xref: /freebsd-src/sys/dev/mlx5/mlx5_core/mlx5_main.c (revision 7db2360401dc8e8ff7f44efc8218bb82015f6f89)
1 /*-
2  * Copyright (c) 2013-2017, 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 #define	LINUXKPI_PARAM_PREFIX mlx5_
29 
30 #include <linux/kmod.h>
31 #include <linux/module.h>
32 #include <linux/errno.h>
33 #include <linux/pci.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/slab.h>
36 #include <linux/io-mapping.h>
37 #include <linux/interrupt.h>
38 #include <dev/mlx5/driver.h>
39 #include <dev/mlx5/cq.h>
40 #include <dev/mlx5/qp.h>
41 #include <dev/mlx5/srq.h>
42 #include <linux/delay.h>
43 #include <dev/mlx5/mlx5_ifc.h>
44 #include "mlx5_core.h"
45 #include "fs_core.h"
46 
47 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
48 MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
49 MODULE_LICENSE("Dual BSD/GPL");
50 #if (__FreeBSD_version >= 1100000)
51 MODULE_DEPEND(mlx5, linuxkpi, 1, 1, 1);
52 #endif
53 MODULE_VERSION(mlx5, 1);
54 
55 int mlx5_core_debug_mask;
56 module_param_named(debug_mask, mlx5_core_debug_mask, int, 0644);
57 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
58 
59 #define MLX5_DEFAULT_PROF	2
60 static int prof_sel = MLX5_DEFAULT_PROF;
61 module_param_named(prof_sel, prof_sel, int, 0444);
62 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
63 
64 static int mlx5_core_msix_eqvec;
65 module_param_named(msix_eqvec, mlx5_core_msix_eqvec, int, 0644);
66 MODULE_PARM_DESC(msix_eqvec, "Maximum number of MSIX event queue vectors");
67 
68 #define NUMA_NO_NODE       -1
69 
70 static LIST_HEAD(intf_list);
71 static LIST_HEAD(dev_list);
72 static DEFINE_MUTEX(intf_mutex);
73 
74 struct mlx5_device_context {
75 	struct list_head	list;
76 	struct mlx5_interface  *intf;
77 	void		       *context;
78 };
79 
80 enum {
81 	MLX5_ATOMIC_REQ_MODE_BE = 0x0,
82 	MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
83 };
84 
85 static struct mlx5_profile profiles[] = {
86 	[0] = {
87 		.mask           = 0,
88 	},
89 	[1] = {
90 		.mask		= MLX5_PROF_MASK_QP_SIZE,
91 		.log_max_qp	= 12,
92 	},
93 	[2] = {
94 		.mask		= MLX5_PROF_MASK_QP_SIZE |
95 				  MLX5_PROF_MASK_MR_CACHE,
96 		.log_max_qp	= 17,
97 		.mr_cache[0]	= {
98 			.size	= 500,
99 			.limit	= 250
100 		},
101 		.mr_cache[1]	= {
102 			.size	= 500,
103 			.limit	= 250
104 		},
105 		.mr_cache[2]	= {
106 			.size	= 500,
107 			.limit	= 250
108 		},
109 		.mr_cache[3]	= {
110 			.size	= 500,
111 			.limit	= 250
112 		},
113 		.mr_cache[4]	= {
114 			.size	= 500,
115 			.limit	= 250
116 		},
117 		.mr_cache[5]	= {
118 			.size	= 500,
119 			.limit	= 250
120 		},
121 		.mr_cache[6]	= {
122 			.size	= 500,
123 			.limit	= 250
124 		},
125 		.mr_cache[7]	= {
126 			.size	= 500,
127 			.limit	= 250
128 		},
129 		.mr_cache[8]	= {
130 			.size	= 500,
131 			.limit	= 250
132 		},
133 		.mr_cache[9]	= {
134 			.size	= 500,
135 			.limit	= 250
136 		},
137 		.mr_cache[10]	= {
138 			.size	= 500,
139 			.limit	= 250
140 		},
141 		.mr_cache[11]	= {
142 			.size	= 500,
143 			.limit	= 250
144 		},
145 		.mr_cache[12]	= {
146 			.size	= 64,
147 			.limit	= 32
148 		},
149 		.mr_cache[13]	= {
150 			.size	= 32,
151 			.limit	= 16
152 		},
153 		.mr_cache[14]	= {
154 			.size	= 16,
155 			.limit	= 8
156 		},
157 	},
158 	[3] = {
159 		.mask		= MLX5_PROF_MASK_QP_SIZE,
160 		.log_max_qp	= 17,
161 	},
162 };
163 
164 static int set_dma_caps(struct pci_dev *pdev)
165 {
166 	int err;
167 
168 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
169 	if (err) {
170 		device_printf((&pdev->dev)->bsddev, "WARN: ""Warning: couldn't set 64-bit PCI DMA mask\n");
171 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
172 		if (err) {
173 			device_printf((&pdev->dev)->bsddev, "ERR: ""Can't set PCI DMA mask, aborting\n");
174 			return err;
175 		}
176 	}
177 
178 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
179 	if (err) {
180 		device_printf((&pdev->dev)->bsddev, "WARN: ""Warning: couldn't set 64-bit consistent PCI DMA mask\n");
181 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
182 		if (err) {
183 			device_printf((&pdev->dev)->bsddev, "ERR: ""Can't set consistent PCI DMA mask, aborting\n");
184 			return err;
185 		}
186 	}
187 
188 	dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
189 	return err;
190 }
191 
192 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev)
193 {
194 	struct pci_dev *pdev = dev->pdev;
195 	int err = 0;
196 
197 	mutex_lock(&dev->pci_status_mutex);
198 	if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) {
199 		err = pci_enable_device(pdev);
200 		if (!err)
201 			dev->pci_status = MLX5_PCI_STATUS_ENABLED;
202 	}
203 	mutex_unlock(&dev->pci_status_mutex);
204 
205 	return err;
206 }
207 
208 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev)
209 {
210 	struct pci_dev *pdev = dev->pdev;
211 
212 	mutex_lock(&dev->pci_status_mutex);
213 	if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) {
214 		pci_disable_device(pdev);
215 		dev->pci_status = MLX5_PCI_STATUS_DISABLED;
216 	}
217 	mutex_unlock(&dev->pci_status_mutex);
218 }
219 
220 static int request_bar(struct pci_dev *pdev)
221 {
222 	int err = 0;
223 
224 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
