xref: /openbmc/linux/drivers/infiniband/hw/mlx5/main.c (revision 0da85d1e)
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 <asm-generic/kmap_types.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 <rdma/ib_smi.h>
44 #include <rdma/ib_umem.h>
45 #include "user.h"
46 #include "mlx5_ib.h"
47 
48 #define DRIVER_NAME "mlx5_ib"
49 #define DRIVER_VERSION "2.2-1"
50 #define DRIVER_RELDATE	"Feb 2014"
51 
52 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
53 MODULE_DESCRIPTION("Mellanox Connect-IB HCA IB driver");
54 MODULE_LICENSE("Dual BSD/GPL");
55 MODULE_VERSION(DRIVER_VERSION);
56 
57 static int deprecated_prof_sel = 2;
58 module_param_named(prof_sel, deprecated_prof_sel, int, 0444);
59 MODULE_PARM_DESC(prof_sel, "profile selector. Deprecated here. Moved to module mlx5_core");
60 
61 static char mlx5_version[] =
62 	DRIVER_NAME ": Mellanox Connect-IB Infiniband driver v"
63 	DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
64 
65 static int mlx5_ib_query_device(struct ib_device *ibdev,
66 				struct ib_device_attr *props)
67 {
68 	struct mlx5_ib_dev *dev = to_mdev(ibdev);
69 	struct ib_smp *in_mad  = NULL;
70 	struct ib_smp *out_mad = NULL;
71 	struct mlx5_general_caps *gen;
72 	int err = -ENOMEM;
73 	int max_rq_sg;
74 	int max_sq_sg;
75 	u64 flags;
76 
77 	gen = &dev->mdev->caps.gen;
78 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
79 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
80 	if (!in_mad || !out_mad)
81 		goto out;
82 
83 	init_query_mad(in_mad);
84 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
85 
86 	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
87 	if (err)
88 		goto out;
89 
90 	memset(props, 0, sizeof(*props));
91 
92 	props->fw_ver = ((u64)fw_rev_maj(dev->mdev) << 32) |
93 		(fw_rev_min(dev->mdev) << 16) |
94 		fw_rev_sub(dev->mdev);
95 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
96 		IB_DEVICE_PORT_ACTIVE_EVENT		|
97 		IB_DEVICE_SYS_IMAGE_GUID		|
98 		IB_DEVICE_RC_RNR_NAK_GEN;
99 	flags = gen->flags;
100 	if (flags & MLX5_DEV_CAP_FLAG_BAD_PKEY_CNTR)
101 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
102 	if (flags & MLX5_DEV_CAP_FLAG_BAD_QKEY_CNTR)
103 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
104 	if (flags & MLX5_DEV_CAP_FLAG_APM)
105 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
106 	props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
107 	if (flags & MLX5_DEV_CAP_FLAG_XRC)
108 		props->device_cap_flags |= IB_DEVICE_XRC;
109 	props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
110 	if (flags & MLX5_DEV_CAP_FLAG_SIG_HAND_OVER) {
111 		props->device_cap_flags |= IB_DEVICE_SIGNATURE_HANDOVER;
112 		/* At this stage no support for signature handover */
113 		props->sig_prot_cap = IB_PROT_T10DIF_TYPE_1 |
114 				      IB_PROT_T10DIF_TYPE_2 |
115 				      IB_PROT_T10DIF_TYPE_3;
116 		props->sig_guard_cap = IB_GUARD_T10DIF_CRC |
117 				       IB_GUARD_T10DIF_CSUM;
118 	}
119 	if (flags & MLX5_DEV_CAP_FLAG_BLOCK_MCAST)
120 		props->device_cap_flags |= IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
121 
122 	props->vendor_id	   = be32_to_cpup((__be32 *)(out_mad->data + 36)) &
123 		0xffffff;
124 	props->vendor_part_id	   = be16_to_cpup((__be16 *)(out_mad->data + 30));
125 	props->hw_ver		   = be32_to_cpup((__be32 *)(out_mad->data + 32));
126 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
127 
128 	props->max_mr_size	   = ~0ull;
129 	props->page_size_cap	   = gen->min_page_sz;
130 	props->max_qp		   = 1 << gen->log_max_qp;
131 	props->max_qp_wr	   = gen->max_wqes;
132 	max_rq_sg = gen->max_rq_desc_sz / sizeof(struct mlx5_wqe_data_seg);
133 	max_sq_sg = (gen->max_sq_desc_sz - sizeof(struct mlx5_wqe_ctrl_seg)) /
134 		sizeof(struct mlx5_wqe_data_seg);
135 	props->max_sge = min(max_rq_sg, max_sq_sg);
136 	props->max_cq		   = 1 << gen->log_max_cq;
137 	props->max_cqe		   = gen->max_cqes - 1;
138 	props->max_mr		   = 1 << gen->log_max_mkey;
139 	props->max_pd		   = 1 << gen->log_max_pd;
140 	props->max_qp_rd_atom	   = 1 << gen->log_max_ra_req_qp;
141 	props->max_qp_init_rd_atom = 1 << gen->log_max_ra_res_qp;
142 	props->max_srq		   = 1 << gen->log_max_srq;
143 	props->max_srq_wr	   = gen->max_srq_wqes - 1;
144 	props->local_ca_ack_delay  = gen->local_ca_ack_delay;
145 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
146 	props->max_srq_sge	   = max_rq_sg - 1;
147 	props->max_fast_reg_page_list_len = (unsigned int)-1;
148 	props->local_ca_ack_delay  = gen->local_ca_ack_delay;
149 	props->atomic_cap	   = IB_ATOMIC_NONE;
150 	props->masked_atomic_cap   = IB_ATOMIC_NONE;
151 	props->max_pkeys	   = be16_to_cpup((__be16 *)(out_mad->data + 28));
152 	props->max_mcast_grp	   = 1 << gen->log_max_mcg;
153 	props->max_mcast_qp_attach = gen->max_qp_mcg;
154 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
155 					   props->max_mcast_grp;
156 	props->max_map_per_fmr = INT_MAX; /* no limit in ConnectIB */
157 
158 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
159 	if (dev->mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_ON_DMND_PG)
160 		props->device_cap_flags |= IB_DEVICE_ON_DEMAND_PAGING;
161 	props->odp_caps = dev->odp_caps;
162 #endif
163 
164 out:
165 	kfree(in_mad);
166 	kfree(out_mad);
167 
168 	return err;
169 }
170 
171 int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
172 		       struct ib_port_attr *props)
173 {
174 	struct mlx5_ib_dev *dev = to_mdev(ibdev);
