)->mdev, pd->pdn);
819 kfree(pd);
820 return ERR_PTR(-EFAULT);
821 }
822 }
823
824 return &pd->ibpd;
825 }
826
827 static int mlx5_ib_dealloc_pd(struct ib_pd *pd)
828 {
829 struct mlx5_ib_dev *mdev = to_mdev(pd->device);
830 struct mlx5_ib_pd *mpd = to_mpd(pd);
831
832 mlx5_core_dealloc_pd(mdev->mdev, mpd->pdn);
833 kfree(mpd);
834
835 return 0;
836 }
837
838 static int mlx5_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
839 {
840 struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
841 int err;
842
843 err = mlx5_core_attach_mcg(dev->mdev, gid, ibqp->qp_num);
844 if (err)
845 mlx5_ib_warn(dev, "failed attaching QPN 0x%x, MGID %pI6\n",
846 ibqp->qp_num, gid->raw);
847
848 return err;
849 }
850
851 static int mlx5_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
852 {
853 struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
854 int err;
855
856 err = mlx5_core_detach_mcg(dev->mdev, gid, ibqp->qp_num);
857 if (err)
858 mlx5_ib_warn(dev, "failed detaching QPN 0x%x, MGID %pI6\n",
859 ibqp->qp_num, gid->raw);
860
861 return err;
862 }
863
864 static int init_node_data(struct mlx5_ib_dev *dev)
865 {
866 int err;
867
868 err = mlx5_query_node_desc(dev, dev->ib_dev.node_desc);
869 if (err)
870 return err;
871
872 dev->mdev->rev_id = dev->mdev->pdev->revision;
873
874 return mlx5_query_node_guid(dev, &dev->ib_dev.node_guid);
875 }
876
877 static ssize_t show_fw_pages(struct device *device, struct device_attribute *attr,
878 char *buf)
879 {
880 struct mlx5_ib_dev *dev =
881 container_of(device, struct mlx5_ib_dev, ib_dev.dev);
882
883 return sprintf(buf, "%d\n", dev->mdev->priv.fw_pages);
884 }
885
886 static ssize_t show_reg_pages(struct device *device,
887 struct device_attribute *attr, char *buf)
888 {
889 struct mlx5_ib_dev *dev =
890 container_of(device, struct mlx5_ib_dev, ib_dev.dev);
891
892 return sprintf(buf, "%d\n", atomic_read(&dev->mdev->priv.reg_pages));
893 }
894
895 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
896 char *buf)
897 {
898 struct mlx5_ib_dev *dev =
899 container_of(device, struct mlx5_ib_dev, ib_dev.dev);
900 return sprintf(buf, "MT%d\n", dev->mdev->pdev->device);
901 }
902
903 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
904 char *buf)
905 {
906 struct mlx5_ib_dev *dev =
907 container_of(device, struct mlx5_ib_dev, ib_dev.dev);
908 return sprintf(buf, "%d.%d.%d\n", fw_rev_maj(dev->mdev),
909 fw_rev_min(dev->mdev), fw_rev_sub(dev->mdev));
910 }
911
912 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
913 char *buf)
914 {
915 struct mlx5_ib_dev *dev =
916 container_of(device, struct mlx5_ib_dev, ib_dev.dev);
917 return sprintf(buf, "%x\n", dev->mdev->rev_id);
918 }
919
920 static ssize_t show_board(struct device *device, struct device_attribute *attr,
921 char *buf)
922 {
923 struct mlx5_ib_dev *dev =
924 container_of(device, struct mlx5_ib_dev, ib_dev.dev);
925 return sprintf(buf, "%.*s\n", MLX5_BOARD_ID_LEN,
926 dev->mdev->board_id);
927 }
928
929 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
