xref: /freebsd-src/sys/dev/mlx5/mlx5_core/mlx5_main.c (revision 3311ff84eac3b7e82f28e331df0586036c6d361c)
1 /*-
2  * Copyright (c) 2013-2015, Mellanox Technologies, Ltd.  All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  *
13  * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND
14  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16  * ARE DISCLAIMED.  IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE
17  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
23  * SUCH DAMAGE.
24  *
25  * $FreeBSD$
26  */
27 
28 #include <linux/kmod.h>
29 #include <linux/module.h>
30 #include <linux/errno.h>
31 #include <linux/pci.h>
32 #include <linux/dma-mapping.h>
33 #include <linux/slab.h>
34 #include <linux/io-mapping.h>
35 #include <linux/interrupt.h>
36 #include <dev/mlx5/driver.h>
37 #include <dev/mlx5/cq.h>
38 #include <dev/mlx5/qp.h>
39 #include <dev/mlx5/srq.h>
40 #include <linux/delay.h>
41 #include <dev/mlx5/mlx5_ifc.h>
42 #include "mlx5_core.h"
43 
44 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
45 MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
46 MODULE_LICENSE("Dual BSD/GPL");
47 #if (__FreeBSD_version >= 1100000)
48 MODULE_DEPEND(mlx5, linuxkpi, 1, 1, 1);
49 #endif
50 MODULE_VERSION(mlx5, 1);
51 
52 int mlx5_core_debug_mask;
53 module_param_named(debug_mask, mlx5_core_debug_mask, int, 0644);
54 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
55 
56 #define MLX5_DEFAULT_PROF	2
57 static int prof_sel = MLX5_DEFAULT_PROF;
58 module_param_named(prof_sel, prof_sel, int, 0444);
59 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
60 
61 #define NUMA_NO_NODE       -1
62 
63 struct workqueue_struct *mlx5_core_wq;
64 static LIST_HEAD(intf_list);
65 static LIST_HEAD(dev_list);
66 static DEFINE_MUTEX(intf_mutex);
67 
68 struct mlx5_device_context {
69 	struct list_head	list;
70 	struct mlx5_interface  *intf;
71 	void		       *context;
72 };
73 
74 static struct mlx5_profile profiles[] = {
75 	[0] = {
76 		.mask           = 0,
77 	},
78 	[1] = {
79 		.mask		= MLX5_PROF_MASK_QP_SIZE,
80 		.log_max_qp	= 12,
81 	},
82 	[2] = {
83 		.mask		= MLX5_PROF_MASK_QP_SIZE |
84 				  MLX5_PROF_MASK_MR_CACHE,
85 		.log_max_qp	= 17,
86 		.mr_cache[0]	= {
87 			.size	= 500,
88 			.limit	= 250
89 		},
90 		.mr_cache[1]	= {
91 			.size	= 500,
92 			.limit	= 250
93 		},
94 		.mr_cache[2]	= {
95 			.size	= 500,
96 			.limit	= 250
97 		},
98 		.mr_cache[3]	= {
99 			.size	= 500,
100 			.limit	= 250
101 		},
102 		.mr_cache[4]	= {
103 			.size	= 500,
104 			.limit	= 250
105 		},
106 		.mr_cache[5]	= {
107 			.size	= 500,
108 			.limit	= 250
109 		},
110 		.mr_cache[6]	= {
111 			.size	= 500,
112 			.limit	= 250
113 		},
114 		.mr_cache[7]	= {
115 			.size	= 500,
116 			.limit	= 250
117 		},
118 		.mr_cache[8]	= {
119 			.size	= 500,
120 			.limit	= 250
121 		},
122 		.mr_cache[9]	= {
123 			.size	= 500,
124 			.limit	= 250
125 		},
126 		.mr_cache[10]	= {
127 			.size	= 500,
128 			.limit	= 250
129 		},
130 		.mr_cache[11]	= {
131 			.size	= 500,
132 			.limit	= 250
133 		},
134 		.mr_cache[12]	= {
135 			.size	= 64,
136 			.limit	= 32
137 		},
138 		.mr_cache[13]	= {
139 			.size	= 32,
140 			.limit	= 16
141 		},
142 		.mr_cache[14]	= {
143 			.size	= 16,
144 			.limit	= 8
145 		},
146 		.mr_cache[15]	= {
147 			.size	= 8,
148 			.limit	= 4
149 		},
150 	},
151 	[3] = {
152 		.mask		= MLX5_PROF_MASK_QP_SIZE,
153 		.log_max_qp	= 17,
154 	},
155 };
156 
157 static int set_dma_caps(struct pci_dev *pdev)
158 {
159 	int err;
160 
161 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
162 	if (err) {
163 		device_printf((&pdev->dev)->bsddev, "WARN: ""Warning: couldn't set 64-bit PCI DMA mask\n");
164 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
165 		if (err) {
166 			device_printf((&pdev->dev)->bsddev, "ERR: ""Can't set PCI DMA mask, aborting\n");
167 			return err;
168 		}