175 	struct ib_smp *in_mad  = NULL;
176 	struct ib_smp *out_mad = NULL;
177 	struct mlx5_general_caps *gen;
178 	int ext_active_speed;
179 	int err = -ENOMEM;
180 
181 	gen = &dev->mdev->caps.gen;
182 	if (port < 1 || port > gen->num_ports) {
183 		mlx5_ib_warn(dev, "invalid port number %d\n", port);
184 		return -EINVAL;
185 	}
186 
187 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
188 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
189 	if (!in_mad || !out_mad)
190 		goto out;
191 
192 	memset(props, 0, sizeof(*props));
193 
194 	init_query_mad(in_mad);
195 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
196 	in_mad->attr_mod = cpu_to_be32(port);
197 
198 	err = mlx5_MAD_IFC(dev, 1, 1, port, NULL, NULL, in_mad, out_mad);
199 	if (err) {
200 		mlx5_ib_warn(dev, "err %d\n", err);
201 		goto out;
202 	}
203 
204 
205 	props->lid		= be16_to_cpup((__be16 *)(out_mad->data + 16));
206 	props->lmc		= out_mad->data[34] & 0x7;
207 	props->sm_lid		= be16_to_cpup((__be16 *)(out_mad->data + 18));
208 	props->sm_sl		= out_mad->data[36] & 0xf;
209 	props->state		= out_mad->data[32] & 0xf;
210 	props->phys_state	= out_mad->data[33] >> 4;
211 	props->port_cap_flags	= be32_to_cpup((__be32 *)(out_mad->data + 20));
212 	props->gid_tbl_len	= out_mad->data[50];
213 	props->max_msg_sz	= 1 << gen->log_max_msg;
214 	props->pkey_tbl_len	= gen->port[port - 1].pkey_table_len;
215 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *)(out_mad->data + 46));
216 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *)(out_mad->data + 48));
217 	props->active_width	= out_mad->data[31] & 0xf;
218 	props->active_speed	= out_mad->data[35] >> 4;
219 	props->max_mtu		= out_mad->data[41] & 0xf;
220 	props->active_mtu	= out_mad->data[36] >> 4;
221 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
222 	props->max_vl_num	= out_mad->data[37] >> 4;
223 	props->init_type_reply	= out_mad->data[41] >> 4;
224 
225 	/* Check if extended speeds (EDR/FDR/...) are supported */
226 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
227 		ext_active_speed = out_mad->data[62] >> 4;
228 
229 		switch (ext_active_speed) {
230 		case 1:
231 			props->active_speed = 16; /* FDR */
232 			break;
233 		case 2:
234 			props->active_speed = 32; /* EDR */
235 			break;
236 		}
237 	}
238 
239 	/* If reported active speed is QDR, check if is FDR-10 */
240 	if (props->active_speed == 4) {
241 		if (gen->ext_port_cap[port - 1] &
242 		    MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO) {
243 			init_query_mad(in_mad);
244 			in_mad->attr_id = MLX5_ATTR_EXTENDED_PORT_INFO;
245 			in_mad->attr_mod = cpu_to_be32(port);
246 
247 			err = mlx5_MAD_IFC(dev, 1, 1, port,
248 					   NULL, NULL, in_mad, out_mad);
249 			if (err)
250 				goto out;
251 
252 			/* Checking LinkSpeedActive for FDR-10 */
253 			if (out_mad->data[15] & 0x1)
254 				props->active_speed = 8;
255 		}
256 	}
257 
258 out:
259 	kfree(in_mad);
260 	kfree(out_mad);
261 
262 	return err;
263 }
264 
265 static int mlx5_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
266 			     union ib_gid *gid)
267 {
268 	struct ib_smp *in_mad  = NULL;
269 	struct ib_smp *out_mad = NULL;
270 	int err = -ENOMEM;
271 
272 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
273 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
274 	if (!in_mad || !out_mad)
275 		goto out;
276 
277 	init_query_mad(in_mad);
278 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
279 	in_mad->attr_mod = cpu_to_be32(port);
280 
281 	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
282 	if (err)
283 		goto out;
284 
285 	memcpy(gid->raw, out_mad->data + 8, 8);
286 
287 	init_query_mad(in_mad);
288 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
289 	in_mad->attr_mod = cpu_to_be32(index / 8);
290 
291 	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
292 	if (err)
293 		goto out;
294 
295 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
296 
297 out:
298 	kfree(in_mad);
299 	kfree(out_mad);
300 	return err;
301 }
302 
303 static int mlx5_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
304 			      u16 *pkey)
305 {
306 	struct ib_smp *in_mad  = NULL;
307 	struct ib_smp *out_mad = NULL;
308 	int err = -ENOMEM;
309 
310 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
311 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
312 	if (!in_mad || !out_mad)
313 		goto out;
314 
315 	init_query_mad(in_mad);
316 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
317 	in_mad->attr_mod = cpu_to_be32(index / 32);
318 
319 	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
320 	if (err)
321 		goto out;
322 
323 	*pkey = be16_to_cpu(((__be16 *)out_mad->data)[index % 32]);
324 
325 out:
326 	kfree(in_mad);
327 	kfree(out_mad);
328 	return err;
329 }
330 
331 struct mlx5_reg_node_desc {
332 	u8	desc[64];
333 };
334 
335 static int mlx5_ib_modify_device(struct ib_device *ibdev, int mask,
336 				 struct ib_device_modify *props)
337 {
338 	struct mlx5_ib_dev *dev = to_mdev(ibdev);
339 	struct mlx5_reg_node_desc in;
340 	struct mlx5_reg_node_desc out;
341 	int err;
342 
343 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
344 		return -EOPNOTSUPP;
345 
346 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
347 		return 0;
348 
349 	/*
350 	 * If possible, pass node desc to FW, so it can generate
351 	 * a 144 trap.  If cmd fails, just ignore.