930 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
931 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
932 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
933 static DEVICE_ATTR(fw_pages, S_IRUGO, show_fw_pages, NULL);
934 static DEVICE_ATTR(reg_pages, S_IRUGO, show_reg_pages, NULL);
935
936 static struct device_attribute *mlx5_class_attributes[] = {
937 &dev_attr_hw_rev,
938 &dev_attr_fw_ver,
939 &dev_attr_hca_type,
940 &dev_attr_board_id,
941 &dev_attr_fw_pages,
942 &dev_attr_reg_pages,
943 };
944
945 static void mlx5_ib_event(struct mlx5_core_dev *dev, void *context,
946 enum mlx5_dev_event event, unsigned long param)
947 {
948 struct mlx5_ib_dev *ibdev = (struct mlx5_ib_dev *)context;
949 struct ib_event ibev;
950
951 u8 port = 0;
952
953 switch (event) {
954 case MLX5_DEV_EVENT_SYS_ERROR:
955 ibdev->ib_active = false;
956 ibev.event = IB_EVENT_DEVICE_FATAL;
957 break;
958
959 case MLX5_DEV_EVENT_PORT_UP:
960 ibev.event = IB_EVENT_PORT_ACTIVE;
961 port = (u8)param;
962 break;
963
964 case MLX5_DEV_EVENT_PORT_DOWN:
965 ibev.event = IB_EVENT_PORT_ERR;
966 port = (u8)param;
967 break;
968
969 case MLX5_DEV_EVENT_PORT_INITIALIZED:
970 /* not used by ULPs */
971 return;
972
973 case MLX5_DEV_EVENT_LID_CHANGE:
974 ibev.event = IB_EVENT_LID_CHANGE;
975 port = (u8)param;
976 break;
977
978 case MLX5_DEV_EVENT_PKEY_CHANGE:
979 ibev.event = IB_EVENT_PKEY_CHANGE;
980 port = (u8)param;
981 break;
982
983 case MLX5_DEV_EVENT_GUID_CHANGE:
984 ibev.event = IB_EVENT_GID_CHANGE;
985 port = (u8)param;
986 break;
987
988 case MLX5_DEV_EVENT_CLIENT_REREG:
989 ibev.event = IB_EVENT_CLIENT_REREGISTER;
990 port = (u8)param;
991 break;
992 }
993
994 ibev.device = &ibdev->ib_dev;
995 ibev.element.port_num = port;
996
997 if (port < 1 || port > ibdev->num_ports) {
998 mlx5_ib_warn(ibdev, "warning: event on port %d\n", port);
999 return;
1000 }
1001
1002 if (ibdev->ib_active)
1003 ib_dispatch_event(&ibev);
1004 }
1005
1006 static void get_ext_port_caps(struct mlx5_ib_dev *dev)
1007 {
1008 int port;
1009
1010 for (port = 1; port <= MLX5_CAP_GEN(dev->mdev, num_ports); port++)
1011 mlx5_query_ext_port_caps(dev, port);
1012 }
1013
1014 static int get_port_caps(struct mlx5_ib_dev *dev)
1015 {
1016 struct ib_device_attr *dprops = NULL;
1017 struct ib_port_attr *pprops = NULL;
1018 int err = -ENOMEM;
1019 int port;
1020 struct ib_udata uhw = {.inlen = 0, .outlen = 0};
1021
1022 pprops = kmalloc(sizeof(*pprops), GFP_KERNEL);
1023 if (!pprops)
1024 goto out;
1025
1026 dprops = kmalloc(sizeof(*dprops), GFP_KERNEL);
1027 if (!dprops)
1028 goto out;
1029
1030 err = mlx5_ib_query_device(&dev->ib_dev, dprops, &uhw);
1031 if (err) {
1032 mlx5_ib_warn(dev, "query_device failed %d\n", err);
1033 goto out;
1034 }
1035
1036 for (port = 1; port <= MLX5_CAP_GEN(dev->mdev, num_ports); port++) {
1037 err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
1038 if (err) {
1039 mlx5_ib_warn(dev, "query_port %d failed %d\n",
1040 port, err);
1041 break;