169 	}
170 
171 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
172 	if (err) {
173 		device_printf((&pdev->dev)->bsddev, "WARN: ""Warning: couldn't set 64-bit consistent PCI DMA mask\n");
174 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
175 		if (err) {
176 			device_printf((&pdev->dev)->bsddev, "ERR: ""Can't set consistent PCI DMA mask, aborting\n");
177 			return err;
178 		}
179 	}
180 
181 	dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
182 	return err;
183 }
184 
185 static int request_bar(struct pci_dev *pdev)
186 {
187 	int err = 0;
188 
189 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
190 		device_printf((&pdev->dev)->bsddev, "ERR: ""Missing registers BAR, aborting\n");
191 		return -ENODEV;
192 	}
193 
194 	err = pci_request_regions(pdev, DRIVER_NAME);
195 	if (err)
196 		device_printf((&pdev->dev)->bsddev, "ERR: ""Couldn't get PCI resources, aborting\n");
197 
198 	return err;
199 }
200 
201 static void release_bar(struct pci_dev *pdev)
202 {
203 	pci_release_regions(pdev);
204 }
205 
206 static int mlx5_enable_msix(struct mlx5_core_dev *dev)
207 {
208 	struct mlx5_priv *priv = &dev->priv;
209 	struct mlx5_eq_table *table = &priv->eq_table;
210 	int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq);
211 	int nvec;
212 	int i;
213 
214 	nvec = MLX5_CAP_GEN(dev, num_ports) * num_online_cpus() +
215 	       MLX5_EQ_VEC_COMP_BASE;
216 	nvec = min_t(int, nvec, num_eqs);
217 	if (nvec <= MLX5_EQ_VEC_COMP_BASE)
218 		return -ENOMEM;
219 
220 	priv->msix_arr = kzalloc(nvec * sizeof(*priv->msix_arr), GFP_KERNEL);
221 
222 	priv->irq_info = kzalloc(nvec * sizeof(*priv->irq_info), GFP_KERNEL);
223 
224 	for (i = 0; i < nvec; i++)
225 		priv->msix_arr[i].entry = i;
226 
227 	nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr,
228 				     MLX5_EQ_VEC_COMP_BASE + 1, nvec);
229 	if (nvec < 0)
230 		return nvec;
231 
232 	table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
233 
234 	return 0;
235 
236 }
237 
238 static void mlx5_disable_msix(struct mlx5_core_dev *dev)
239 {
240 	struct mlx5_priv *priv = &dev->priv;
241 
242 	pci_disable_msix(dev->pdev);
243 	kfree(priv->irq_info);
244 	kfree(priv->msix_arr);
245 }
246 
247 struct mlx5_reg_host_endianess {
248 	u8	he;
249 	u8      rsvd[15];
250 };
251 
252 
253 #define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos))
254 
255 enum {
256 	MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) |
257 				MLX5_DEV_CAP_FLAG_DCT,
258 };
259 
260 static u16 to_fw_pkey_sz(u32 size)
261 {
262 	switch (size) {
263 	case 128:
264 		return 0;
265 	case 256:
266 		return 1;
267 	case 512:
268 		return 2;
269 	case 1024:
270 		return 3;
271 	case 2048:
272 		return 4;
273 	case 4096:
274 		return 5;
275 	default:
276 		printf("mlx5_core: WARN: ""invalid pkey table size %d\n", size);
277 		return 0;
278 	}
279 }
280 
281 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type,
282 		       enum mlx5_cap_mode cap_mode)
283 {
284 	u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
285 	int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
286 	void *out, *hca_caps;
287 	u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
288 	int err;
289 
290 	memset(in, 0, sizeof(in));
291 	out = kzalloc(out_sz, GFP_KERNEL);
292 
293 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
294 	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
295 	err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz);
296 	if (err)
297 		goto query_ex;
298 
299 	err = mlx5_cmd_status_to_err_v2(out);
300 	if (err) {
301 		mlx5_core_warn(dev,
302 			       "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
303 			       cap_type, cap_mode, err);
304 		goto query_ex;
305 	}
306 
307 	hca_caps =  MLX5_ADDR_OF(query_hca_cap_out, out, capability);
308 
309 	switch (cap_mode) {
310 	case HCA_CAP_OPMOD_GET_MAX:
311 		memcpy(dev->hca_caps_max[cap_type], hca_caps,
312 		       MLX5_UN_SZ_BYTES(hca_cap_union));
313 		break;
314 	case HCA_CAP_OPMOD_GET_CUR:
315 		memcpy(dev->hca_caps_cur[cap_type], hca_caps,
316 		       MLX5_UN_SZ_BYTES(hca_cap_union));
317 		break;
318 	default:
319 		mlx5_core_warn(dev,
320 			       "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
321 			       cap_type, cap_mode);
322 		err = -EINVAL;
323 		break;
324 	}
325 query_ex:
326 	kfree(out);
327 	return err;
328 }
329 
330 static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz)
331 {
332 	u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)];
333 	int err;
334 
335 	memset(out, 0, sizeof(out));
336 
337 	MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
338 	err = mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out));
339 	if (err)
340 		return err;
341 
342 	err = mlx5_cmd_status_to_err_v2(out);
343 
344 	return err;
345 }
346 
347 static int handle_hca_cap(struct mlx5_core_dev *dev)
348 {
349 	void *set_ctx = NULL;
350 	struct mlx5_profile *prof = dev->profile;
351 	int err = -ENOMEM;
352 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
353 	void *set_hca_cap;
354 
355 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
356 
357 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL, HCA_CAP_OPMOD_GET_MAX);
358 	if (err)
359 		goto query_ex;
360 
361 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL, HCA_CAP_OPMOD_GET_CUR);
362 	if (err)
363 		goto query_ex;
364 
365 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
366 				   capability);
367 	memcpy(set_hca_cap, dev->hca_caps_cur[MLX5_CAP_GENERAL],
368 	       MLX5_ST_SZ_BYTES(cmd_hca_cap));
369 
370 	mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
371 		      mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
372 		      128);
373 	/* we limit the size of the pkey table to 128 entries for now */
374 	MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
375 		 to_fw_pkey_sz(128));
376 
377 	if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
378 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
379 			 prof->log_max_qp);
380 
381 	/* disable cmdif checksum */
382 	MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
383 
384 	MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
385 
386 	err = set_caps(dev, set_ctx, set_sz);
387 
388 query_ex:
389 	kfree(set_ctx);
390 	return err;
391 }
392 
393 static int set_hca_ctrl(struct mlx5_core_dev *dev)
394 {
395 	struct mlx5_reg_host_endianess he_in;
396 	struct mlx5_reg_host_endianess he_out;
397 	int err;
398 
399 	memset(&he_in, 0, sizeof(he_in));
400 	he_in.he = MLX5_SET_HOST_ENDIANNESS;
401 	err = mlx5_core_access_reg(dev, &he_in,  sizeof(he_in),
402 					&he_out, sizeof(he_out),
403 					MLX5_REG_HOST_ENDIANNESS, 0, 1);
404 	return err;
405 }
406 
407 static int mlx5_core_enable_hca(struct mlx5_core_dev *dev)
408 {
409 	u32 in[MLX5_ST_SZ_DW(enable_hca_in)];
410 	u32 out[MLX5_ST_SZ_DW(enable_hca_out)];
411 
412 	memset(in, 0, sizeof(in));
413 	MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
414 	memset(out, 0, sizeof(out));
415 	return mlx5_cmd_exec_check_status(dev, in,  sizeof(in),
416 					       out, sizeof(out));
417 }
418 
419 static int mlx5_core_disable_hca(struct mlx5_core_dev *dev)
420 {
421 	u32 in[MLX5_ST_SZ_DW(disable_hca_in)];
422 	u32 out[MLX5_ST_SZ_DW(disable_hca_out)];
423 
424 	memset(in, 0, sizeof(in));
425 
426 	MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
427 	memset(out, 0, sizeof(out));
428 	return mlx5_cmd_exec_check_status(dev, in,  sizeof(in),
429 					       out, sizeof(out));
430 }
431 
432 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
433 {
434 	u32 query_in[MLX5_ST_SZ_DW(query_issi_in)];
435 	u32 query_out[MLX5_ST_SZ_DW(query_issi_out)];
436 	u32 set_in[MLX5_ST_SZ_DW(set_issi_in)];
437 	u32 set_out[MLX5_ST_SZ_DW(set_issi_out)];
