xref: /freebsd-src/sys/dev/mlx5/mlx5_core/mlx5_main.c (revision 0b57cec536236d46e3dba9bd041533462f33dbb7)
1 /*-
2  * Copyright (c) 2013-2019, 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 <linux/hardirq.h>
37 #include <dev/mlx5/driver.h>
38 #include <dev/mlx5/cq.h>
39 #include <dev/mlx5/qp.h>
40 #include <dev/mlx5/srq.h>
41 #include <dev/mlx5/mpfs.h>
42 #include <linux/delay.h>
43 #include <dev/mlx5/mlx5_ifc.h>
44 #include <dev/mlx5/mlx5_fpga/core.h>
45 #include <dev/mlx5/mlx5_lib/mlx5.h>
46 #include "mlx5_core.h"
47 #include "fs_core.h"
48 
49 static const char mlx5_version[] = "Mellanox Core driver "
50 	DRIVER_VERSION " (" DRIVER_RELDATE ")";
51 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
52 MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver");
53 MODULE_LICENSE("Dual BSD/GPL");
54 MODULE_DEPEND(mlx5, linuxkpi, 1, 1, 1);
55 MODULE_DEPEND(mlx5, mlxfw, 1, 1, 1);
56 MODULE_DEPEND(mlx5, firmware, 1, 1, 1);
57 MODULE_VERSION(mlx5, 1);
58 
59 SYSCTL_NODE(_hw, OID_AUTO, mlx5, CTLFLAG_RW, 0, "mlx5 hardware controls");
60 
61 int mlx5_core_debug_mask;
62 SYSCTL_INT(_hw_mlx5, OID_AUTO, debug_mask, CTLFLAG_RWTUN,
63     &mlx5_core_debug_mask, 0,
64     "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
65 
66 #define MLX5_DEFAULT_PROF	2
67 static int mlx5_prof_sel = MLX5_DEFAULT_PROF;
68 SYSCTL_INT(_hw_mlx5, OID_AUTO, prof_sel, CTLFLAG_RWTUN,
69     &mlx5_prof_sel, 0,
70     "profile selector. Valid range 0 - 2");
71 
72 static int mlx5_fast_unload_enabled = 1;
73 SYSCTL_INT(_hw_mlx5, OID_AUTO, fast_unload_enabled, CTLFLAG_RWTUN,
74     &mlx5_fast_unload_enabled, 0,
75     "Set to enable fast unload. Clear to disable.");
76 
77 #define NUMA_NO_NODE       -1
78 
79 static LIST_HEAD(intf_list);
80 static LIST_HEAD(dev_list);
81 static DEFINE_MUTEX(intf_mutex);
82 
83 struct mlx5_device_context {
84 	struct list_head	list;
85 	struct mlx5_interface  *intf;
86 	void		       *context;
87 };
88 
89 enum {
90 	MLX5_ATOMIC_REQ_MODE_BE = 0x0,
91 	MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
92 };
93 
94 static struct mlx5_profile profiles[] = {
95 	[0] = {
96 		.mask           = 0,
97 	},
98 	[1] = {
99 		.mask		= MLX5_PROF_MASK_QP_SIZE,
100 		.log_max_qp	= 12,
101 	},
102 	[2] = {
103 		.mask		= MLX5_PROF_MASK_QP_SIZE |
104 				  MLX5_PROF_MASK_MR_CACHE,
105 		.log_max_qp	= 17,
106 		.mr_cache[0]	= {
107 			.size	= 500,
108 			.limit	= 250
109 		},
110 		.mr_cache[1]	= {
111 			.size	= 500,
112 			.limit	= 250
113 		},
114 		.mr_cache[2]	= {
115 			.size	= 500,
116 			.limit	= 250
117 		},
118 		.mr_cache[3]	= {
119 			.size	= 500,
120 			.limit	= 250
121 		},
122 		.mr_cache[4]	= {
123 			.size	= 500,
124 			.limit	= 250
125 		},
126 		.mr_cache[5]	= {
127 			.size	= 500,
128 			.limit	= 250
129 		},
130 		.mr_cache[6]	= {
131 			.size	= 500,
132 			.limit	= 250
133 		},
134 		.mr_cache[7]	= {
135 			.size	= 500,
136 			.limit	= 250
137 		},
138 		.mr_cache[8]	= {
139 			.size	= 500,
140 			.limit	= 250
141 		},
142 		.mr_cache[9]	= {
143 			.size	= 500,
144 			.limit	= 250
145 		},
146 		.mr_cache[10]	= {
147 			.size	= 500,
148 			.limit	= 250
149 		},
150 		.mr_cache[11]	= {
151 			.size	= 500,
152 			.limit	= 250
153 		},
154 		.mr_cache[12]	= {
155 			.size	= 64,
156 			.limit	= 32
157 		},
158 		.mr_cache[13]	= {
159 			.size	= 32,
160 			.limit	= 16
161 		},
162 		.mr_cache[14]	= {
163 			.size	= 16,
164 			.limit	= 8
165 		},
166 	},
167 	[3] = {
168 		.mask		= MLX5_PROF_MASK_QP_SIZE,
169 		.log_max_qp	= 17,
170 	},
171 };
172 
173 static int set_dma_caps(struct pci_dev *pdev)
174 {
175 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
176 	int err;
177 
178 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
179 	if (err) {
180 		mlx5_core_warn(dev, "couldn't set 64-bit PCI DMA mask\n");
181 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
182 		if (err) {
183 			mlx5_core_err(dev, "Can't set PCI DMA mask, aborting\n");
184 			return err;
185 		}
186 	}
187 
188 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
189 	if (err) {
190 		mlx5_core_warn(dev, "couldn't set 64-bit consistent PCI DMA mask\n");
191 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
192 		if (err) {
193 			mlx5_core_err(dev, "Can't set consistent PCI DMA mask, aborting\n");