1042 }
1043 dev->mdev->port_caps[port - 1].pkey_table_len =
1044 dprops->max_pkeys;
1045 dev->mdev->port_caps[port - 1].gid_table_len =
1046 pprops->gid_tbl_len;
1047 mlx5_ib_dbg(dev, "pkey_table_len %d, gid_table_len %d\n",
1048 dprops->max_pkeys, pprops->gid_tbl_len);
1049 }
1050
1051 out:
1052 kfree(pprops);
1053 kfree(dprops);
1054
1055 return err;
1056 }
1057
1058 static void destroy_umrc_res(struct mlx5_ib_dev *dev)
1059 {
1060 int err;
1061
1062 err = mlx5_mr_cache_cleanup(dev);
1063 if (err)
1064 mlx5_ib_warn(dev, "mr cache cleanup failed\n");
1065
1066 mlx5_ib_destroy_qp(dev->umrc.qp);
1067 ib_destroy_cq(dev->umrc.cq);
1068 ib_dealloc_pd(dev->umrc.pd);
1069 }
1070
1071 enum {
1072 MAX_UMR_WR = 128,
1073 };
1074
1075 static int create_umr_res(struct mlx5_ib_dev *dev)
1076 {
1077 struct ib_qp_init_attr *init_attr = NULL;
1078 struct ib_qp_attr *attr = NULL;
1079 struct ib_pd *pd;
1080 struct ib_cq *cq;
1081 struct ib_qp *qp;
1082 struct ib_cq_init_attr cq_attr = {};
1083 int ret;
1084
1085 attr = kzalloc(sizeof(*attr), GFP_KERNEL);
1086 init_attr = kzalloc(sizeof(*init_attr), GFP_KERNEL);
1087 if (!attr || !init_attr) {
1088 ret = -ENOMEM;
1089 goto error_0;
1090 }
1091
1092 pd = ib_alloc_pd(&dev->ib_dev);
1093 if (IS_ERR(pd)) {
1094 mlx5_ib_dbg(dev, "Couldn't create PD for sync UMR QP\n");
1095 ret = PTR_ERR(pd);
1096 goto error_0;
1097 }
1098
1099 cq_attr.cqe = 128;
1100 cq = ib_create_cq(&dev->ib_dev, mlx5_umr_cq_handler, NULL, NULL,
1101 &cq_attr);
1102 if (IS_ERR(cq)) {
1103 mlx5_ib_dbg(dev, "Couldn't create CQ for sync UMR QP\n");
1104 ret = PTR_ERR(cq);
1105 goto error_2;
1106 }
1107 ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
1108
1109 init_attr->send_cq = cq;
1110 init_attr->recv_cq = cq;
1111 init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
1112 init_attr->cap.max_send_wr = MAX_UMR_WR;
1113 init_attr->cap.max_send_sge = 1;
1114 init_attr->qp_type = MLX5_IB_QPT_REG_UMR;
1115 init_attr->port_num = 1;
1116 qp = mlx5_ib_create_qp(pd, init_attr, NULL);
1117 if (IS_ERR(qp)) {
1118 mlx5_ib_dbg(dev, "Couldn't create sync UMR QP\n");
1119 ret = PTR_ERR(qp);
1120 goto error_3;
1121 }
1122 qp->device = &dev->ib_dev;
1123 qp->real_qp = qp;
1124 qp->uobject = NULL;
1125 qp->qp_type = MLX5_IB_QPT_REG_UMR;
1126
1127 attr->qp_state = IB_QPS_INIT;
1128 attr->port_num = 1;
1129 ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE | IB_QP_PKEY_INDEX |
1130 IB_QP_PORT, NULL);
1131 if (ret) {
1132 mlx5_ib_dbg(dev, "Couldn't modify UMR QP\n");
1133 goto error_4;
1134 }
1135
1136 memset(attr, 0, sizeof(*attr));
1137 attr->qp_state = IB_QPS_RTR;
1138 attr->path_mtu = IB_MTU_256;
1139
1140 ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
1141 if (ret) {
1142 mlx5_ib_dbg(dev, "Couldn't modify umr QP to rtr\n");
1143 goto error_4;
1144 }
1145
1146 memset(attr, 0, sizeof(*attr));
1147 attr->qp_state = IB_QPS_RTS;
1148 ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
1149 if (ret) {
1150 mlx5_ib_dbg(dev, "Couldn't modify umr QP to rts\n");