225 		device_printf((&pdev->dev)->bsddev, "ERR: ""Missing registers BAR, aborting\n");
226 		return -ENODEV;
227 	}
228 
229 	err = pci_request_regions(pdev, DRIVER_NAME);
230 	if (err)
231 		device_printf((&pdev->dev)->bsddev, "ERR: ""Couldn't get PCI resources, aborting\n");
232 
233 	return err;
234 }
235 
236 static void release_bar(struct pci_dev *pdev)
237 {
238 	pci_release_regions(pdev);
239 }
240 
241 static int mlx5_enable_msix(struct mlx5_core_dev *dev)
242 {
243 	struct mlx5_priv *priv = &dev->priv;
244 	struct mlx5_eq_table *table = &priv->eq_table;
245 	int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq);
246 	int limit = mlx5_core_msix_eqvec;
247 	int nvec = MLX5_EQ_VEC_COMP_BASE;
248 	int i;
249 
250 	if (limit > 0)
251 		nvec += limit;
252 	else
253 		nvec += MLX5_CAP_GEN(dev, num_ports) * num_online_cpus();
254 
255 	nvec = min_t(int, nvec, num_eqs);
256 	if (nvec <= MLX5_EQ_VEC_COMP_BASE)
257 		return -ENOMEM;
258 
259 	priv->msix_arr = kzalloc(nvec * sizeof(*priv->msix_arr), GFP_KERNEL);
260 
261 	priv->irq_info = kzalloc(nvec * sizeof(*priv->irq_info), GFP_KERNEL);
262 
263 	for (i = 0; i < nvec; i++)
264 		priv->msix_arr[i].entry = i;
265 
266 	nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr,
267 				     MLX5_EQ_VEC_COMP_BASE + 1, nvec);
268 	if (nvec < 0)
269 		return nvec;
270 
271 	table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
272 
273 	return 0;
274 
275 }
276 
277 static void mlx5_disable_msix(struct mlx5_core_dev *dev)
278 {
279 	struct mlx5_priv *priv = &dev->priv;
280 
281 	pci_disable_msix(dev->pdev);
282 	kfree(priv->irq_info);
283 	kfree(priv->msix_arr);
284 }
285 
286 struct mlx5_reg_host_endianess {
287 	u8	he;
288 	u8      rsvd[15];
289 };
290 
291 
292 #define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos))
293 
294 enum {
295 	MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) |
296 				MLX5_DEV_CAP_FLAG_DCT |
297 				MLX5_DEV_CAP_FLAG_DRAIN_SIGERR,
298 };
299 
300 static u16 to_fw_pkey_sz(u32 size)
301 {
302 	switch (size) {
303 	case 128:
304 		return 0;
305 	case 256:
306 		return 1;
307 	case 512:
308 		return 2;
309 	case 1024:
310 		return 3;
311 	case 2048:
312 		return 4;
313 	case 4096:
314 		return 5;
315 	default:
316 		printf("mlx5_core: WARN: ""invalid pkey table size %d\n", size);
317 		return 0;
318 	}
319 }
320 
321 static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev,
322 				   enum mlx5_cap_type cap_type,
323 				   enum mlx5_cap_mode cap_mode)
324 {
325 	u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
326 	int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
327 	void *out, *hca_caps;
328 	u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
329 	int err;
330 
331 	memset(in, 0, sizeof(in));
332 	out = kzalloc(out_sz, GFP_KERNEL);
333 
334 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
335 	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
336 	err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz);
337 	if (err) {
338 		mlx5_core_warn(dev,
339 			       "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
340 			       cap_type, cap_mode, err);
341 		goto query_ex;
342 	}
343 
344 	hca_caps =  MLX5_ADDR_OF(query_hca_cap_out, out, capability);
345 
346 	switch (cap_mode) {
347 	case HCA_CAP_OPMOD_GET_MAX:
348 		memcpy(dev->hca_caps_max[cap_type], hca_caps,
349 		       MLX5_UN_SZ_BYTES(hca_cap_union));
350 		break;
351 	case HCA_CAP_OPMOD_GET_CUR:
352 		memcpy(dev->hca_caps_cur[cap_type], hca_caps,
353 		       MLX5_UN_SZ_BYTES(hca_cap_union));
354 		break;
355 	default:
356 		mlx5_core_warn(dev,
357 			       "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
358 			       cap_type, cap_mode);
359 		err = -EINVAL;
360 		break;
361 	}
362 query_ex:
363 	kfree(out);
364 	return err;
365 }
366 
367 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
368 {
369 	int ret;
370 
371 	ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
372 	if (ret)
373 		return ret;
374 
375 	return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
376 }
377 
378 static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz)
379 {
380 	u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)] = {0};
381 
382 	MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
383 
384 	return mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out));
385 }
386 
387 static int handle_hca_cap(struct mlx5_core_dev *dev)
388 {
389 	void *set_ctx = NULL;
390 	struct mlx5_profile *prof = dev->profile;
391 	int err = -ENOMEM;
392 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
393 	void *set_hca_cap;
394 
395 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
396 
397 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
398 	if (err)
399 		goto query_ex;
400 
401 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
402 				   capability);
403 	memcpy(set_hca_cap, dev->hca_caps_cur[MLX5_CAP_GENERAL],
404 	       MLX5_ST_SZ_BYTES(cmd_hca_cap));
405 
406 	mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
407 		      mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
408 		      128);
409 	/* we limit the size of the pkey table to 128 entries for now */
410 	MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
411 		 to_fw_pkey_sz(128));
412 