352 	 */
353 	memcpy(&in, props->node_desc, 64);
354 	err = mlx5_core_access_reg(dev->mdev, &in, sizeof(in), &out,
355 				   sizeof(out), MLX5_REG_NODE_DESC, 0, 1);
356 	if (err)
357 		return err;
358 
359 	memcpy(ibdev->node_desc, props->node_desc, 64);
360 
361 	return err;
362 }
363 
364 static int mlx5_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
365 			       struct ib_port_modify *props)
366 {
367 	struct mlx5_ib_dev *dev = to_mdev(ibdev);
368 	struct ib_port_attr attr;
369 	u32 tmp;
370 	int err;
371 
372 	mutex_lock(&dev->cap_mask_mutex);
373 
374 	err = mlx5_ib_query_port(ibdev, port, &attr);
375 	if (err)
376 		goto out;
377 
378 	tmp = (attr.port_cap_flags | props->set_port_cap_mask) &
379 		~props->clr_port_cap_mask;
380 
381 	err = mlx5_set_port_caps(dev->mdev, port, tmp);
382 
383 out:
384 	mutex_unlock(&dev->cap_mask_mutex);
385 	return err;
386 }
387 
388 static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev,
389 						  struct ib_udata *udata)
390 {
391 	struct mlx5_ib_dev *dev = to_mdev(ibdev);
392 	struct mlx5_ib_alloc_ucontext_req_v2 req;
393 	struct mlx5_ib_alloc_ucontext_resp resp;
394 	struct mlx5_ib_ucontext *context;
395 	struct mlx5_general_caps *gen;
396 	struct mlx5_uuar_info *uuari;
397 	struct mlx5_uar *uars;
398 	int gross_uuars;
399 	int num_uars;
400 	int ver;
401 	int uuarn;
402 	int err;
403 	int i;
404 	size_t reqlen;
405 
406 	gen = &dev->mdev->caps.gen;
407 	if (!dev->ib_active)
408 		return ERR_PTR(-EAGAIN);
409 
410 	memset(&req, 0, sizeof(req));
411 	reqlen = udata->inlen - sizeof(struct ib_uverbs_cmd_hdr);
412 	if (reqlen == sizeof(struct mlx5_ib_alloc_ucontext_req))
413 		ver = 0;
414 	else if (reqlen == sizeof(struct mlx5_ib_alloc_ucontext_req_v2))
415 		ver = 2;
416 	else
417 		return ERR_PTR(-EINVAL);
418 
419 	err = ib_copy_from_udata(&req, udata, reqlen);
420 	if (err)
421 		return ERR_PTR(err);
422 
423 	if (req.flags || req.reserved)
424 		return ERR_PTR(-EINVAL);
425 
426 	if (req.total_num_uuars > MLX5_MAX_UUARS)
427 		return ERR_PTR(-ENOMEM);
428 
429 	if (req.total_num_uuars == 0)
430 		return ERR_PTR(-EINVAL);
431 
432 	req.total_num_uuars = ALIGN(req.total_num_uuars,
433 				    MLX5_NON_FP_BF_REGS_PER_PAGE);
434 	if (req.num_low_latency_uuars > req.total_num_uuars - 1)
435 		return ERR_PTR(-EINVAL);
436 
437 	num_uars = req.total_num_uuars / MLX5_NON_FP_BF_REGS_PER_PAGE;
438 	gross_uuars = num_uars * MLX5_BF_REGS_PER_PAGE;
439 	resp.qp_tab_size      = 1 << gen->log_max_qp;
440 	resp.bf_reg_size      = gen->bf_reg_size;
441 	resp.cache_line_size  = L1_CACHE_BYTES;
442 	resp.max_sq_desc_sz = gen->max_sq_desc_sz;
443 	resp.max_rq_desc_sz = gen->max_rq_desc_sz;
444 	resp.max_send_wqebb = gen->max_wqes;
445 	resp.max_recv_wr = gen->max_wqes;
446 	resp.max_srq_recv_wr = gen->max_srq_wqes;
447 
448 	context = kzalloc(sizeof(*context), GFP_KERNEL);
449 	if (!context)
450 		return ERR_PTR(-ENOMEM);
451 
452 	uuari = &context->uuari;
453 	mutex_init(&uuari->lock);
454 	uars = kcalloc(num_uars, sizeof(*uars), GFP_KERNEL);
455 	if (!uars) {
456 		err = -ENOMEM;
457 		goto out_ctx;
458 	}
459 
460 	uuari->bitmap = kcalloc(BITS_TO_LONGS(gross_uuars),
461 				sizeof(*uuari->bitmap),
462 				GFP_KERNEL);
463 	if (!uuari->bitmap) {
464 		err = -ENOMEM;
465 		goto out_uar_ctx;
466 	}
467 	/*
468 	 * clear all fast path uuars
469 	 */
470 	for (i = 0; i < gross_uuars; i++) {
471 		uuarn = i & 3;
472 		if (uuarn == 2 || uuarn == 3)
473 			set_bit(i, uuari->bitmap);
474 	}
475 
476 	uuari->count = kcalloc(gross_uuars, sizeof(*uuari->count), GFP_KERNEL);
477 	if (!uuari->count) {
478 		err = -ENOMEM;
479 		goto out_bitmap;
480 	}
481 
482 	for (i = 0; i < num_uars; i++) {
483 		err = mlx5_cmd_alloc_uar(dev->mdev, &uars[i].index);
484 		if (err)
485 			goto out_count;
486 	}
487 
488 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
489 	context->ibucontext.invalidate_range = &mlx5_ib_invalidate_range;
490 #endif
491 
492 	INIT_LIST_HEAD(&context->db_page_list);
493 	mutex_init(&context->db_page_mutex);
494 
495 	resp.tot_uuars = req.total_num_uuars;
496 	resp.num_ports = gen->num_ports;
497 	err = ib_copy_to_udata(udata, &resp,
498 			       sizeof(resp) - sizeof(resp.reserved));
499 	if (err)
500 		goto out_uars;
501 
502 	uuari->ver = ver;
503 	uuari->num_low_latency_uuars = req.num_low_latency_uuars;
504 	uuari->uars = uars;
505 	uuari->num_uars = num_uars;
506 	return &context->ibucontext;
507 
508 out_uars:
509 	for (i--; i >= 0; i--)
510 		mlx5_cmd_free_uar(dev->mdev, uars[i].index);
511 out_count:
512 	kfree(uuari->count);
513 
514 out_bitmap:
515 	kfree(uuari->bitmap);
516 
517 out_uar_ctx:
518 	kfree(uars);
519 
520 out_ctx:
521 	kfree(context);
522 	return ERR_PTR(err);
523 }
524 
525 static int mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
526 {
527 	struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
528 	struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
529 	struct mlx5_uuar_info *uuari = &context->uuari;
530 	int i;
531 
532 	for (i = 0; i < uuari->num_uars; i++) {
533 		if (mlx5_cmd_free_uar(dev->mdev, uuari->uars[i].index))