1151 goto error_4;
1152 }
1153
1154 dev->umrc.qp = qp;
1155 dev->umrc.cq = cq;
1156 dev->umrc.pd = pd;
1157
1158 sema_init(&dev->umrc.sem, MAX_UMR_WR);
1159 ret = mlx5_mr_cache_init(dev);
1160 if (ret) {
1161 mlx5_ib_warn(dev, "mr cache init failed %d\n", ret);
1162 goto error_4;
1163 }
1164
1165 kfree(attr);
1166 kfree(init_attr);
1167
1168 return 0;
1169
1170 error_4:
1171 mlx5_ib_destroy_qp(qp);
1172
1173 error_3:
1174 ib_destroy_cq(cq);
1175
1176 error_2:
1177 ib_dealloc_pd(pd);
1178
1179 error_0:
1180 kfree(attr);
1181 kfree(init_attr);
1182 return ret;
1183 }
1184
1185 static int create_dev_resources(struct mlx5_ib_resources *devr)
1186 {
1187 struct ib_srq_init_attr attr;
1188 struct mlx5_ib_dev *dev;
1189 struct ib_cq_init_attr cq_attr = {.cqe = 1};
1190 int ret = 0;
1191
1192 dev = container_of(devr, struct mlx5_ib_dev, devr);
1193
1194 devr->p0 = mlx5_ib_alloc_pd(&dev->ib_dev, NULL, NULL);
1195 if (IS_ERR(devr->p0)) {
1196 ret = PTR_ERR(devr->p0);
1197 goto error0;
1198 }
1199 devr->p0->device = &dev->ib_dev;
1200 devr->p0->uobject = NULL;
1201 atomic_set(&devr->p0->usecnt, 0);
1202
1203 devr->c0 = mlx5_ib_create_cq(&dev->ib_dev, &cq_attr, NULL, NULL);
1204 if (IS_ERR(devr->c0)) {
1205 ret = PTR_ERR(devr->c0);
1206 goto error1;
1207 }
1208 devr->c0->device = &dev->ib_dev;
1209 devr->c0->uobject = NULL;
1210 devr->c0->comp_handler = NULL;
1211 devr->c0->event_handler = NULL;
1212 devr->c0->cq_context = NULL;
1213 atomic_set(&devr->c0->usecnt, 0);
1214
1215 devr->x0 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
1216 if (IS_ERR(devr->x0)) {
1217 ret = PTR_ERR(devr->x0);
1218 goto error2;
1219 }
1220 devr->x0->device = &dev->ib_dev;
1221 devr->x0->inode = NULL;
1222 atomic_set(&devr->x0->usecnt, 0);
1223 mutex_init(&devr->x0->tgt_qp_mutex);
1224 INIT_LIST_HEAD(&devr->x0->tgt_qp_list);
1225
1226 devr->x1 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
1227 if (IS_ERR(devr->x1)) {
1228 ret = PTR_ERR(devr->x1);
1229 goto error3;
1230 }
1231 devr->x1->device = &dev->ib_dev;
1232 devr->x1->inode = NULL;
1233 atomic_set(&devr->x1->usecnt, 0);
1234 mutex_init(&devr->x1->tgt_qp_mutex);
1235 INIT_LIST_HEAD(&devr->x1->tgt_qp_list);
1236
1237 memset(&attr, 0, sizeof(attr));
1238 attr.attr.max_sge = 1;
1239 attr.attr.max_wr = 1;
1240 attr.srq_type = IB_SRQT_XRC;
1241 attr.ext.xrc.cq = devr->c0;
1242 attr.ext.xrc.xrcd = devr->x0;
1243
1244 devr->s0 = mlx5_ib_create_srq(devr->p0, &attr, NULL);
1245 if (IS_ERR(devr->s0)) {
1246 ret = PTR_ERR(devr->s0);
1247 goto error4;
1248 }
1249 devr->s0->device = &dev->ib_dev;
1250 devr->s0->pd = devr->p0;
1251 devr->s0->uobject = NULL;
1252 devr->s0->event_handler = NULL;
1253 devr->s0->srq_context = NULL;
1254 devr->s0->srq_type = IB_SRQT_XRC;
1255 devr->s0->ext.xrc.xrcd = devr->x0;
1256 devr->s0->ext.xrc.cq = devr->c0;
1257 atomic_inc(&devr->s0->ext.xrc.xrcd->usecnt);
1258 atomic_inc(&devr->s0->ext.xrc.cq->usecnt);
1259 atomic_inc(&devr->p0->usecnt);
1260 atomic_set(&devr->s0->usecnt, 0);