413 	if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
414 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
415 			 prof->log_max_qp);
416 
417 	/* disable cmdif checksum */
418 	MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
419 
420 	/* enable drain sigerr */
421 	MLX5_SET(cmd_hca_cap, set_hca_cap, drain_sigerr, 1);
422 
423 	MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
424 
425 	err = set_caps(dev, set_ctx, set_sz);
426 
427 query_ex:
428 	kfree(set_ctx);
429 	return err;
430 }
431 
432 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev)
433 {
434 	void *set_ctx;
435 	void *set_hca_cap;
436 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
437 	int req_endianness;
438 	int err;
439 
440 	if (MLX5_CAP_GEN(dev, atomic)) {
441 		err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
442 		if (err)
443 			return err;
444 	} else {
445 		return 0;
446 	}
447 
448 	req_endianness =
449 		MLX5_CAP_ATOMIC(dev,
450 				supported_atomic_req_8B_endianess_mode_1);
451 
452 	if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
453 		return 0;
454 
455 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
456 	if (!set_ctx)
457 		return -ENOMEM;
458 
459 	MLX5_SET(set_hca_cap_in, set_ctx, op_mod,
460 		 MLX5_SET_HCA_CAP_OP_MOD_ATOMIC << 1);
461 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
462 
463 	/* Set requestor to host endianness */
464 	MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianess_mode,
465 		 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
466 
467 	err = set_caps(dev, set_ctx, set_sz);
468 
469 	kfree(set_ctx);
470 	return err;
471 }
472 
473 static int set_hca_ctrl(struct mlx5_core_dev *dev)
474 {
475 	struct mlx5_reg_host_endianess he_in;
476 	struct mlx5_reg_host_endianess he_out;
477 	int err;
478 
479 	if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH &&
480 	    !MLX5_CAP_GEN(dev, roce))
481 		return 0;
482 
483 	memset(&he_in, 0, sizeof(he_in));
484 	he_in.he = MLX5_SET_HOST_ENDIANNESS;
485 	err = mlx5_core_access_reg(dev, &he_in,  sizeof(he_in),
486 					&he_out, sizeof(he_out),
487 					MLX5_REG_HOST_ENDIANNESS, 0, 1);
488 	return err;
489 }
490 
491 static int mlx5_core_enable_hca(struct mlx5_core_dev *dev)
492 {
493 	u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0};
494 	u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0};
495 
496 	MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
497 	return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
498 }
499 
500 static int mlx5_core_disable_hca(struct mlx5_core_dev *dev)
501 {
502 	u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0};
503 	u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0};
504 
505 	MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
506 	return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
507 }
508 
509 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
510 {
511 	u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0};
512 	u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0};
513 	u32 sup_issi;
514 	int err;
515 
516 	MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
517 
518 	err = mlx5_cmd_exec(dev, query_in, sizeof(query_in), query_out, sizeof(query_out));
519 	if (err) {
520 		u32 syndrome;
521 		u8 status;
522 
523 		mlx5_cmd_mbox_status(query_out, &status, &syndrome);
524 		if (status == MLX5_CMD_STAT_BAD_OP_ERR) {
525 			pr_debug("Only ISSI 0 is supported\n");
526 			return 0;
527 		}
528 
529 		printf("mlx5_core: ERR: ""failed to query ISSI\n");
530 		return err;
531 	}
532 
533 	sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
534 
535 	if (sup_issi & (1 << 1)) {
536 		u32 set_in[MLX5_ST_SZ_DW(set_issi_in)]	 = {0};
537 		u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0};
538 
539 		MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
540 		MLX5_SET(set_issi_in, set_in, current_issi, 1);
541 
542 		err = mlx5_cmd_exec(dev, set_in, sizeof(set_in), set_out, sizeof(set_out));
543 		if (err) {
544 			printf("mlx5_core: ERR: ""failed to set ISSI=1 err(%d)\n", err);
545 			return err;
546 		}
547 
548 		dev->issi = 1;
549 
550 		return 0;
551 	} else if (sup_issi & (1 << 0)) {
552 		return 0;
553 	}
554 
555 	return -ENOTSUPP;
556 }
557 
558 
559 int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn, int *irqn)
560 {
561 	struct mlx5_eq_table *table = &dev->priv.eq_table;
562 	struct mlx5_eq *eq;
563 	int err = -ENOENT;
564 
565 	spin_lock(&table->lock);
566 	list_for_each_entry(eq, &table->comp_eqs_list, list) {
567 		if (eq->index == vector) {
568 			*eqn = eq->eqn;
569 			*irqn = eq->irqn;
570 			err = 0;
571 			break;
572 		}
573 	}
574 	spin_unlock(&table->lock);
575 
576 	return err;
577 }
578 EXPORT_SYMBOL(mlx5_vector2eqn);
579 
580 int mlx5_rename_eq(struct mlx5_core_dev *dev, int eq_ix, char *name)
581 {
582 	struct mlx5_priv *priv = &dev->priv;
583 	struct mlx5_eq_table *table = &priv->eq_table;
584 	struct mlx5_eq *eq;
585 	int err = -ENOENT;
586 
587 	spin_lock(&table->lock);
588 	list_for_each_entry(eq, &table->comp_eqs_list, list) {
589 		if (eq->index == eq_ix) {
590 			int irq_ix = eq_ix + MLX5_EQ_VEC_COMP_BASE;
591 
592 			snprintf(priv->irq_info[irq_ix].name, MLX5_MAX_IRQ_NAME,
593 				 "%s-%d", name, eq_ix);
594 