438 	int err;
439 	u32 sup_issi;
440 
441 	memset(query_in, 0, sizeof(query_in));
442 	memset(query_out, 0, sizeof(query_out));
443 
444 	MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
445 
446 	err = mlx5_cmd_exec_check_status(dev, query_in, sizeof(query_in),
447 					 query_out, sizeof(query_out));
448 	if (err) {
449 		if (((struct mlx5_outbox_hdr *)query_out)->status ==
450 		    MLX5_CMD_STAT_BAD_OP_ERR) {
451 			pr_debug("Only ISSI 0 is supported\n");
452 			return 0;
453 		}
454 
455 		printf("mlx5_core: ERR: ""failed to query ISSI\n");
456 		return err;
457 	}
458 
459 	sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
460 
461 	if (sup_issi & (1 << 1)) {
462 		memset(set_in, 0, sizeof(set_in));
463 		memset(set_out, 0, sizeof(set_out));
464 
465 		MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
466 		MLX5_SET(set_issi_in, set_in, current_issi, 1);
467 
468 		err = mlx5_cmd_exec_check_status(dev, set_in, sizeof(set_in),
469 						 set_out, sizeof(set_out));
470 		if (err) {
471 			printf("mlx5_core: ERR: ""failed to set ISSI=1\n");
472 			return err;
473 		}
474 
475 		dev->issi = 1;
476 
477 		return 0;
478 	} else if (sup_issi & (1 << 0)) {
479 		return 0;
480 	}
481 
482 	return -ENOTSUPP;
483 }
484 
485 
486 int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn, int *irqn)
487 {
488 	struct mlx5_eq_table *table = &dev->priv.eq_table;
489 	struct mlx5_eq *eq;
490 	int err = -ENOENT;
491 
492 	spin_lock(&table->lock);
493 	list_for_each_entry(eq, &table->comp_eqs_list, list) {
494 		if (eq->index == vector) {
495 			*eqn = eq->eqn;
496 			*irqn = eq->irqn;
497 			err = 0;
498 			break;
499 		}
500 	}
501 	spin_unlock(&table->lock);
502 
503 	return err;
504 }
505 EXPORT_SYMBOL(mlx5_vector2eqn);
506 
507 int mlx5_rename_eq(struct mlx5_core_dev *dev, int eq_ix, char *name)
508 {
509 	struct mlx5_priv *priv = &dev->priv;
510 	struct mlx5_eq_table *table = &priv->eq_table;
511 	struct mlx5_eq *eq;
512 	int err = -ENOENT;
513 
514 	spin_lock(&table->lock);
515 	list_for_each_entry(eq, &table->comp_eqs_list, list) {
516 		if (eq->index == eq_ix) {
517 			int irq_ix = eq_ix + MLX5_EQ_VEC_COMP_BASE;
518 
519 			snprintf(priv->irq_info[irq_ix].name, MLX5_MAX_IRQ_NAME,
520 				 "%s-%d", name, eq_ix);
521 
522 			err = 0;
523 			break;
524 		}
525 	}
526 	spin_unlock(&table->lock);
527 
528 	return err;
529 }
530 
531 static void free_comp_eqs(struct mlx5_core_dev *dev)
532 {
533 	struct mlx5_eq_table *table = &dev->priv.eq_table;
534 	struct mlx5_eq *eq, *n;
535 
536 	spin_lock(&table->lock);
537 	list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
538 		list_del(&eq->list);
539 		spin_unlock(&table->lock);
540 		if (mlx5_destroy_unmap_eq(dev, eq))
541 			mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n",
542 				       eq->eqn);
543 		kfree(eq);
544 		spin_lock(&table->lock);
545 	}
546 	spin_unlock(&table->lock);
547 }
548 
549 static int alloc_comp_eqs(struct mlx5_core_dev *dev)
550 {
551 	struct mlx5_eq_table *table = &dev->priv.eq_table;
552 	char name[MLX5_MAX_IRQ_NAME];
553 	struct mlx5_eq *eq;
554 	int ncomp_vec;
555 	int nent;
556 	int err;
557 	int i;
558 
559 	INIT_LIST_HEAD(&table->comp_eqs_list);
560 	ncomp_vec = table->num_comp_vectors;
561 	nent = MLX5_COMP_EQ_SIZE;
562 	for (i = 0; i < ncomp_vec; i++) {
563 		eq = kzalloc(sizeof(*eq), GFP_KERNEL);
564 
565 		snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i);
566 		err = mlx5_create_map_eq(dev, eq,
567 					 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
568 					 name, &dev->priv.uuari.uars[0]);
569 		if (err) {
570 			kfree(eq);
571 			goto clean;
572 		}
573 		mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
574 		eq->index = i;
575 		spin_lock(&table->lock);
576 		list_add_tail(&eq->list, &table->comp_eqs_list);
577 		spin_unlock(&table->lock);
578 	}
579 
580 	return 0;