534 			mlx5_ib_warn(dev, "failed to free UAR 0x%x\n", uuari->uars[i].index);
535 	}
536 
537 	kfree(uuari->count);
538 	kfree(uuari->bitmap);
539 	kfree(uuari->uars);
540 	kfree(context);
541 
542 	return 0;
543 }
544 
545 static phys_addr_t uar_index2pfn(struct mlx5_ib_dev *dev, int index)
546 {
547 	return (pci_resource_start(dev->mdev->pdev, 0) >> PAGE_SHIFT) + index;
548 }
549 
550 static int get_command(unsigned long offset)
551 {
552 	return (offset >> MLX5_IB_MMAP_CMD_SHIFT) & MLX5_IB_MMAP_CMD_MASK;
553 }
554 
555 static int get_arg(unsigned long offset)
556 {
557 	return offset & ((1 << MLX5_IB_MMAP_CMD_SHIFT) - 1);
558 }
559 
560 static int get_index(unsigned long offset)
561 {
562 	return get_arg(offset);
563 }
564 
565 static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
566 {
567 	struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
568 	struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
569 	struct mlx5_uuar_info *uuari = &context->uuari;
570 	unsigned long command;
571 	unsigned long idx;
572 	phys_addr_t pfn;
573 
574 	command = get_command(vma->vm_pgoff);
575 	switch (command) {
576 	case MLX5_IB_MMAP_REGULAR_PAGE:
577 		if (vma->vm_end - vma->vm_start != PAGE_SIZE)
578 			return -EINVAL;
579 
580 		idx = get_index(vma->vm_pgoff);
581 		if (idx >= uuari->num_uars)
582 			return -EINVAL;
583 
584 		pfn = uar_index2pfn(dev, uuari->uars[idx].index);
585 		mlx5_ib_dbg(dev, "uar idx 0x%lx, pfn 0x%llx\n", idx,
586 			    (unsigned long long)pfn);
587 
588 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
589 		if (io_remap_pfn_range(vma, vma->vm_start, pfn,
590 				       PAGE_SIZE, vma->vm_page_prot))
591 			return -EAGAIN;
592 
593 		mlx5_ib_dbg(dev, "mapped WC at 0x%lx, PA 0x%llx\n",
594 			    vma->vm_start,
595 			    (unsigned long long)pfn << PAGE_SHIFT);
596 		break;
597 
598 	case MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES:
599 		return -ENOSYS;
600 
601 	default:
602 		return -EINVAL;
603 	}
604 
605 	return 0;
606 }
607 
608 static int alloc_pa_mkey(struct mlx5_ib_dev *dev, u32 *key, u32 pdn)
609 {
610 	struct mlx5_create_mkey_mbox_in *in;
611 	struct mlx5_mkey_seg *seg;
612 	struct mlx5_core_mr mr;
613 	int err;
614 
615 	in = kzalloc(sizeof(*in), GFP_KERNEL);
616 	if (!in)
617 		return -ENOMEM;
618 
619 	seg = &in->seg;
620 	seg->flags = MLX5_PERM_LOCAL_READ | MLX5_ACCESS_MODE_PA;
621 	seg->flags_pd = cpu_to_be32(pdn | MLX5_MKEY_LEN64);
622 	seg->qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
623 	seg->start_addr = 0;
624 
625 	err = mlx5_core_create_mkey(dev->mdev, &mr, in, sizeof(*in),
626 				    NULL, NULL, NULL);
627 	if (err) {
628 		mlx5_ib_warn(dev, "failed to create mkey, %d\n", err);
629 		goto err_in;
630 	}
631 
632 	kfree(in);
633 	*key = mr.key;
634 
635 	return 0;
636 
637 err_in:
638 	kfree(in);
639 
640 	return err;
641 }
642 
643 static void free_pa_mkey(struct mlx5_ib_dev *dev, u32 key)
644 {
645 	struct mlx5_core_mr mr;
646 	int err;
647 
648 	memset(&mr, 0, sizeof(mr));
649 	mr.key = key;
650 	err = mlx5_core_destroy_mkey(dev->mdev, &mr);
651 	if (err)
652 		mlx5_ib_warn(dev, "failed to destroy mkey 0x%x\n", key);
653 }
654 
655 static struct ib_pd *mlx5_ib_alloc_pd(struct ib_device *ibdev,
656 				      struct ib_ucontext *context,
657 				      struct ib_udata *udata)
658 {
659 	struct mlx5_ib_alloc_pd_resp resp;
660 	struct mlx5_ib_pd *pd;
661 	int err;
662 
663 	pd = kmalloc(sizeof(*pd), GFP_KERNEL);
664 	if (!pd)
665 		return ERR_PTR(-ENOMEM);
666 
667 	err = mlx5_core_alloc_pd(to_mdev(ibdev)->mdev, &pd->pdn);
668 	if (err) {
669 		kfree(pd);
670 		return ERR_PTR(err);
671 	}
672 
673 	if (context) {
674 		resp.pdn = pd->pdn;
675 		if (ib_copy_to_udata(udata, &resp, sizeof(resp))) {
676 			mlx5_core_dealloc_pd(to_mdev(ibdev)->mdev, pd->pdn);
677 			kfree(pd);
678 			return ERR_PTR(-EFAULT);
679 		}
680 	} else {
681 		err = alloc_pa_mkey(to_mdev(ibdev), &pd->pa_lkey, pd->pdn);
682 		if (err) {
683 			mlx5_core_dealloc_pd(to_mdev(ibdev)->mdev, pd->pdn);
684 			kfree(pd);
685 			return ERR_PTR(err);
686 		}
687 	}
688 
689 	return &pd->ibpd;
690 }
691 
692 static int mlx5_ib_dealloc_pd(struct ib_pd *pd)
693 {
694 	struct mlx5_ib_dev *mdev = to_mdev(pd->device);
695 	struct mlx5_ib_pd *mpd = to_mpd(pd);
696 
697 	if (!pd->uobject)
698 		free_pa_mkey(mdev, mpd->pa_lkey);
699 
700 	mlx5_core_dealloc_pd(mdev->mdev, mpd->pdn);
701 	kfree(mpd);
702 
703 	return 0;
704 }
705 
706 static int mlx5_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
707 {
708 	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
709 	int err;
710 
711 	err = mlx5_core_attach_mcg(dev->mdev, gid, ibqp->qp_num);
712 	if (err)
713 		mlx5_ib_warn(dev, "failed attaching QPN 0x%x, MGID %pI6\n",
714 			     ibqp->qp_num, gid->raw);
715 
716 	return err;
717 }
718 
719 static int mlx5_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
720 {
721 	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
722 	int err;
723 
724 	err = mlx5_core_detach_mcg(dev->mdev, gid, ibqp->qp_num);
725 	if (err)
726 		mlx5_ib_warn(dev, "failed detaching QPN 0x%x, MGID %pI6\n",