194 			return err;
195 		}
196 	}
197 
198 	dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
199 	return err;
200 }
201 
202 int mlx5_pci_read_power_status(struct mlx5_core_dev *dev,
203 			       u16 *p_power, u8 *p_status)
204 {
205 	u32 in[MLX5_ST_SZ_DW(mpein_reg)] = {};
206 	u32 out[MLX5_ST_SZ_DW(mpein_reg)] = {};
207 	int err;
208 
209 	err = mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out),
210 	    MLX5_ACCESS_REG_SUMMARY_CTRL_ID_MPEIN, 0, 0);
211 
212 	*p_status = MLX5_GET(mpein_reg, out, pwr_status);
213 	*p_power = MLX5_GET(mpein_reg, out, pci_power);
214 	return err;
215 }
216 
217 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev)
218 {
219 	struct pci_dev *pdev = dev->pdev;
220 	int err = 0;
221 
222 	mutex_lock(&dev->pci_status_mutex);
223 	if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) {
224 		err = pci_enable_device(pdev);
225 		if (!err)
226 			dev->pci_status = MLX5_PCI_STATUS_ENABLED;
227 	}
228 	mutex_unlock(&dev->pci_status_mutex);
229 
230 	return err;
231 }
232 
233 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev)
234 {
235 	struct pci_dev *pdev = dev->pdev;
236 
237 	mutex_lock(&dev->pci_status_mutex);
238 	if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) {
239 		pci_disable_device(pdev);
240 		dev->pci_status = MLX5_PCI_STATUS_DISABLED;
241 	}
242 	mutex_unlock(&dev->pci_status_mutex);
243 }
244 
245 static int request_bar(struct pci_dev *pdev)
246 {
247 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
248 	int err = 0;
249 
250 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
251 		mlx5_core_err(dev, "Missing registers BAR, aborting\n");
252 		return -ENODEV;
253 	}
254 
255 	err = pci_request_regions(pdev, DRIVER_NAME);
256 	if (err)
257 		mlx5_core_err(dev, "Couldn't get PCI resources, aborting\n");
258 
259 	return err;
260 }
261 
262 static void release_bar(struct pci_dev *pdev)
263 {
264 	pci_release_regions(pdev);
265 }
266 
267 static int mlx5_enable_msix(struct mlx5_core_dev *dev)
268 {
269 	struct mlx5_priv *priv = &dev->priv;
270 	struct mlx5_eq_table *table = &priv->eq_table;
271 	int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq);
272 	int limit = dev->msix_eqvec;
273 	int nvec = MLX5_EQ_VEC_COMP_BASE;
274 	int i;
275 
276 	if (limit > 0)
277 		nvec += limit;
278 	else
279 		nvec += MLX5_CAP_GEN(dev, num_ports) * num_online_cpus();
280 
281 	if (nvec > num_eqs)
282 		nvec = num_eqs;
283 	if (nvec > 256)
284 		nvec = 256;	/* limit of firmware API */
285 	if (nvec <= MLX5_EQ_VEC_COMP_BASE)
286 		return -ENOMEM;
287 
288 	priv->msix_arr = kzalloc(nvec * sizeof(*priv->msix_arr), GFP_KERNEL);
289 
290 	for (i = 0; i < nvec; i++)
291 		priv->msix_arr[i].entry = i;
292 
293 	nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr,
294 				     MLX5_EQ_VEC_COMP_BASE + 1, nvec);
295 	if (nvec < 0)
296 		return nvec;
297 
298 	table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE;
299 	return 0;
300 }
301 
302 static void mlx5_disable_msix(struct mlx5_core_dev *dev)
303 {
304 	struct mlx5_priv *priv = &dev->priv;
305 
306 	pci_disable_msix(dev->pdev);
307 	kfree(priv->msix_arr);
308 }
309 
310 struct mlx5_reg_host_endianess {
311 	u8	he;
312 	u8      rsvd[15];
313 };
314 
315 
316 #define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos))
317 
318 enum {
319 	MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) |
320 				MLX5_DEV_CAP_FLAG_DCT |
321 				MLX5_DEV_CAP_FLAG_DRAIN_SIGERR,
322 };
323 
324 static u16 to_fw_pkey_sz(struct mlx5_core_dev *dev, u32 size)
325 {
326 	switch (size) {
327 	case 128:
328 		return 0;
329 	case 256:
330 		return 1;
331 	case 512:
332 		return 2;
333 	case 1024:
334 		return 3;
335 	case 2048:
336 		return 4;
337 	case 4096:
338 		return 5;
339 	default:
340 		mlx5_core_warn(dev, "invalid pkey table size %d\n", size);
341 		return 0;
342 	}
343 }
344 
345 static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev,
346 				   enum mlx5_cap_type cap_type,
347 				   enum mlx5_cap_mode cap_mode)
348 {
349 	u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
350 	int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
351 	void *out, *hca_caps;
352 	u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