1261
1262 memset(&attr, 0, sizeof(attr));
1263 attr.attr.max_sge = 1;
1264 attr.attr.max_wr = 1;
1265 attr.srq_type = IB_SRQT_BASIC;
1266 devr->s1 = mlx5_ib_create_srq(devr->p0, &attr, NULL);
1267 if (IS_ERR(devr->s1)) {
1268 ret = PTR_ERR(devr->s1);
1269 goto error5;
1270 }
1271 devr->s1->device = &dev->ib_dev;
1272 devr->s1->pd = devr->p0;
1273 devr->s1->uobject = NULL;
1274 devr->s1->event_handler = NULL;
1275 devr->s1->srq_context = NULL;
1276 devr->s1->srq_type = IB_SRQT_BASIC;
1277 devr->s1->ext.xrc.cq = devr->c0;
1278 atomic_inc(&devr->p0->usecnt);
1279 atomic_set(&devr->s0->usecnt, 0);
1280
1281 return 0;
1282
1283 error5:
1284 mlx5_ib_destroy_srq(devr->s0);
1285 error4:
1286 mlx5_ib_dealloc_xrcd(devr->x1);
1287 error3:
1288 mlx5_ib_dealloc_xrcd(devr->x0);
1289 error2:
1290 mlx5_ib_destroy_cq(devr->c0);
1291 error1:
1292 mlx5_ib_dealloc_pd(devr->p0);
1293 error0:
1294 return ret;
1295 }
1296
1297 static void destroy_dev_resources(struct mlx5_ib_resources *devr)
1298 {
1299 mlx5_ib_destroy_srq(devr->s1);
1300 mlx5_ib_destroy_srq(devr->s0);
1301 mlx5_ib_dealloc_xrcd(devr->x0);
1302 mlx5_ib_dealloc_xrcd(devr->x1);
1303 mlx5_ib_destroy_cq(devr->c0);
1304 mlx5_ib_dealloc_pd(devr->p0);
1305 }
1306
1307 static int mlx5_port_immutable(struct ib_device *ibdev, u8 port_num,
1308 struct ib_port_immutable *immutable)
1309 {
1310 struct ib_port_attr attr;
1311 int err;
1312
1313 err = mlx5_ib_query_port(ibdev, port_num, &attr);
1314 if (err)
1315 return err;
1316
1317 immutable->pkey_tbl_len = attr.pkey_tbl_len;
1318 immutable->gid_tbl_len = attr.gid_tbl_len;
1319 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
1320 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
1321
1322 return 0;
1323 }
1324
1325 static void *mlx5_ib_add(struct mlx5_core_dev *mdev)
1326 {
1327 struct mlx5_ib_dev *dev;
1328 int err;
1329 int i;
1330
1331 /* don't create IB instance over Eth ports, no RoCE yet! */
1332 if (MLX5_CAP_GEN(mdev, port_type) == MLX5_CAP_PORT_TYPE_ETH)
1333 return NULL;
1334
1335 printk_once(KERN_INFO "%s", mlx5_version);
1336
1337 dev = (struct mlx5_ib_dev *)ib_alloc_device(sizeof(*dev));
1338 if (!dev)
1339 return NULL;
1340
1341 dev->mdev = mdev;
1342
1343 err = get_port_caps(dev);
1344 if (err)
1345 goto err_dealloc;
1346
1347 if (mlx5_use_mad_ifc(dev))
1348 get_ext_port_caps(dev);
1349
1350 MLX5_INIT_DOORBELL_LOCK(&dev->uar_lock);
1351
1352 strlcpy(dev->ib_dev.name, "mlx5_%d", IB_DEVICE_NAME_MAX);
1353 dev->ib_dev.owner = THIS_MODULE;
1354 dev->ib_dev.node_type = RDMA_NODE_IB_CA;
1355 dev->ib_dev.local_dma_lkey = 0 /* not supported for now */;
1356 dev->num_ports = MLX5_CAP_GEN(mdev, num_ports);
1357 dev->ib_dev.phys_port_cnt = dev->num_ports;
1358 dev->ib_dev.num_comp_vectors =
1359 dev->mdev->priv.eq_table.num_comp_vectors;
1360 dev->ib_dev.dma_device = &mdev->pdev->dev;
1361
1362 dev->ib_dev.uverbs_abi_ver = MLX5_IB_UVERBS_ABI_VERSION;
1363 dev->ib_dev.uverbs_cmd_mask =