581 
582 clean:
583 	free_comp_eqs(dev);
584 	return err;
585 }
586 
587 static int map_bf_area(struct mlx5_core_dev *dev)
588 {
589 	resource_size_t bf_start = pci_resource_start(dev->pdev, 0);
590 	resource_size_t bf_len = pci_resource_len(dev->pdev, 0);
591 
592 	dev->priv.bf_mapping = io_mapping_create_wc(bf_start, bf_len);
593 
594 	return dev->priv.bf_mapping ? 0 : -ENOMEM;
595 }
596 
597 static void unmap_bf_area(struct mlx5_core_dev *dev)
598 {
599 	if (dev->priv.bf_mapping)
600 		io_mapping_free(dev->priv.bf_mapping);
601 }
602 
603 static inline int fw_initializing(struct mlx5_core_dev *dev)
604 {
605 	return ioread32be(&dev->iseg->initializing) >> 31;
606 }
607 
608 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili)
609 {
610 	u64 end = jiffies + msecs_to_jiffies(max_wait_mili);
611 	int err = 0;
612 
613 	while (fw_initializing(dev)) {
614 		if (time_after(jiffies, end)) {
615 			err = -EBUSY;
616 			break;
617 		}
618 		msleep(FW_INIT_WAIT_MS);
619 	}
620 
621 	return err;
622 }
623 
624 static int mlx5_dev_init(struct mlx5_core_dev *dev, struct pci_dev *pdev)
625 {
626 	struct mlx5_priv *priv = &dev->priv;
627 	int err;
628 
629 	dev->pdev = pdev;
630 	pci_set_drvdata(dev->pdev, dev);
631 	strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
632 	priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
633 
634 	mutex_init(&priv->pgdir_mutex);
635 	INIT_LIST_HEAD(&priv->pgdir_list);
636 	spin_lock_init(&priv->mkey_lock);
637 
638 	priv->numa_node = NUMA_NO_NODE;
639 
640 	err = pci_enable_device(pdev);
641 	if (err) {
642 		device_printf((&pdev->dev)->bsddev, "ERR: ""Cannot enable PCI device, aborting\n");
643 		goto err_dbg;
644 	}
645 
646 	err = request_bar(pdev);
647 	if (err) {
648 		device_printf((&pdev->dev)->bsddev, "ERR: ""error requesting BARs, aborting\n");
649 		goto err_disable;
650 	}
651 
652 	pci_set_master(pdev);
653 
654 	err = set_dma_caps(pdev);
655 	if (err) {
656 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed setting DMA capabilities mask, aborting\n");
657 		goto err_clr_master;
658 	}
659 
660 	dev->iseg = ioremap(pci_resource_start(dev->pdev, 0),
661 			    sizeof(*dev->iseg));
662 	if (!dev->iseg) {
663 		err = -ENOMEM;
664 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed mapping initialization segment, aborting\n");
665 		goto err_clr_master;
666 	}
667 	device_printf((&pdev->dev)->bsddev, "INFO: ""firmware version: %d.%d.%d\n", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev));
668 
669 	err = mlx5_cmd_init(dev);
670 	if (err) {
671 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed initializing command interface, aborting\n");
672 		goto err_unmap;
673 	}
674 
675 	err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI);
676 	if (err) {
677 		device_printf((&dev->pdev->dev)->bsddev, "ERR: ""Firmware over %d MS in initializing state, aborting\n", FW_INIT_TIMEOUT_MILI);
678 		goto err_cmd_cleanup;
679 	}
680 
681 	mlx5_pagealloc_init(dev);
682 
683 	err = mlx5_core_enable_hca(dev);
684 	if (err) {
685 		device_printf((&pdev->dev)->bsddev, "ERR: ""enable hca failed\n");
686 		goto err_pagealloc_cleanup;
687 	}
688 
689 	err = mlx5_core_set_issi(dev);
690 	if (err) {
691 		device_printf((&pdev->dev)->bsddev, "ERR: ""failed to set issi\n");
692 		goto err_disable_hca;
693 	}
694 
695 	err = mlx5_pagealloc_start(dev);
696 	if (err) {
697 		device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_pagealloc_start failed\n");
698 		goto err_disable_hca;
699 	}
700 
701 	err = mlx5_satisfy_startup_pages(dev, 1);
702 	if (err) {
703 		device_printf((&pdev->dev)->bsddev, "ERR: ""failed to allocate boot pages\n");
704 		goto err_pagealloc_stop;
705 	}
706 
707 	err = set_hca_ctrl(dev);
708 	if (err) {
709 		device_printf((&pdev->dev)->bsddev, "ERR: ""set_hca_ctrl failed\n");
710 		goto reclaim_boot_pages;
711 	}
712 