727 			     ibqp->qp_num, gid->raw);
728 
729 	return err;
730 }
731 
732 static int init_node_data(struct mlx5_ib_dev *dev)
733 {
734 	struct ib_smp *in_mad  = NULL;
735 	struct ib_smp *out_mad = NULL;
736 	int err = -ENOMEM;
737 
738 	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
739 	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
740 	if (!in_mad || !out_mad)
741 		goto out;
742 
743 	init_query_mad(in_mad);
744 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
745 
746 	err = mlx5_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
747 	if (err)
748 		goto out;
749 
750 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
751 
752 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
753 
754 	err = mlx5_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
755 	if (err)
756 		goto out;
757 
758 	dev->mdev->rev_id = be32_to_cpup((__be32 *)(out_mad->data + 32));
759 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
760 
761 out:
762 	kfree(in_mad);
763 	kfree(out_mad);
764 	return err;
765 }
766 
767 static ssize_t show_fw_pages(struct device *device, struct device_attribute *attr,
768 			     char *buf)
769 {
770 	struct mlx5_ib_dev *dev =
771 		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
772 
773 	return sprintf(buf, "%d\n", dev->mdev->priv.fw_pages);
774 }
775 
776 static ssize_t show_reg_pages(struct device *device,
777 			      struct device_attribute *attr, char *buf)
778 {
779 	struct mlx5_ib_dev *dev =
780 		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
781 
782 	return sprintf(buf, "%d\n", atomic_read(&dev->mdev->priv.reg_pages));
783 }
784 
785 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
786 			char *buf)
787 {
788 	struct mlx5_ib_dev *dev =
789 		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
790 	return sprintf(buf, "MT%d\n", dev->mdev->pdev->device);
791 }
792 
793 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
794 			   char *buf)
795 {
796 	struct mlx5_ib_dev *dev =
797 		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
798 	return sprintf(buf, "%d.%d.%d\n", fw_rev_maj(dev->mdev),
799 		       fw_rev_min(dev->mdev), fw_rev_sub(dev->mdev));
800 }
801 
802 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
803 			char *buf)
804 {
805 	struct mlx5_ib_dev *dev =
806 		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
807 	return sprintf(buf, "%x\n", dev->mdev->rev_id);
808 }
809 
810 static ssize_t show_board(struct device *device, struct device_attribute *attr,
811 			  char *buf)
812 {
813 	struct mlx5_ib_dev *dev =
814 		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
815 	return sprintf(buf, "%.*s\n", MLX5_BOARD_ID_LEN,
816 		       dev->mdev->board_id);
817 }
818 
819 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
820 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
821 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
822 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
823 static DEVICE_ATTR(fw_pages, S_IRUGO, show_fw_pages, NULL);
824 static DEVICE_ATTR(reg_pages, S_IRUGO, show_reg_pages, NULL);
825 
826 static struct device_attribute *mlx5_class_attributes[] = {
827 	&dev_attr_hw_rev,
828 	&dev_attr_fw_ver,
829 	&dev_attr_hca_type,
830 	&dev_attr_board_id,
831 	&dev_attr_fw_pages,
832 	&dev_attr_reg_pages,
833 };
834 
835 static void mlx5_ib_event(struct mlx5_core_dev *dev, void *context,
836 			  enum mlx5_dev_event event, unsigned long param)
837 {
838 	struct mlx5_ib_dev *ibdev = (struct mlx5_ib_dev *)context;
839 	struct ib_event ibev;
840 
841 	u8 port = 0;
842 
843 	switch (event) {
844 	case MLX5_DEV_EVENT_SYS_ERROR:
845 		ibdev->ib_active = false;
846 		ibev.event = IB_EVENT_DEVICE_FATAL;
847 		break;
848 
849 	case MLX5_DEV_EVENT_PORT_UP:
850 		ibev.event = IB_EVENT_PORT_ACTIVE;
851 		port = (u8)param;
852 		break;
853 
854 	case MLX5_DEV_EVENT_PORT_DOWN:
855 		ibev.event = IB_EVENT_PORT_ERR;
856 		port = (u8)param;
857 		break;
858 
859 	case MLX5_DEV_EVENT_PORT_INITIALIZED:
860 		/* not used by ULPs */
861 		return;
862 
863 	case MLX5_DEV_EVENT_LID_CHANGE:
864 		ibev.event = IB_EVENT_LID_CHANGE;
865 		port = (u8)param;
866 		break;
867 
868 	case MLX5_DEV_EVENT_PKEY_CHANGE:
869 		ibev.event = IB_EVENT_PKEY_CHANGE;
870 		port = (u8)param;
871 		break;
872 
873 	case MLX5_DEV_EVENT_GUID_CHANGE:
874 		ibev.event = IB_EVENT_GID_CHANGE;
875 		port = (u8)param;
876 		break;
877 
878 	case MLX5_DEV_EVENT_CLIENT_REREG:
879 		ibev.event = IB_EVENT_CLIENT_REREGISTER;
880 		port = (u8)param;
881 		break;
882 	}
883 
884 	ibev.device	      = &ibdev->ib_dev;
885 	ibev.element.port_num = port;
886 
887 	if (port < 1 || port > ibdev->num_ports) {
888 		mlx5_ib_warn(ibdev, "warning: event on port %d\n", port);
889 		return;
890 	}
891 
892 	if (ibdev->ib_active)
893 		ib_dispatch_event(&ibev);
894 }
895 
896 static void get_ext_port_caps(struct mlx5_ib_dev *dev)
897 {
898 	struct mlx5_general_caps *gen;
899 	int port;
900 
901 	gen = &dev->mdev->caps.gen;
902 	for (port = 1; port <= gen->num_ports; port++)
903 		mlx5_query_ext_port_caps(dev, port);
904 }