595 			err = 0;
596 			break;
597 		}
598 	}
599 	spin_unlock(&table->lock);
600 
601 	return err;
602 }
603 
604 static void free_comp_eqs(struct mlx5_core_dev *dev)
605 {
606 	struct mlx5_eq_table *table = &dev->priv.eq_table;
607 	struct mlx5_eq *eq, *n;
608 
609 	spin_lock(&table->lock);
610 	list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) {
611 		list_del(&eq->list);
612 		spin_unlock(&table->lock);
613 		if (mlx5_destroy_unmap_eq(dev, eq))
614 			mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n",
615 				       eq->eqn);
616 		kfree(eq);
617 		spin_lock(&table->lock);
618 	}
619 	spin_unlock(&table->lock);
620 }
621 
622 static int alloc_comp_eqs(struct mlx5_core_dev *dev)
623 {
624 	struct mlx5_eq_table *table = &dev->priv.eq_table;
625 	char name[MLX5_MAX_IRQ_NAME];
626 	struct mlx5_eq *eq;
627 	int ncomp_vec;
628 	int nent;
629 	int err;
630 	int i;
631 
632 	INIT_LIST_HEAD(&table->comp_eqs_list);
633 	ncomp_vec = table->num_comp_vectors;
634 	nent = MLX5_COMP_EQ_SIZE;
635 	for (i = 0; i < ncomp_vec; i++) {
636 		eq = kzalloc(sizeof(*eq), GFP_KERNEL);
637 
638 		snprintf(name, MLX5_MAX_IRQ_NAME, "mlx5_comp%d", i);
639 		err = mlx5_create_map_eq(dev, eq,
640 					 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
641 					 name, &dev->priv.uuari.uars[0]);
642 		if (err) {
643 			kfree(eq);
644 			goto clean;
645 		}
646 		mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
647 		eq->index = i;
648 		spin_lock(&table->lock);
649 		list_add_tail(&eq->list, &table->comp_eqs_list);
650 		spin_unlock(&table->lock);
651 	}
652 
653 	return 0;
654 
655 clean:
656 	free_comp_eqs(dev);
657 	return err;
658 }
659 
660 static int map_bf_area(struct mlx5_core_dev *dev)
661 {
662 	resource_size_t bf_start = pci_resource_start(dev->pdev, 0);
663 	resource_size_t bf_len = pci_resource_len(dev->pdev, 0);
664 
665 	dev->priv.bf_mapping = io_mapping_create_wc(bf_start, bf_len);
666 
667 	return dev->priv.bf_mapping ? 0 : -ENOMEM;
668 }
669 
670 static void unmap_bf_area(struct mlx5_core_dev *dev)
671 {
672 	if (dev->priv.bf_mapping)
673 		io_mapping_free(dev->priv.bf_mapping);
674 }
675 
676 static inline int fw_initializing(struct mlx5_core_dev *dev)
677 {
678 	return ioread32be(&dev->iseg->initializing) >> 31;
679 }
680 
681 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili)
682 {
683 	u64 end = jiffies + msecs_to_jiffies(max_wait_mili);
684 	int err = 0;
685 
686 	while (fw_initializing(dev)) {
687 		if (time_after(jiffies, end)) {
688 			err = -EBUSY;
689 			break;
690 		}
691 		msleep(FW_INIT_WAIT_MS);
692 	}
693 
694 	return err;
695 }
696 
697 static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
698 {
699 	struct mlx5_device_context *dev_ctx;
700 	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
701 
702 	dev_ctx = kzalloc(sizeof(*dev_ctx), GFP_KERNEL);
703 	if (!dev_ctx)
704 		return;
705 
706 	dev_ctx->intf    = intf;
707 	CURVNET_SET_QUIET(vnet0);
708 	dev_ctx->context = intf->add(dev);
709 	CURVNET_RESTORE();
710 
711 	if (dev_ctx->context) {
712 		spin_lock_irq(&priv->ctx_lock);
713 		list_add_tail(&dev_ctx->list, &priv->ctx_list);
714 		spin_unlock_irq(&priv->ctx_lock);
715 	} else {
716 		kfree(dev_ctx);
717 	}
718 }
719 
720 static void mlx5_remove_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
721 {
722 	struct mlx5_device_context *dev_ctx;
723 	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
724 
725 	list_for_each_entry(dev_ctx, &priv->ctx_list, list)
726 		if (dev_ctx->intf == intf) {
727 			spin_lock_irq(&priv->ctx_lock);
728 			list_del(&dev_ctx->list);
729 			spin_unlock_irq(&priv->ctx_lock);
730 
731 			intf->remove(dev, dev_ctx->context);
732 			kfree(dev_ctx);
733 			return;
734 		}
735 }
736 
737 static int mlx5_register_device(struct mlx5_core_dev *dev)
738 {
739 	struct mlx5_priv *priv = &dev->priv;
740 	struct mlx5_interface *intf;
741 
742 	mutex_lock(&intf_mutex);
743 	list_add_tail(&priv->dev_list, &dev_list);
744 	list_for_each_entry(intf, &intf_list, list)
745 		mlx5_add_device(intf, priv);
746 	mutex_unlock(&intf_mutex);
747 
748 	return 0;
749 }
750 
751 static void mlx5_unregister_device(struct mlx5_core_dev *dev)
752 {
753 	struct mlx5_priv *priv = &dev->priv;
754 	struct mlx5_interface *intf;
755 
756 	mutex_lock(&intf_mutex);
757 	list_for_each_entry(intf, &intf_list, list)
758 		mlx5_remove_device(intf, priv);
759 	list_del(&priv->dev_list);
760 	mutex_unlock(&intf_mutex);
761 }
762 
763 int mlx5_register_interface(struct mlx5_interface *intf)
764 {
765 	struct mlx5_priv *priv;
766 
767 	if (!intf->add || !intf->remove)
768 		return -EINVAL;
769 
770 	mutex_lock(&intf_mutex);
771 	list_add_tail(&intf->list, &intf_list);
772 	list_for_each_entry(priv, &dev_list, dev_list)
773 		mlx5_add_device(intf, priv);
774 	mutex_unlock(&intf_mutex);
775 
776 	return 0;
777 }
778 EXPORT_SYMBOL(mlx5_register_interface);
779 
780 void mlx5_unregister_interface(struct mlx5_interface *intf)
781 {
782 	struct mlx5_priv *priv;
783 
784 	mutex_lock(&intf_mutex);
785 	list_for_each_entry(priv, &dev_list, dev_list)
786 		mlx5_remove_device(intf, priv);
787 	list_del(&intf->list);
788 	mutex_unlock(&intf_mutex);
789 }
790 EXPORT_SYMBOL(mlx5_unregister_interface);
791 