713 	err = handle_hca_cap(dev);
714 	if (err) {
715 		device_printf((&pdev->dev)->bsddev, "ERR: ""handle_hca_cap failed\n");
716 		goto reclaim_boot_pages;
717 	}
718 
719 	err = mlx5_satisfy_startup_pages(dev, 0);
720 	if (err) {
721 		device_printf((&pdev->dev)->bsddev, "ERR: ""failed to allocate init pages\n");
722 		goto reclaim_boot_pages;
723 	}
724 
725 	err = mlx5_cmd_init_hca(dev);
726 	if (err) {
727 		device_printf((&pdev->dev)->bsddev, "ERR: ""init hca failed\n");
728 		goto reclaim_boot_pages;
729 	}
730 
731 	mlx5_start_health_poll(dev);
732 
733 	err = mlx5_query_hca_caps(dev);
734 	if (err) {
735 		device_printf((&pdev->dev)->bsddev, "ERR: ""query hca failed\n");
736 		goto err_stop_poll;
737 	}
738 
739 	err = mlx5_query_board_id(dev);
740 	if (err) {
741 		device_printf((&pdev->dev)->bsddev, "ERR: ""query board id failed\n");
742 		goto err_stop_poll;
743 	}
744 
745 	err = mlx5_enable_msix(dev);
746 	if (err) {
747 		device_printf((&pdev->dev)->bsddev, "ERR: ""enable msix failed\n");
748 		goto err_stop_poll;
749 	}
750 
751 	err = mlx5_eq_init(dev);
752 	if (err) {
753 		device_printf((&pdev->dev)->bsddev, "ERR: ""failed to initialize eq\n");
754 		goto disable_msix;
755 	}
756 
757 	err = mlx5_alloc_uuars(dev, &priv->uuari);
758 	if (err) {
759 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed allocating uar, aborting\n");
760 		goto err_eq_cleanup;
761 	}
762 
763 	err = mlx5_start_eqs(dev);
764 	if (err) {
765 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to start pages and async EQs\n");
766 		goto err_free_uar;
767 	}
768 
769 	err = alloc_comp_eqs(dev);
770 	if (err) {
771 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to alloc completion EQs\n");
772 		goto err_stop_eqs;
773 	}
774 
775 	if (map_bf_area(dev))
776 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to map blue flame area\n");
777 
778 	MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock);
779 
780 	mlx5_init_cq_table(dev);
781 	mlx5_init_qp_table(dev);
782 	mlx5_init_srq_table(dev);
783 	mlx5_init_mr_table(dev);
784 
785 	return 0;
786 
787 err_stop_eqs:
788 	mlx5_stop_eqs(dev);
789 
790 err_free_uar:
791 	mlx5_free_uuars(dev, &priv->uuari);
792 
793 err_eq_cleanup:
794 	mlx5_eq_cleanup(dev);
795 
796 disable_msix:
797 	mlx5_disable_msix(dev);
798 
799 err_stop_poll:
800 	mlx5_stop_health_poll(dev);
801 	if (mlx5_cmd_teardown_hca(dev)) {
802 		device_printf((&dev->pdev->dev)->bsddev, "ERR: ""tear_down_hca failed, skip cleanup\n");
803 		return err;
804 	}
805 
806 reclaim_boot_pages:
807 	mlx5_reclaim_startup_pages(dev);
808 
809 err_pagealloc_stop:
810 	mlx5_pagealloc_stop(dev);
811 
812 err_disable_hca:
813 	mlx5_core_disable_hca(dev);
814 
815 err_pagealloc_cleanup:
816 	mlx5_pagealloc_cleanup(dev);
817 err_cmd_cleanup:
818 	mlx5_cmd_cleanup(dev);
819 
820 err_unmap:
821 	iounmap(dev->iseg);
822 
823 err_clr_master:
824 	pci_clear_master(dev->pdev);
825 	release_bar(dev->pdev);
826 
827 err_disable:
828 	pci_disable_device(dev->pdev);
829 
830 err_dbg:
831 	return err;
832 }
833 
834 static void mlx5_dev_cleanup(struct mlx5_core_dev *dev)
835 {
836 	struct mlx5_priv *priv = &dev->priv;
837 
838 	mlx5_cleanup_mr_table(dev);
839 	mlx5_cleanup_srq_table(dev);
840 	mlx5_cleanup_qp_table(dev);
841 	mlx5_cleanup_cq_table(dev);
842 	unmap_bf_area(dev);
843 	free_comp_eqs(dev);
844 	mlx5_stop_eqs(dev);
845 	mlx5_free_uuars(dev, &priv->uuari);
846 	mlx5_eq_cleanup(dev);
847 	mlx5_disable_msix(dev);
848 	mlx5_stop_health_poll(dev);
849 	if (mlx5_cmd_teardown_hca(dev)) {
850 		device_printf((&dev->pdev->dev)->bsddev, "ERR: ""tear_down_hca failed, skip cleanup\n");
851 		return;
852 	}
853 	mlx5_pagealloc_stop(dev);
854 	mlx5_reclaim_startup_pages(dev);
855 	mlx5_core_disable_hca(dev);
856 	mlx5_pagealloc_cleanup(dev);
857 	mlx5_cmd_cleanup(dev);
858 	iounmap(dev->iseg);
859 	pci_clear_master(dev->pdev);