1364 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1365 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1366 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1367 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1368 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1369 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1370 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1371 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1372 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1373 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1374 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1375 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1376 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1377 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1378 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1379 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1380 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
1381 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1382 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1383 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1384 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
1385 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ) |
1386 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1387 dev->ib_dev.uverbs_ex_cmd_mask =
1388 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE);
1389
1390 dev->ib_dev.query_device = mlx5_ib_query_device;
1391 dev->ib_dev.query_port = mlx5_ib_query_port;
1392 dev->ib_dev.query_gid = mlx5_ib_query_gid;
1393 dev->ib_dev.query_pkey = mlx5_ib_query_pkey;
1394 dev->ib_dev.modify_device = mlx5_ib_modify_device;
1395 dev->ib_dev.modify_port = mlx5_ib_modify_port;
1396 dev->ib_dev.alloc_ucontext = mlx5_ib_alloc_ucontext;
1397 dev->ib_dev.dealloc_ucontext = mlx5_ib_dealloc_ucontext;
1398 dev->ib_dev.mmap = mlx5_ib_mmap;
1399 dev->ib_dev.alloc_pd = mlx5_ib_alloc_pd;
1400 dev->ib_dev.dealloc_pd = mlx5_ib_dealloc_pd;
1401 dev->ib_dev.create_ah = mlx5_ib_create_ah;
1402 dev->ib_dev.query_ah = mlx5_ib_query_ah;
1403 dev->ib_dev.destroy_ah = mlx5_ib_destroy_ah;
1404 dev->ib_dev.create_srq = mlx5_ib_create_srq;
1405 dev->ib_dev.modify_srq = mlx5_ib_modify_srq;
1406 dev->ib_dev.query_srq = mlx5_ib_query_srq;
1407 dev->ib_dev.destroy_srq = mlx5_ib_destroy_srq;
1408 dev->ib_dev.post_srq_recv = mlx5_ib_post_srq_recv;
1409 dev->ib_dev.create_qp = mlx5_ib_create_qp;
1410 dev->ib_dev.modify_qp = mlx5_ib_modify_qp;
1411 dev->ib_dev.query_qp = mlx5_ib_query_qp;
1412 dev->ib_dev.destroy_qp = mlx5_ib_destroy_qp;
1413 dev->ib_dev.post_send = mlx5_ib_post_send;
1414 dev->ib_dev.post_recv = mlx5_ib_post_recv;
1415 dev->ib_dev.create_cq = mlx5_ib_create_cq;
1416 dev->ib_dev.modify_cq = mlx5_ib_modify_cq;
1417 dev->ib_dev.resize_cq = mlx5_ib_resize_cq;
1418 dev->ib_dev.destroy_cq = mlx5_ib_destroy_cq;
1419 dev->ib_dev.poll_cq = mlx5_ib_poll_cq;
1420 dev->ib_dev.req_notify_cq = mlx5_ib_arm_cq;
1421 dev->ib_dev.get_dma_mr = mlx5_ib_get_dma_mr;
1422 dev->ib_dev.reg_user_mr = mlx5_ib_reg_user_mr;