905 
906 static int get_port_caps(struct mlx5_ib_dev *dev)
907 {
908 	struct ib_device_attr *dprops = NULL;
909 	struct ib_port_attr *pprops = NULL;
910 	struct mlx5_general_caps *gen;
911 	int err = -ENOMEM;
912 	int port;
913 
914 	gen = &dev->mdev->caps.gen;
915 	pprops = kmalloc(sizeof(*pprops), GFP_KERNEL);
916 	if (!pprops)
917 		goto out;
918 
919 	dprops = kmalloc(sizeof(*dprops), GFP_KERNEL);
920 	if (!dprops)
921 		goto out;
922 
923 	err = mlx5_ib_query_device(&dev->ib_dev, dprops);
924 	if (err) {
925 		mlx5_ib_warn(dev, "query_device failed %d\n", err);
926 		goto out;
927 	}
928 
929 	for (port = 1; port <= gen->num_ports; port++) {
930 		err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
931 		if (err) {
932 			mlx5_ib_warn(dev, "query_port %d failed %d\n", port, err);
933 			break;
934 		}
935 		gen->port[port - 1].pkey_table_len = dprops->max_pkeys;
936 		gen->port[port - 1].gid_table_len = pprops->gid_tbl_len;
937 		mlx5_ib_dbg(dev, "pkey_table_len %d, gid_table_len %d\n",
938 			    dprops->max_pkeys, pprops->gid_tbl_len);
939 	}
940 
941 out:
942 	kfree(pprops);
943 	kfree(dprops);
944 
945 	return err;
946 }
947 
948 static void destroy_umrc_res(struct mlx5_ib_dev *dev)
949 {
950 	int err;
951 
952 	err = mlx5_mr_cache_cleanup(dev);
953 	if (err)
954 		mlx5_ib_warn(dev, "mr cache cleanup failed\n");
955 
956 	mlx5_ib_destroy_qp(dev->umrc.qp);
957 	ib_destroy_cq(dev->umrc.cq);
958 	ib_dereg_mr(dev->umrc.mr);
959 	ib_dealloc_pd(dev->umrc.pd);
960 }
961 
962 enum {
963 	MAX_UMR_WR = 128,
964 };
965 
966 static int create_umr_res(struct mlx5_ib_dev *dev)
967 {
968 	struct ib_qp_init_attr *init_attr = NULL;
969 	struct ib_qp_attr *attr = NULL;
970 	struct ib_pd *pd;
971 	struct ib_cq *cq;
972 	struct ib_qp *qp;
973 	struct ib_mr *mr;
974 	int ret;
975 
976 	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
977 	init_attr = kzalloc(sizeof(*init_attr), GFP_KERNEL);
978 	if (!attr || !init_attr) {
979 		ret = -ENOMEM;
980 		goto error_0;
981 	}
982 
983 	pd = ib_alloc_pd(&dev->ib_dev);
984 	if (IS_ERR(pd)) {
985 		mlx5_ib_dbg(dev, "Couldn't create PD for sync UMR QP\n");
986 		ret = PTR_ERR(pd);
987 		goto error_0;
988 	}
989 
990 	mr = ib_get_dma_mr(pd,  IB_ACCESS_LOCAL_WRITE);
991 	if (IS_ERR(mr)) {
992 		mlx5_ib_dbg(dev, "Couldn't create DMA MR for sync UMR QP\n");
993 		ret = PTR_ERR(mr);
994 		goto error_1;
995 	}
996 
997 	cq = ib_create_cq(&dev->ib_dev, mlx5_umr_cq_handler, NULL, NULL, 128,
998 			  0);
999 	if (IS_ERR(cq)) {
1000 		mlx5_ib_dbg(dev, "Couldn't create CQ for sync UMR QP\n");
1001 		ret = PTR_ERR(cq);
1002 		goto error_2;
1003 	}
1004 	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
1005 
1006 	init_attr->send_cq = cq;
1007 	init_attr->recv_cq = cq;
1008 	init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
1009 	init_attr->cap.max_send_wr = MAX_UMR_WR;
1010 	init_attr->cap.max_send_sge = 1;
1011 	init_attr->qp_type = MLX5_IB_QPT_REG_UMR;
1012 	init_attr->port_num = 1;
1013 	qp = mlx5_ib_create_qp(pd, init_attr, NULL);
1014 	if (IS_ERR(qp)) {
1015 		mlx5_ib_dbg(dev, "Couldn't create sync UMR QP\n");
1016 		ret = PTR_ERR(qp);
1017 		goto error_3;
1018 	}
1019 	qp->device     = &dev->ib_dev;
1020 	qp->real_qp    = qp;
1021 	qp->uobject    = NULL;
1022 	qp->qp_type    = MLX5_IB_QPT_REG_UMR;
1023 
1024 	attr->qp_state = IB_QPS_INIT;
1025 	attr->port_num = 1;
1026 	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE | IB_QP_PKEY_INDEX |
1027 				IB_QP_PORT, NULL);
1028 	if (ret) {
1029 		mlx5_ib_dbg(dev, "Couldn't modify UMR QP\n");
1030 		goto error_4;
1031 	}
1032 
1033 	memset(attr, 0, sizeof(*attr));
1034 	attr->qp_state = IB_QPS_RTR;
1035 	attr->path_mtu = IB_MTU_256;
1036 
1037 	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
1038 	if (ret) {
1039 		mlx5_ib_dbg(dev, "Couldn't modify umr QP to rtr\n");
1040 		goto error_4;
1041 	}
1042 
1043 	memset(attr, 0, sizeof(*attr));
1044 	attr->qp_state = IB_QPS_RTS;
1045 	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
1046 	if (ret) {
1047 		mlx5_ib_dbg(dev, "Couldn't modify umr QP to rts\n");
1048 		goto error_4;
1049 	}
1050 
1051 	dev->umrc.qp = qp;
1052 	dev->umrc.cq = cq;
1053 	dev->umrc.mr = mr;
1054 	dev->umrc.pd = pd;
1055 
1056 	sema_init(&dev->umrc.sem, MAX_UMR_WR);
1057 	ret = mlx5_mr_cache_init(dev);
1058 	if (ret) {
1059 		mlx5_ib_warn(dev, "mr cache init failed %d\n", ret);
1060 		goto error_4;
1061 	}
1062 
1063 	kfree(attr);
1064 	kfree(init_attr);
1065 
1066 	return 0;
1067 
1068 error_4:
1069 	mlx5_ib_destroy_qp(qp);
1070 
1071 error_3:
1072 	ib_destroy_cq(cq);
1073 
1074 error_2:
1075 	ib_dereg_mr(mr);
1076 
1077 error_1:
1078 	ib_dealloc_pd(pd);
1079 
1080 error_0:
1081 	kfree(attr);
1082 	kfree(init_attr);
1083 	return ret;
1084 }
1085 
1086 static int create_dev_resources(struct mlx5_ib_resources *devr)
1087 {
1088 	struct ib_srq_init_attr attr;
1089 	struct mlx5_ib_dev *dev;
1090 	int ret = 0;
1091 
1092 	dev = container_of(devr, struct mlx5_ib_dev, devr);
1093 
1094 	devr->p0 = mlx5_ib_alloc_pd(&dev->ib_dev, NULL, NULL);
1095 	if (IS_ERR(devr->p0)) {
1096 		ret = PTR_ERR(devr->p0);