860 	release_bar(dev->pdev);
861 	pci_disable_device(dev->pdev);
862 }
863 
864 static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
865 {
866 	struct mlx5_device_context *dev_ctx;
867 	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
868 
869 	dev_ctx = kmalloc(sizeof(*dev_ctx), GFP_KERNEL);
870 
871 	dev_ctx->intf    = intf;
872 	dev_ctx->context = intf->add(dev);
873 
874 	if (dev_ctx->context) {
875 		spin_lock_irq(&priv->ctx_lock);
876 		list_add_tail(&dev_ctx->list, &priv->ctx_list);
877 		spin_unlock_irq(&priv->ctx_lock);
878 	} else {
879 		kfree(dev_ctx);
880 	}
881 }
882 
883 static void mlx5_remove_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
884 {
885 	struct mlx5_device_context *dev_ctx;
886 	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
887 
888 	list_for_each_entry(dev_ctx, &priv->ctx_list, list)
889 		if (dev_ctx->intf == intf) {
890 			spin_lock_irq(&priv->ctx_lock);
891 			list_del(&dev_ctx->list);
892 			spin_unlock_irq(&priv->ctx_lock);
893 
894 			intf->remove(dev, dev_ctx->context);
895 			kfree(dev_ctx);
896 			return;
897 		}
898 }
899 static int mlx5_register_device(struct mlx5_core_dev *dev)
900 {
901 	struct mlx5_priv *priv = &dev->priv;
902 	struct mlx5_interface *intf;
903 
904 	mutex_lock(&intf_mutex);
905 	list_add_tail(&priv->dev_list, &dev_list);
906 	list_for_each_entry(intf, &intf_list, list)
907 		mlx5_add_device(intf, priv);
908 	mutex_unlock(&intf_mutex);
909 
910 	return 0;
911 }
912 static void mlx5_unregister_device(struct mlx5_core_dev *dev)
913 {
914 	struct mlx5_priv *priv = &dev->priv;
915 	struct mlx5_interface *intf;
916 
917 	mutex_lock(&intf_mutex);
918 	list_for_each_entry(intf, &intf_list, list)
919 		mlx5_remove_device(intf, priv);
920 	list_del(&priv->dev_list);
921 	mutex_unlock(&intf_mutex);
922 }
923 
924 int mlx5_register_interface(struct mlx5_interface *intf)
925 {
926 	struct mlx5_priv *priv;
927 
928 	if (!intf->add || !intf->remove)
929 		return -EINVAL;
930 
931 	mutex_lock(&intf_mutex);
932 	list_add_tail(&intf->list, &intf_list);
933 	list_for_each_entry(priv, &dev_list, dev_list)
934 		mlx5_add_device(intf, priv);
935 	mutex_unlock(&intf_mutex);
936 
937 	return 0;
938 }
939 EXPORT_SYMBOL(mlx5_register_interface);
940 
941 void mlx5_unregister_interface(struct mlx5_interface *intf)
942 {
943 	struct mlx5_priv *priv;
944 
945 	mutex_lock(&intf_mutex);
946 	list_for_each_entry(priv, &dev_list, dev_list)
947 	       mlx5_remove_device(intf, priv);
948 	list_del(&intf->list);
949 	mutex_unlock(&intf_mutex);
950 }
951 EXPORT_SYMBOL(mlx5_unregister_interface);
952 
953 void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol)
954 {
955 	struct mlx5_priv *priv = &mdev->priv;
956 	struct mlx5_device_context *dev_ctx;
957 	unsigned long flags;
958 	void *result = NULL;
959 
960 	spin_lock_irqsave(&priv->ctx_lock, flags);
961 
962 	list_for_each_entry(dev_ctx, &mdev->priv.ctx_list, list)
963 		if ((dev_ctx->intf->protocol == protocol) &&
964 		    dev_ctx->intf->get_dev) {
965 			result = dev_ctx->intf->get_dev(dev_ctx->context);
966 			break;
967 		}
968 
969 	spin_unlock_irqrestore(&priv->ctx_lock, flags);
970 
971 	return result;
972 }
973 EXPORT_SYMBOL(mlx5_get_protocol_dev);
974 
975 static void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event,
976 			    unsigned long param)
977 {
978 	struct mlx5_priv *priv = &dev->priv;
979 	struct mlx5_device_context *dev_ctx;
980 	unsigned long flags;
981 
982 	spin_lock_irqsave(&priv->ctx_lock, flags);
983 
984 	list_for_each_entry(dev_ctx, &priv->ctx_list, list)
985 		if (dev_ctx->intf->event)
986 			dev_ctx->intf->event(dev, dev_ctx->context, event, param);
987 
988 	spin_unlock_irqrestore(&priv->ctx_lock, flags);
989 }
990 
991 struct mlx5_core_event_handler {
992 	void (*event)(struct mlx5_core_dev *dev,
993 		      enum mlx5_dev_event event,