792 void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol)
793 {
794 	struct mlx5_priv *priv = &mdev->priv;
795 	struct mlx5_device_context *dev_ctx;
796 	unsigned long flags;
797 	void *result = NULL;
798 
799 	spin_lock_irqsave(&priv->ctx_lock, flags);
800 
801 	list_for_each_entry(dev_ctx, &mdev->priv.ctx_list, list)
802 		if ((dev_ctx->intf->protocol == protocol) &&
803 		    dev_ctx->intf->get_dev) {
804 			result = dev_ctx->intf->get_dev(dev_ctx->context);
805 			break;
806 		}
807 
808 	spin_unlock_irqrestore(&priv->ctx_lock, flags);
809 
810 	return result;
811 }
812 EXPORT_SYMBOL(mlx5_get_protocol_dev);
813 
814 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
815 {
816 	struct pci_dev *pdev = dev->pdev;
817 	int err = 0;
818 
819 	pci_set_drvdata(dev->pdev, dev);
820 	strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
821 	priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
822 
823 	mutex_init(&priv->pgdir_mutex);
824 	INIT_LIST_HEAD(&priv->pgdir_list);
825 	spin_lock_init(&priv->mkey_lock);
826 
827 	priv->numa_node = NUMA_NO_NODE;
828 
829 	err = mlx5_pci_enable_device(dev);
830 	if (err) {
831 		device_printf((&pdev->dev)->bsddev, "ERR: ""Cannot enable PCI device, aborting\n");
832 		goto err_dbg;
833 	}
834 
835 	err = request_bar(pdev);
836 	if (err) {
837 		device_printf((&pdev->dev)->bsddev, "ERR: ""error requesting BARs, aborting\n");
838 		goto err_disable;
839 	}
840 
841 	pci_set_master(pdev);
842 
843 	err = set_dma_caps(pdev);
844 	if (err) {
845 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed setting DMA capabilities mask, aborting\n");
846 		goto err_clr_master;
847 	}
848 
849 	dev->iseg_base = pci_resource_start(dev->pdev, 0);
850 	dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
851 	if (!dev->iseg) {
852 		err = -ENOMEM;
853 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed mapping initialization segment, aborting\n");
854 		goto err_clr_master;
855 	}
856 
857 	return 0;
858 
859 err_clr_master:
860 	pci_clear_master(dev->pdev);
861 	release_bar(dev->pdev);
862 err_disable:
863 	mlx5_pci_disable_device(dev);
864 err_dbg:
865 	return err;
866 }
867 
868 static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
869 {
870 	iounmap(dev->iseg);
871 	pci_clear_master(dev->pdev);
872 	release_bar(dev->pdev);
873 	mlx5_pci_disable_device(dev);
874 }
875 
876 static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
877 {
878 	struct pci_dev *pdev = dev->pdev;
879 	int err;
880 
881 	err = mlx5_vsc_find_cap(dev);
882 	if (err)
883 		dev_err(&pdev->dev, "Unable to find vendor specific capabilities\n");
884 
885 	err = mlx5_query_hca_caps(dev);
886 	if (err) {
887 		dev_err(&pdev->dev, "query hca failed\n");
888 		goto out;
889 	}
890 
891 	err = mlx5_query_board_id(dev);
892 	if (err) {
893 		dev_err(&pdev->dev, "query board id failed\n");
894 		goto out;
895 	}
896 
897 	err = mlx5_eq_init(dev);
898 	if (err) {
899 		dev_err(&pdev->dev, "failed to initialize eq\n");
900 		goto out;
901 	}
902 
903 	MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock);
904 
905 	err = mlx5_init_cq_table(dev);
906 	if (err) {
907 		dev_err(&pdev->dev, "failed to initialize cq table\n");
908 		goto err_eq_cleanup;
909 	}
910 
911 	mlx5_init_qp_table(dev);
912 	mlx5_init_srq_table(dev);
913 	mlx5_init_mr_table(dev);
914 
915 #ifdef RATELIMIT
916 	err = mlx5_init_rl_table(dev);
917 	if (err) {
918 		dev_err(&pdev->dev, "Failed to init rate limiting\n");
919 		goto err_tables_cleanup;
920 	}
921 #endif
922 	return 0;
923 
924 #ifdef RATELIMIT
925 err_tables_cleanup:
926 	mlx5_cleanup_mr_table(dev);
927 	mlx5_cleanup_srq_table(dev);
928 	mlx5_cleanup_qp_table(dev);
929 	mlx5_cleanup_cq_table(dev);
930 #endif
931 
932 err_eq_cleanup:
933 	mlx5_eq_cleanup(dev);
934 
935 out:
936 	return err;
937 }
938 
939 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
940 {
941 #ifdef RATELIMIT
942 	mlx5_cleanup_rl_table(dev);
943 #endif
944 	mlx5_cleanup_mr_table(dev);
945 	mlx5_cleanup_srq_table(dev);
946 	mlx5_cleanup_qp_table(dev);
947 	mlx5_cleanup_cq_table(dev);
948 	mlx5_eq_cleanup(dev);
949 }
950 
951 static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
952 			 bool boot)
953 {
954 	struct pci_dev *pdev = dev->pdev;
955 	int err;
956 
957 	mutex_lock(&dev->intf_state_mutex);
958 	if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
959 		dev_warn(&dev->pdev->dev, "%s: interface is up, NOP\n",
960 			 __func__);
961 		goto out;
962 	}
963 
964 	device_printf((&pdev->dev)->bsddev, "INFO: ""firmware version: %d.%d.%d\n", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev));
965 
966 	/*
967 	 * On load removing any previous indication of internal error,
968 	 * device is up
969 	 */
970 	dev->state = MLX5_DEVICE_STATE_UP;
971 
972 	err = mlx5_cmd_init(dev);
973 	if (err) {
974 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed initializing command interface, aborting\n");
975 		goto out_err;
976 	}
977 
978 	err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI);
979 	if (err) {
980 		device_printf((&dev->pdev->dev)->bsddev, "ERR: ""Firmware over %d MS in initializing state, aborting\n", FW_INIT_TIMEOUT_MILI);
981 		goto err_cmd_cleanup;
982 	}
983 
984 	err = mlx5_core_enable_hca(dev);