1097 		goto error0;
1098 	}
1099 	devr->p0->device  = &dev->ib_dev;
1100 	devr->p0->uobject = NULL;
1101 	atomic_set(&devr->p0->usecnt, 0);
1102 
1103 	devr->c0 = mlx5_ib_create_cq(&dev->ib_dev, 1, 0, NULL, NULL);
1104 	if (IS_ERR(devr->c0)) {
1105 		ret = PTR_ERR(devr->c0);
1106 		goto error1;
1107 	}
1108 	devr->c0->device        = &dev->ib_dev;
1109 	devr->c0->uobject       = NULL;
1110 	devr->c0->comp_handler  = NULL;
1111 	devr->c0->event_handler = NULL;
1112 	devr->c0->cq_context    = NULL;
1113 	atomic_set(&devr->c0->usecnt, 0);
1114 
1115 	devr->x0 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
1116 	if (IS_ERR(devr->x0)) {
1117 		ret = PTR_ERR(devr->x0);
1118 		goto error2;
1119 	}
1120 	devr->x0->device = &dev->ib_dev;
1121 	devr->x0->inode = NULL;
1122 	atomic_set(&devr->x0->usecnt, 0);
1123 	mutex_init(&devr->x0->tgt_qp_mutex);
1124 	INIT_LIST_HEAD(&devr->x0->tgt_qp_list);
1125 
1126 	devr->x1 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
1127 	if (IS_ERR(devr->x1)) {
1128 		ret = PTR_ERR(devr->x1);
1129 		goto error3;
1130 	}
1131 	devr->x1->device = &dev->ib_dev;
1132 	devr->x1->inode = NULL;
1133 	atomic_set(&devr->x1->usecnt, 0);
1134 	mutex_init(&devr->x1->tgt_qp_mutex);
1135 	INIT_LIST_HEAD(&devr->x1->tgt_qp_list);
1136 
1137 	memset(&attr, 0, sizeof(attr));
1138 	attr.attr.max_sge = 1;
1139 	attr.attr.max_wr = 1;
1140 	attr.srq_type = IB_SRQT_XRC;
1141 	attr.ext.xrc.cq = devr->c0;
1142 	attr.ext.xrc.xrcd = devr->x0;
1143 
1144 	devr->s0 = mlx5_ib_create_srq(devr->p0, &attr, NULL);
1145 	if (IS_ERR(devr->s0)) {
1146 		ret = PTR_ERR(devr->s0);
1147 		goto error4;
1148 	}
1149 	devr->s0->device	= &dev->ib_dev;
1150 	devr->s0->pd		= devr->p0;
1151 	devr->s0->uobject       = NULL;
1152 	devr->s0->event_handler = NULL;
1153 	devr->s0->srq_context   = NULL;
1154 	devr->s0->srq_type      = IB_SRQT_XRC;
1155 	devr->s0->ext.xrc.xrcd	= devr->x0;
1156 	devr->s0->ext.xrc.cq	= devr->c0;
1157 	atomic_inc(&devr->s0->ext.xrc.xrcd->usecnt);
1158 	atomic_inc(&devr->s0->ext.xrc.cq->usecnt);
1159 	atomic_inc(&devr->p0->usecnt);
1160 	atomic_set(&devr->s0->usecnt, 0);
1161 
1162 	return 0;
1163 
1164 error4:
1165 	mlx5_ib_dealloc_xrcd(devr->x1);
1166 error3:
1167 	mlx5_ib_dealloc_xrcd(devr->x0);
1168 error2:
1169 	mlx5_ib_destroy_cq(devr->c0);
1170 error1:
1171 	mlx5_ib_dealloc_pd(devr->p0);
1172 error0:
1173 	return ret;
1174 }
1175 
1176 static void destroy_dev_resources(struct mlx5_ib_resources *devr)
1177 {
1178 	mlx5_ib_destroy_srq(devr->s0);
1179 	mlx5_ib_dealloc_xrcd(devr->x0);
1180 	mlx5_ib_dealloc_xrcd(devr->x1);
1181 	mlx5_ib_destroy_cq(devr->c0);
1182 	mlx5_ib_dealloc_pd(devr->p0);
1183 }
1184 
1185 static void *mlx5_ib_add(struct mlx5_core_dev *mdev)
1186 {
1187 	struct mlx5_ib_dev *dev;
1188 	int err;
1189 	int i;
1190 
1191 	printk_once(KERN_INFO "%s", mlx5_version);
1192 
1193 	dev = (struct mlx5_ib_dev *)ib_alloc_device(sizeof(*dev));
1194 	if (!dev)
1195 		return NULL;
1196 
1197 	dev->mdev = mdev;
1198 
1199 	err = get_port_caps(dev);
1200 	if (err)
1201 		goto err_dealloc;
1202 
1203 	get_ext_port_caps(dev);
1204 
1205 	MLX5_INIT_DOORBELL_LOCK(&dev->uar_lock);
1206 
1207 	strlcpy(dev->ib_dev.name, "mlx5_%d", IB_DEVICE_NAME_MAX);
1208 	dev->ib_dev.owner		= THIS_MODULE;
1209 	dev->ib_dev.node_type		= RDMA_NODE_IB_CA;
1210 	dev->ib_dev.local_dma_lkey	= mdev->caps.gen.reserved_lkey;
1211 	dev->num_ports		= mdev->caps.gen.num_ports;
1212 	dev->ib_dev.phys_port_cnt     = dev->num_ports;
1213 	dev->ib_dev.num_comp_vectors    =
1214 		dev->mdev->priv.eq_table.num_comp_vectors;
1215 	dev->ib_dev.dma_device	= &mdev->pdev->dev;
1216 
1217 	dev->ib_dev.uverbs_abi_ver	= MLX5_IB_UVERBS_ABI_VERSION;
1218 	dev->ib_dev.uverbs_cmd_mask	=
1219 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1220 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1221 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1222 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
1223 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
1224 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
1225 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
1226 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
1227 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
1228 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
1229 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
1230 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
1231 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
1232 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
1233 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
1234 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
1235 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
1236 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
1237 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
1238 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
1239 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
1240 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
1241 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