994 		      void *data);
995 };
996 
997 
998 static int init_one(struct pci_dev *pdev,
999 		    const struct pci_device_id *id)
1000 {
1001 	struct mlx5_core_dev *dev;
1002 	struct mlx5_priv *priv;
1003 	int err;
1004 
1005 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1006 	priv = &dev->priv;
1007 
1008 	if (prof_sel < 0 || prof_sel >= ARRAY_SIZE(profiles)) {
1009 		printf("mlx5_core: WARN: ""selected profile out of range, selecting default (%d)\n", MLX5_DEFAULT_PROF);
1010 		prof_sel = MLX5_DEFAULT_PROF;
1011 	}
1012 	dev->profile = &profiles[prof_sel];
1013 	dev->event = mlx5_core_event;
1014 
1015 	INIT_LIST_HEAD(&priv->ctx_list);
1016 	spin_lock_init(&priv->ctx_lock);
1017 	err = mlx5_dev_init(dev, pdev);
1018 	if (err) {
1019 		device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_dev_init failed %d\n", err);
1020 		goto out;
1021 	}
1022 
1023 	err = mlx5_register_device(dev);
1024 	if (err) {
1025 		device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_register_device failed %d\n", err);
1026 		goto out_init;
1027 	}
1028 
1029 
1030 	return 0;
1031 
1032 out_init:
1033 	mlx5_dev_cleanup(dev);
1034 out:
1035 	kfree(dev);
1036 	return err;
1037 }
1038 
1039 static void remove_one(struct pci_dev *pdev)
1040 {
1041 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1042 
1043 	mlx5_unregister_device(dev);
1044 	mlx5_dev_cleanup(dev);
1045 	kfree(dev);
1046 }
1047 
1048 static const struct pci_device_id mlx5_core_pci_table[] = {
1049 	{ PCI_VDEVICE(MELLANOX, 4113) }, /* Connect-IB */
1050 	{ PCI_VDEVICE(MELLANOX, 4114) }, /* Connect-IB VF */
1051 	{ PCI_VDEVICE(MELLANOX, 4115) }, /* ConnectX-4 */
1052 	{ PCI_VDEVICE(MELLANOX, 4116) }, /* ConnectX-4 VF */
1053 	{ PCI_VDEVICE(MELLANOX, 4117) }, /* ConnectX-4LX */
1054 	{ PCI_VDEVICE(MELLANOX, 4118) }, /* ConnectX-4LX VF */
1055 	{ PCI_VDEVICE(MELLANOX, 4119) },
1056 	{ PCI_VDEVICE(MELLANOX, 4120) },
1057 	{ PCI_VDEVICE(MELLANOX, 4121) },
1058 	{ PCI_VDEVICE(MELLANOX, 4122) },
1059 	{ PCI_VDEVICE(MELLANOX, 4123) },
1060 	{ PCI_VDEVICE(MELLANOX, 4124) },
1061 	{ PCI_VDEVICE(MELLANOX, 4125) },
1062 	{ PCI_VDEVICE(MELLANOX, 4126) },
1063 	{ PCI_VDEVICE(MELLANOX, 4127) },
1064 	{ PCI_VDEVICE(MELLANOX, 4128) },
1065 	{ PCI_VDEVICE(MELLANOX, 4129) },
1066 	{ PCI_VDEVICE(MELLANOX, 4130) },
1067 	{ PCI_VDEVICE(MELLANOX, 4131) },
1068 	{ PCI_VDEVICE(MELLANOX, 4132) },
1069 	{ PCI_VDEVICE(MELLANOX, 4133) },
1070 	{ PCI_VDEVICE(MELLANOX, 4134) },
1071 	{ PCI_VDEVICE(MELLANOX, 4135) },
1072 	{ PCI_VDEVICE(MELLANOX, 4136) },
1073 	{ PCI_VDEVICE(MELLANOX, 4137) },
1074 	{ PCI_VDEVICE(MELLANOX, 4138) },
1075 	{ PCI_VDEVICE(MELLANOX, 4139) },
1076 	{ PCI_VDEVICE(MELLANOX, 4140) },
1077 	{ PCI_VDEVICE(MELLANOX, 4141) },
1078 	{ PCI_VDEVICE(MELLANOX, 4142) },
1079 	{ PCI_VDEVICE(MELLANOX, 4143) },
1080 	{ PCI_VDEVICE(MELLANOX, 4144) },
1081 	{ 0, }
1082 };
1083 
1084 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1085 
1086 static struct pci_driver mlx5_core_driver = {
1087 	.name           = DRIVER_NAME,
1088 	.id_table       = mlx5_core_pci_table,
1089 	.probe          = init_one,
1090 	.remove         = remove_one
1091 };
1092 
1093 static int __init init(void)
1094 {
1095 	int err;
1096 
1097 	mlx5_core_wq = create_singlethread_workqueue("mlx5_core_wq");
1098 	if (!mlx5_core_wq) {
1099 		err = -ENOMEM;
1100 		goto err_debug;
1101 	}
1102 	mlx5_health_init();
1103 
1104 	err = pci_register_driver(&mlx5_core_driver);
1105 	if (err)
1106 		goto err_health;
1107 
1108 
1109 	return 0;
1110 
1111 err_health:
1112 	mlx5_health_cleanup();
1113 	destroy_workqueue(mlx5_core_wq);
1114 err_debug:
1115 	return err;
1116 }
1117 
1118 static void __exit cleanup(void)
1119 {
1120 	pci_unregister_driver(&mlx5_core_driver);
1121 	mlx5_health_cleanup();
1122 	destroy_workqueue(mlx5_core_wq);
1123 }
1124 
1125 module_init(init);
1126 module_exit(cleanup);
1127