985 	if (err) {
986 		device_printf((&pdev->dev)->bsddev, "ERR: ""enable hca failed\n");
987 		goto err_cmd_cleanup;
988 	}
989 
990 	err = mlx5_core_set_issi(dev);
991 	if (err) {
992 		device_printf((&pdev->dev)->bsddev, "ERR: ""failed to set issi\n");
993 		goto err_disable_hca;
994 	}
995 
996 	err = mlx5_pagealloc_start(dev);
997 	if (err) {
998 		device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_pagealloc_start failed\n");
999 		goto err_disable_hca;
1000 	}
1001 
1002 	err = mlx5_satisfy_startup_pages(dev, 1);
1003 	if (err) {
1004 		device_printf((&pdev->dev)->bsddev, "ERR: ""failed to allocate boot pages\n");
1005 		goto err_pagealloc_stop;
1006 	}
1007 
1008 	err = set_hca_ctrl(dev);
1009 	if (err) {
1010 		device_printf((&pdev->dev)->bsddev, "ERR: ""set_hca_ctrl failed\n");
1011 		goto reclaim_boot_pages;
1012 	}
1013 
1014 	err = handle_hca_cap(dev);
1015 	if (err) {
1016 		device_printf((&pdev->dev)->bsddev, "ERR: ""handle_hca_cap failed\n");
1017 		goto reclaim_boot_pages;
1018 	}
1019 
1020 	err = handle_hca_cap_atomic(dev);
1021 	if (err) {
1022 		device_printf((&pdev->dev)->bsddev, "ERR: ""handle_hca_cap_atomic failed\n");
1023 		goto reclaim_boot_pages;
1024 	}
1025 
1026 	err = mlx5_satisfy_startup_pages(dev, 0);
1027 	if (err) {
1028 		device_printf((&pdev->dev)->bsddev, "ERR: ""failed to allocate init pages\n");
1029 		goto reclaim_boot_pages;
1030 	}
1031 
1032 	err = mlx5_cmd_init_hca(dev);
1033 	if (err) {
1034 		device_printf((&pdev->dev)->bsddev, "ERR: ""init hca failed\n");
1035 		goto reclaim_boot_pages;
1036 	}
1037 
1038 	mlx5_start_health_poll(dev);
1039 
1040 	if (boot && mlx5_init_once(dev, priv)) {
1041 		dev_err(&pdev->dev, "sw objs init failed\n");
1042 		goto err_stop_poll;
1043 	}
1044 
1045 	err = mlx5_enable_msix(dev);
1046 	if (err) {
1047 		device_printf((&pdev->dev)->bsddev, "ERR: ""enable msix failed\n");
1048 		goto err_cleanup_once;
1049 	}
1050 
1051 	err = mlx5_alloc_uuars(dev, &priv->uuari);
1052 	if (err) {
1053 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed allocating uar, aborting\n");
1054 		goto err_disable_msix;
1055 	}
1056 
1057 	err = mlx5_start_eqs(dev);
1058 	if (err) {
1059 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to start pages and async EQs\n");
1060 		goto err_free_uar;
1061 	}
1062 
1063 	err = alloc_comp_eqs(dev);
1064 	if (err) {
1065 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to alloc completion EQs\n");
1066 		goto err_stop_eqs;
1067 	}
1068 
1069 	if (map_bf_area(dev))
1070 		device_printf((&pdev->dev)->bsddev, "ERR: ""Failed to map blue flame area\n");
1071 
1072 	err = mlx5_init_fs(dev);
1073 	if (err) {
1074 		mlx5_core_err(dev, "flow steering init %d\n", err);
1075 		goto err_free_comp_eqs;
1076 	}
1077 
1078 	err = mlx5_register_device(dev);
1079 	if (err) {
1080 		dev_err(&pdev->dev, "mlx5_register_device failed %d\n", err);
1081 		goto err_fs;
1082 	}
1083 
1084 	clear_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state);
1085 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1086 
1087 out:
1088 	mutex_unlock(&dev->intf_state_mutex);
1089 	return 0;
1090 
1091 err_fs:
1092 	mlx5_cleanup_fs(dev);
1093 
1094 err_free_comp_eqs:
1095 	free_comp_eqs(dev);
1096 	unmap_bf_area(dev);
1097 
1098 err_stop_eqs:
1099 	mlx5_stop_eqs(dev);
1100 
1101 err_free_uar:
1102 	mlx5_free_uuars(dev, &priv->uuari);
1103 
1104 err_disable_msix:
1105 	mlx5_disable_msix(dev);
1106 
1107 err_cleanup_once:
1108 	if (boot)
1109 		mlx5_cleanup_once(dev);
1110 
1111 err_stop_poll:
1112 	mlx5_stop_health_poll(dev);
1113 	if (mlx5_cmd_teardown_hca(dev)) {
1114 		device_printf((&dev->pdev->dev)->bsddev, "ERR: ""tear_down_hca failed, skip cleanup\n");
1115 		goto out_err;
1116 	}
1117 
1118 reclaim_boot_pages:
1119 	mlx5_reclaim_startup_pages(dev);
1120 
1121 err_pagealloc_stop:
1122 	mlx5_pagealloc_stop(dev);
1123 
1124 err_disable_hca:
1125 	mlx5_core_disable_hca(dev);
1126 
1127 err_cmd_cleanup:
1128 	mlx5_cmd_cleanup(dev);
1129 
1130 out_err:
1131 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1132 	mutex_unlock(&dev->intf_state_mutex);
1133 
1134 	return err;
1135 }
1136 
1137 static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
1138 			   bool cleanup)
1139 {
1140 	int err = 0;
1141 
1142 	if (cleanup)
1143 		mlx5_drain_health_recovery(dev);
1144 
1145 	mutex_lock(&dev->intf_state_mutex);
1146 	if (test_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state)) {
1147 		dev_warn(&dev->pdev->dev, "%s: interface is down, NOP\n", __func__);
1148                 if (cleanup)
1149                         mlx5_cleanup_once(dev);
1150 		goto out;
1151 	}
1152 
1153 	mlx5_unregister_device(dev);
1154 
1155 	mlx5_cleanup_fs(dev);
1156 	unmap_bf_area(dev);
1157 	mlx5_wait_for_reclaim_vfs_pages(dev);
1158 	free_comp_eqs(dev);
1159 	mlx5_stop_eqs(dev);
1160 	mlx5_free_uuars(dev, &priv->uuari);
1161 	mlx5_disable_msix(dev);
1162         if (cleanup)
1163                 mlx5_cleanup_once(dev);
1164 	mlx5_stop_health_poll(dev);
1165 	err = mlx5_cmd_teardown_hca(dev);
1166 	if (err) {
1167 		device_printf((&dev->pdev->dev)->bsddev, "ERR: ""tear_down_hca failed, skip cleanup\n");
1168 		goto out;
1169 	}
1170 	mlx5_pagealloc_stop(dev);