1423 dev->ib_dev.dereg_mr = mlx5_ib_dereg_mr;
1424 dev->ib_dev.attach_mcast = mlx5_ib_mcg_attach;
1425 dev->ib_dev.detach_mcast = mlx5_ib_mcg_detach;
1426 dev->ib_dev.process_mad = mlx5_ib_process_mad;
1427 dev->ib_dev.alloc_mr = mlx5_ib_alloc_mr;
1428 dev->ib_dev.map_mr_sg = mlx5_ib_map_mr_sg;
1429 dev->ib_dev.check_mr_status = mlx5_ib_check_mr_status;
1430 dev->ib_dev.get_port_immutable = mlx5_port_immutable;
1431
1432 mlx5_ib_internal_fill_odp_caps(dev);
1433
1434 if (MLX5_CAP_GEN(mdev, xrc)) {
1435 dev->ib_dev.alloc_xrcd = mlx5_ib_alloc_xrcd;
1436 dev->ib_dev.dealloc_xrcd = mlx5_ib_dealloc_xrcd;
1437 dev->ib_dev.uverbs_cmd_mask |=
1438 (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1439 (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1440 }
1441
1442 err = init_node_data(dev);
1443 if (err)
1444 goto err_dealloc;
1445
1446 mutex_init(&dev->cap_mask_mutex);
1447
1448 err = create_dev_resources(&dev->devr);
1449 if (err)
1450 goto err_dealloc;
1451
1452 err = mlx5_ib_odp_init_one(dev);
1453 if (err)
1454 goto err_rsrc;
1455
1456 err = ib_register_device(&dev->ib_dev, NULL);
1457 if (err)
1458 goto err_odp;
1459
1460 err = create_umr_res(dev);
1461 if (err)
1462 goto err_dev;
1463
1464 for (i = 0; i < ARRAY_SIZE(mlx5_class_attributes); i++) {
1465 err = device_create_file(&dev->ib_dev.dev,
1466 mlx5_class_attributes[i]);
1467 if (err)
1468 goto err_umrc;
1469 }
1470
1471 dev->ib_active = true;
1472
1473 return dev;
1474
1475 err_umrc:
1476 destroy_umrc_res(dev);
1477
1478 err_dev:
1479 ib_unregister_device(&dev->ib_dev);
1480
1481 err_odp:
1482 mlx5_ib_odp_remove_one(dev);
1483
1484 err_rsrc:
1485 destroy_dev_resources(&dev->devr);
1486
1487 err_dealloc:
1488 ib_dealloc_device((struct ib_device *)dev);
1489
1490 return NULL;
1491 }
1492
1493 static void mlx5_ib_remove(struct mlx5_core_dev *mdev, void *context)
1494 {
1495 struct mlx5_ib_dev *dev = context;
1496
1497 ib_unregister_device(&dev->ib_dev);
1498 destroy_umrc_res(dev);
1499 mlx5_ib_odp_remove_one(dev);
1500 destroy_dev_resources(&dev->devr);
1501 ib_dealloc_device(&dev->ib_dev);
1502 }
1503
1504 static struct mlx5_interface mlx5_ib_interface = {
1505 .add = mlx5_ib_add,
1506 .remove = mlx5_ib_remove,
1507 .event = mlx5_ib_event,
1508 .protocol = MLX5_INTERFACE_PROTOCOL_IB,
1509 };
1510
1511 static int __init mlx5_ib_init(void)
1512 {
1513 int err;
1514
1515 if (deprecated_prof_sel != 2)
1516 pr_warn("prof_sel is deprecated for mlx5_ib, set it for mlx5_core\n");
1517
1518 err = mlx5_ib_odp_init();
1519 if (err)
1520 return err;
1521
1522 err = mlx5_register_interface(&mlx5_ib_interface);
1523 if (err)
1524 goto clean_odp;
1525
1526 return err;
1527
1528 clean_odp:
1529 mlx5_ib_odp_cleanup();
1530 return err;
1531 }
1532
1533 static void __exit mlx5_ib_cleanup(void)
1534 {
1535 mlx5_unregister_interface(&mlx5_ib_interface);
1536 mlx5_ib_odp_cleanup();
1537 }
1538
1539 module_init(mlx5_ib_init);
1540 module_exit(mlx5_ib_cleanup);
Ubuntu Artful kernel mirror