1242 	dev->ib_dev.uverbs_ex_cmd_mask =
1243 		(1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE);
1244 
1245 	dev->ib_dev.query_device	= mlx5_ib_query_device;
1246 	dev->ib_dev.query_port		= mlx5_ib_query_port;
1247 	dev->ib_dev.query_gid		= mlx5_ib_query_gid;
1248 	dev->ib_dev.query_pkey		= mlx5_ib_query_pkey;
1249 	dev->ib_dev.modify_device	= mlx5_ib_modify_device;
1250 	dev->ib_dev.modify_port		= mlx5_ib_modify_port;
1251 	dev->ib_dev.alloc_ucontext	= mlx5_ib_alloc_ucontext;
1252 	dev->ib_dev.dealloc_ucontext	= mlx5_ib_dealloc_ucontext;
1253 	dev->ib_dev.mmap		= mlx5_ib_mmap;
1254 	dev->ib_dev.alloc_pd		= mlx5_ib_alloc_pd;
1255 	dev->ib_dev.dealloc_pd		= mlx5_ib_dealloc_pd;
1256 	dev->ib_dev.create_ah		= mlx5_ib_create_ah;
1257 	dev->ib_dev.query_ah		= mlx5_ib_query_ah;
1258 	dev->ib_dev.destroy_ah		= mlx5_ib_destroy_ah;
1259 	dev->ib_dev.create_srq		= mlx5_ib_create_srq;
1260 	dev->ib_dev.modify_srq		= mlx5_ib_modify_srq;
1261 	dev->ib_dev.query_srq		= mlx5_ib_query_srq;
1262 	dev->ib_dev.destroy_srq		= mlx5_ib_destroy_srq;
1263 	dev->ib_dev.post_srq_recv	= mlx5_ib_post_srq_recv;
1264 	dev->ib_dev.create_qp		= mlx5_ib_create_qp;
1265 	dev->ib_dev.modify_qp		= mlx5_ib_modify_qp;
1266 	dev->ib_dev.query_qp		= mlx5_ib_query_qp;
1267 	dev->ib_dev.destroy_qp		= mlx5_ib_destroy_qp;
1268 	dev->ib_dev.post_send		= mlx5_ib_post_send;
1269 	dev->ib_dev.post_recv		= mlx5_ib_post_recv;
1270 	dev->ib_dev.create_cq		= mlx5_ib_create_cq;
1271 	dev->ib_dev.modify_cq		= mlx5_ib_modify_cq;
1272 	dev->ib_dev.resize_cq		= mlx5_ib_resize_cq;
1273 	dev->ib_dev.destroy_cq		= mlx5_ib_destroy_cq;
1274 	dev->ib_dev.poll_cq		= mlx5_ib_poll_cq;
1275 	dev->ib_dev.req_notify_cq	= mlx5_ib_arm_cq;
1276 	dev->ib_dev.get_dma_mr		= mlx5_ib_get_dma_mr;
1277 	dev->ib_dev.reg_user_mr		= mlx5_ib_reg_user_mr;
1278 	dev->ib_dev.dereg_mr		= mlx5_ib_dereg_mr;
1279 	dev->ib_dev.destroy_mr		= mlx5_ib_destroy_mr;
1280 	dev->ib_dev.attach_mcast	= mlx5_ib_mcg_attach;
1281 	dev->ib_dev.detach_mcast	= mlx5_ib_mcg_detach;
1282 	dev->ib_dev.process_mad		= mlx5_ib_process_mad;
1283 	dev->ib_dev.create_mr		= mlx5_ib_create_mr;
1284 	dev->ib_dev.alloc_fast_reg_mr	= mlx5_ib_alloc_fast_reg_mr;
1285 	dev->ib_dev.alloc_fast_reg_page_list = mlx5_ib_alloc_fast_reg_page_list;
1286 	dev->ib_dev.free_fast_reg_page_list  = mlx5_ib_free_fast_reg_page_list;
1287 	dev->ib_dev.check_mr_status	= mlx5_ib_check_mr_status;
1288 
1289 	mlx5_ib_internal_query_odp_caps(dev);
1290 
1291 	if (mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_XRC) {
1292 		dev->ib_dev.alloc_xrcd = mlx5_ib_alloc_xrcd;
1293 		dev->ib_dev.dealloc_xrcd = mlx5_ib_dealloc_xrcd;
1294 		dev->ib_dev.uverbs_cmd_mask |=
1295 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1296 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1297 	}
1298 
1299 	err = init_node_data(dev);
1300 	if (err)
1301 		goto err_dealloc;
1302 
1303 	mutex_init(&dev->cap_mask_mutex);
1304 
1305 	err = create_dev_resources(&dev->devr);
1306 	if (err)
1307 		goto err_dealloc;
1308 
1309 	err = mlx5_ib_odp_init_one(dev);
1310 	if (err)
1311 		goto err_rsrc;
1312 
1313 	err = ib_register_device(&dev->ib_dev, NULL);
1314 	if (err)
1315 		goto err_odp;
1316 
1317 	err = create_umr_res(dev);
1318 	if (err)
1319 		goto err_dev;
1320 
1321 	for (i = 0; i < ARRAY_SIZE(mlx5_class_attributes); i++) {
1322 		err = device_create_file(&dev->ib_dev.dev,
1323 					 mlx5_class_attributes[i]);
1324 		if (err)
1325 			goto err_umrc;
1326 	}
1327 
1328 	dev->ib_active = true;
1329 
1330 	return dev;
1331 
1332 err_umrc:
1333 	destroy_umrc_res(dev);
1334 
1335 err_dev:
1336 	ib_unregister_device(&dev->ib_dev);
1337 
1338 err_odp:
1339 	mlx5_ib_odp_remove_one(dev);
1340 
1341 err_rsrc:
1342 	destroy_dev_resources(&dev->devr);
1343 
1344 err_dealloc:
1345 	ib_dealloc_device((struct ib_device *)dev);
1346 
1347 	return NULL;
1348 }
1349 
1350 static void mlx5_ib_remove(struct mlx5_core_dev *mdev, void *context)
1351 {
1352 	struct mlx5_ib_dev *dev = context;
1353 
1354 	ib_unregister_device(&dev->ib_dev);
1355 	destroy_umrc_res(dev);
1356 	mlx5_ib_odp_remove_one(dev);
1357 	destroy_dev_resources(&dev->devr);
1358 	ib_dealloc_device(&dev->ib_dev);
1359 }
1360 
1361 static struct mlx5_interface mlx5_ib_interface = {
1362 	.add            = mlx5_ib_add,
1363 	.remove         = mlx5_ib_remove,
1364 	.event          = mlx5_ib_event,
1365 	.protocol	= MLX5_INTERFACE_PROTOCOL_IB,
1366 };
1367 
1368 static int __init mlx5_ib_init(void)
1369 {
1370 	int err;
1371 
1372 	if (deprecated_prof_sel != 2)
1373 		pr_warn("prof_sel is deprecated for mlx5_ib, set it for mlx5_core\n");
1374 
1375 	err = mlx5_ib_odp_init();
1376 	if (err)
1377 		return err;
1378 
1379 	err = mlx5_register_interface(&mlx5_ib_interface);
1380 	if (err)
1381 		goto clean_odp;
1382 
1383 	return err;
1384 
1385 clean_odp:
1386 	mlx5_ib_odp_cleanup();
1387 	return err;
1388 }
1389 
1390 static void __exit mlx5_ib_cleanup(void)
1391 {
1392 	mlx5_unregister_interface(&mlx5_ib_interface);
1393 	mlx5_ib_odp_cleanup();
1394 }
1395 
1396 module_init(mlx5_ib_init);
1397 module_exit(mlx5_ib_cleanup);
1398