1171 	mlx5_reclaim_startup_pages(dev);
1172 	mlx5_core_disable_hca(dev);
1173 	mlx5_cmd_cleanup(dev);
1174 
1175 out:
1176 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1177 	set_bit(MLX5_INTERFACE_STATE_DOWN, &dev->intf_state);
1178 	mutex_unlock(&dev->intf_state_mutex);
1179 	return err;
1180 }
1181 
1182 void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event,
1183 		     unsigned long param)
1184 {
1185 	struct mlx5_priv *priv = &dev->priv;
1186 	struct mlx5_device_context *dev_ctx;
1187 	unsigned long flags;
1188 
1189 	spin_lock_irqsave(&priv->ctx_lock, flags);
1190 
1191 	list_for_each_entry(dev_ctx, &priv->ctx_list, list)
1192 		if (dev_ctx->intf->event)
1193 			dev_ctx->intf->event(dev, dev_ctx->context, event, param);
1194 
1195 	spin_unlock_irqrestore(&priv->ctx_lock, flags);
1196 }
1197 
1198 struct mlx5_core_event_handler {
1199 	void (*event)(struct mlx5_core_dev *dev,
1200 		      enum mlx5_dev_event event,
1201 		      void *data);
1202 };
1203 
1204 static int init_one(struct pci_dev *pdev,
1205 		    const struct pci_device_id *id)
1206 {
1207 	struct mlx5_core_dev *dev;
1208 	struct mlx5_priv *priv;
1209 	int err;
1210 
1211 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1212 	priv = &dev->priv;
1213 	if (id)
1214 		priv->pci_dev_data = id->driver_data;
1215 
1216 	if (prof_sel < 0 || prof_sel >= ARRAY_SIZE(profiles)) {
1217 		printf("mlx5_core: WARN: ""selected profile out of range, selecting default (%d)\n", MLX5_DEFAULT_PROF);
1218 		prof_sel = MLX5_DEFAULT_PROF;
1219 	}
1220 	dev->profile = &profiles[prof_sel];
1221 	dev->pdev = pdev;
1222 	dev->event = mlx5_core_event;
1223 
1224 	INIT_LIST_HEAD(&priv->ctx_list);
1225 	spin_lock_init(&priv->ctx_lock);
1226         mutex_init(&dev->pci_status_mutex);
1227         mutex_init(&dev->intf_state_mutex);
1228 	err = mlx5_pci_init(dev, priv);
1229 	if (err) {
1230 		device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_pci_init failed %d\n", err);
1231 		goto clean_dev;
1232 	}
1233 
1234         err = mlx5_health_init(dev);
1235         if (err) {
1236                 device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_health_init failed %d\n", err);
1237                 goto close_pci;
1238         }
1239 
1240 	mlx5_pagealloc_init(dev);
1241 
1242 	err = mlx5_load_one(dev, priv, true);
1243 	if (err) {
1244 		device_printf((&pdev->dev)->bsddev, "ERR: ""mlx5_register_device failed %d\n", err);
1245 		goto clean_health;
1246 	}
1247 
1248 	mlx5_fwdump_prep(dev);
1249 
1250 	pci_save_state(pdev->dev.bsddev);
1251 	return 0;
1252 
1253 clean_health:
1254 	mlx5_pagealloc_cleanup(dev);
1255         mlx5_health_cleanup(dev);
1256 close_pci:
1257         mlx5_pci_close(dev, priv);
1258 clean_dev:
1259 	kfree(dev);
1260 	return err;
1261 }
1262 
1263 static void remove_one(struct pci_dev *pdev)
1264 {
1265 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1266 	struct mlx5_priv *priv = &dev->priv;
1267 
1268 	if (mlx5_unload_one(dev, priv, true)) {
1269 		dev_err(&dev->pdev->dev, "mlx5_unload_one failed\n");
1270 		mlx5_health_cleanup(dev);
1271 		return;
1272 	}
1273 
1274 	mlx5_fwdump_clean(dev);
1275 	mlx5_pagealloc_cleanup(dev);
1276 	mlx5_health_cleanup(dev);
1277 	mlx5_pci_close(dev, priv);
1278 	pci_set_drvdata(pdev, NULL);
1279 	kfree(dev);
1280 }
1281 
1282 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
1283 					      pci_channel_state_t state)
1284 {
1285 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1286 	struct mlx5_priv *priv = &dev->priv;
1287 
1288 	dev_info(&pdev->dev, "%s was called\n", __func__);
1289 	mlx5_enter_error_state(dev, false);
1290 	mlx5_unload_one(dev, priv, false);
1291 
1292 	if (state) {
1293 		mlx5_drain_health_wq(dev);
1294 		mlx5_pci_disable_device(dev);
1295 	}
1296 
1297 	return state == pci_channel_io_perm_failure ?
1298 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1299 }
1300 
1301 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
1302 {
1303 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1304 	int err = 0;
1305 
1306 	dev_info(&pdev->dev, "%s was called\n", __func__);
1307 
1308 	err = mlx5_pci_enable_device(dev);
1309 	if (err) {
1310 		dev_err(&pdev->dev, "%s: mlx5_pci_enable_device failed with error code: %d\n"
1311 			, __func__, err);
1312 		return PCI_ERS_RESULT_DISCONNECT;
1313 	}
1314 	pci_set_master(pdev);
1315 	pci_set_powerstate(pdev->dev.bsddev, PCI_POWERSTATE_D0);
1316 	pci_restore_state(pdev->dev.bsddev);
1317 	pci_save_state(pdev->dev.bsddev);
1318 
1319 	return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
1320 }
1321 
1322 /* wait for the device to show vital signs. For now we check
1323  * that we can read the device ID and that the health buffer
1324  * shows a non zero value which is different than 0xffffffff
1325  */
1326 static void wait_vital(struct pci_dev *pdev)
1327 {
1328 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1329 	struct mlx5_core_health *health = &dev->priv.health;
1330 	const int niter = 100;
1331 	u32 count;
1332 	u16 did;
1333 	int i;
1334 
1335 	/* Wait for firmware to be ready after reset */
1336 	msleep(1000);
1337 	for (i = 0; i < niter; i++) {
1338 		if (pci_read_config_word(pdev, 2, &did)) {
1339 			dev_warn(&pdev->dev, "failed reading config word\n");
1340 			break;
1341 		}
1342 		if (did == pdev->device) {
1343 			dev_info(&pdev->dev, "device ID correctly read after %d iterations\n", i);
1344 			break;
1345 		}
1346 		msleep(50);
1347 	}
1348 	if (i == niter)
1349 		dev_warn(&pdev->dev, "%s-%d: could not read device ID\n", __func__, __LINE__);
1350 
1351 	for (i = 0; i < niter; i++) {
1352 		count = ioread32be(health->health_counter);
1353 		if (count && count != 0xffffffff) {
1354 			dev_info(&pdev->dev, "Counter value 0x%x after %d iterations\n", count, i);
1355 			break;
1356 		}
1357 		msleep(50);
1358 	}
1359 
1360 	if (i == niter)
1361 		dev_warn(&pdev->dev, "%s-%d: could not read device ID\n", __func__, __LINE__);
1362 }
1363 
1364 static void mlx5_pci_resume(struct pci_dev *pdev)
1365 {
1366 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1367 	struct mlx5_priv *priv = &dev->priv;
1368 	int err;
1369 
1370 	dev_info(&pdev->dev, "%s was called\n", __func__);
1371 
1372 	wait_vital(pdev);
1373 
1374 	err = mlx5_load_one(dev, priv, false);
1375 	if (err)
1376 		dev_err(&pdev->dev, "%s: mlx5_load_one failed with error code: %d\n"
1377 			, __func__, err);
1378 	else
1379 		dev_info(&pdev->dev, "%s: device recovered\n", __func__);
1380 }
1381 
1382 static const struct pci_error_handlers mlx5_err_handler = {
1383 	.error_detected = mlx5_pci_err_detected,
1384 	.slot_reset	= mlx5_pci_slot_reset,
1385 	.resume		= mlx5_pci_resume
1386 };
1387 
1388 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
1389 {
1390 	int err;
1391 
1392 	if (!MLX5_CAP_GEN(dev, force_teardown)) {
1393 		mlx5_core_dbg(dev, "force teardown is not supported in the firmware\n");
1394 		return -EOPNOTSUPP;
1395 	}
1396 
1397 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1398 		mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
1399 		return -EAGAIN;
1400 	}
1401 
1402 	err = mlx5_cmd_force_teardown_hca(dev);
1403 	if (err) {
1404 		mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", err);
1405 		return err;
1406 	}
1407 
1408 	mlx5_enter_error_state(dev, true);
1409 
1410 	return 0;
1411 }
1412 
1413 static void shutdown_one(struct pci_dev *pdev)
1414 {
1415 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1416 	struct mlx5_priv *priv = &dev->priv;
1417 	int err;
1418 
1419 	set_bit(MLX5_INTERFACE_STATE_SHUTDOWN, &dev->intf_state);
1420 	err = mlx5_try_fast_unload(dev);
1421 	if (err)
1422 	        mlx5_unload_one(dev, priv, false);
1423 	mlx5_pci_disable_device(dev);
1424 }
1425 
1426 static const struct pci_device_id mlx5_core_pci_table[] = {
1427 	{ PCI_VDEVICE(MELLANOX, 4113) }, /* Connect-IB */
1428 	{ PCI_VDEVICE(MELLANOX, 4114) }, /* Connect-IB VF */
1429 	{ PCI_VDEVICE(MELLANOX, 4115) }, /* ConnectX-4 */
1430 	{ PCI_VDEVICE(MELLANOX, 4116) }, /* ConnectX-4 VF */
1431 	{ PCI_VDEVICE(MELLANOX, 4117) }, /* ConnectX-4LX */
1432 	{ PCI_VDEVICE(MELLANOX, 4118) }, /* ConnectX-4LX VF */
1433 	{ PCI_VDEVICE(MELLANOX, 4119) }, /* ConnectX-5 */
1434 	{ PCI_VDEVICE(MELLANOX, 4120) }, /* ConnectX-5 VF */
1435 	{ PCI_VDEVICE(MELLANOX, 4121) },
1436 	{ PCI_VDEVICE(MELLANOX, 4122) },
1437 	{ PCI_VDEVICE(MELLANOX, 4123) },
1438 	{ PCI_VDEVICE(MELLANOX, 4124) },
1439 	{ PCI_VDEVICE(MELLANOX, 4125) },
1440 	{ PCI_VDEVICE(MELLANOX, 4126) },
1441 	{ PCI_VDEVICE(MELLANOX, 4127) },
1442 	{ PCI_VDEVICE(MELLANOX, 4128) },
1443 	{ PCI_VDEVICE(MELLANOX, 4129) },
1444 	{ PCI_VDEVICE(MELLANOX, 4130) },
1445 	{ PCI_VDEVICE(MELLANOX, 4131) },
1446 	{ PCI_VDEVICE(MELLANOX, 4132) },
1447 	{ PCI_VDEVICE(MELLANOX, 4133) },
1448 	{ PCI_VDEVICE(MELLANOX, 4134) },
1449 	{ PCI_VDEVICE(MELLANOX, 4135) },
1450 	{ PCI_VDEVICE(MELLANOX, 4136) },
1451 	{ PCI_VDEVICE(MELLANOX, 4137) },
1452 	{ PCI_VDEVICE(MELLANOX, 4138) },
1453 	{ PCI_VDEVICE(MELLANOX, 4139) },
1454 	{ PCI_VDEVICE(MELLANOX, 4140) },
1455 	{ PCI_VDEVICE(MELLANOX, 4141) },
1456 	{ PCI_VDEVICE(MELLANOX, 4142) },
1457 	{ PCI_VDEVICE(MELLANOX, 4143) },
1458 	{ PCI_VDEVICE(MELLANOX, 4144) },
1459 	{ 0, }
1460 };
1461 
1462 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1463 
1464 void mlx5_disable_device(struct mlx5_core_dev *dev)
1465 {
1466 	mlx5_pci_err_detected(dev->pdev, 0);
1467 }
1468 
1469 void mlx5_recover_device(struct mlx5_core_dev *dev)
1470 {
1471 	mlx5_pci_disable_device(dev);
1472 	if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED)
1473 		mlx5_pci_resume(dev->pdev);
1474 }
1475 
1476 struct pci_driver mlx5_core_driver = {
1477 	.name           = DRIVER_NAME,
1478 	.id_table       = mlx5_core_pci_table,
1479 	.shutdown	= shutdown_one,
1480 	.probe          = init_one,
1481 	.remove         = remove_one,
1482 	.err_handler	= &mlx5_err_handler
1483 };
1484 
1485 static int __init init(void)
1486 {
1487 	int err;
1488 
1489 	err = pci_register_driver(&mlx5_core_driver);
1490 	if (err)
1491 		goto err_debug;
1492 
1493 	err = mlx5_fwdump_init();
1494 	if (err)
1495 		goto err_fwdump;
1496 
1497  	return 0;
1498 
1499 err_fwdump:
1500 	pci_unregister_driver(&mlx5_core_driver);
1501 
1502 err_debug:
1503 	return err;
1504 }
1505 
1506 static void __exit cleanup(void)
1507 {
1508 	mlx5_fwdump_fini();
1509 	pci_unregister_driver(&mlx5_core_driver);
1510 }
1511 
1512 module_init(init);
1513 module_exit(cleanup);
1514