353 	int err;
354 
355 	memset(in, 0, sizeof(in));
356 	out = kzalloc(out_sz, GFP_KERNEL);
357 
358 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
359 	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
360 	err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz);
361 	if (err) {
362 		mlx5_core_warn(dev,
363 			       "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
364 			       cap_type, cap_mode, err);
365 		goto query_ex;
366 	}
367 
368 	hca_caps =  MLX5_ADDR_OF(query_hca_cap_out, out, capability);
369 
370 	switch (cap_mode) {
371 	case HCA_CAP_OPMOD_GET_MAX:
372 		memcpy(dev->hca_caps_max[cap_type], hca_caps,
373 		       MLX5_UN_SZ_BYTES(hca_cap_union));
374 		break;
375 	case HCA_CAP_OPMOD_GET_CUR:
376 		memcpy(dev->hca_caps_cur[cap_type], hca_caps,
377 		       MLX5_UN_SZ_BYTES(hca_cap_union));
378 		break;
379 	default:
380 		mlx5_core_warn(dev,
381 			       "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
382 			       cap_type, cap_mode);
383 		err = -EINVAL;
384 		break;
385 	}
386 query_ex:
387 	kfree(out);
388 	return err;
389 }
390 
391 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
392 {
393 	int ret;
394 
395 	ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
396 	if (ret)
397 		return ret;
398 
399 	return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
400 }
401 
402 static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz)
403 {
404 	u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)] = {0};
405 
406 	MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
407 
408 	return mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out));
409 }
410 
411 static int handle_hca_cap(struct mlx5_core_dev *dev)
412 {
413 	void *set_ctx = NULL;
414 	struct mlx5_profile *prof = dev->profile;
415 	int err = -ENOMEM;
416 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
417 	void *set_hca_cap;
418 
419 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
420 
421 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
422 	if (err)
423 		goto query_ex;
424 
425 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
426 				   capability);
427 	memcpy(set_hca_cap, dev->hca_caps_cur[MLX5_CAP_GENERAL],
428 	       MLX5_ST_SZ_BYTES(cmd_hca_cap));
429 
430 	mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
431 		      mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
432 		      128);
433 	/* we limit the size of the pkey table to 128 entries for now */
434 	MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
435 		 to_fw_pkey_sz(dev, 128));
436 
437 	if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
438 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
439 			 prof->log_max_qp);
440 
441 	/* disable cmdif checksum */
442 	MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
443 
444 	/* enable drain sigerr */
445 	MLX5_SET(cmd_hca_cap, set_hca_cap, drain_sigerr, 1);
446 
447 	MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
448 
449 	err = set_caps(dev, set_ctx, set_sz);
450 
451 query_ex:
452 	kfree(set_ctx);
453 	return err;
454 }
455 
456 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev)
457 {
458 	void *set_ctx;
459 	void *set_hca_cap;
460 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
461 	int req_endianness;
462 	int err;
463 
464 	if (MLX5_CAP_GEN(dev, atomic)) {
465 		err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
466 		if (err)
467 			return err;
468 	} else {
469 		return 0;
470 	}
471 
472 	req_endianness =
473 		MLX5_CAP_ATOMIC(dev,
474 				supported_atomic_req_8B_endianess_mode_1);
475 
476 	if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
477 		return 0;
478 
479 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
480 	if (!set_ctx)
481 		return -ENOMEM;
482 
483 	MLX5_SET(set_hca_cap_in, set_ctx, op_mod,
484 		 MLX5_SET_HCA_CAP_OP_MOD_ATOMIC << 1);
485 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
486 
487 	/* Set requestor to host endianness */
488 	MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianess_mode,
489 		 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
490 
491 	err = set_caps(dev, set_ctx, set_sz);
492 
493 	kfree(set_ctx);
494 	return err;
495 }
496 
497 static int set_hca_ctrl(struct mlx5_core_dev *dev)
498 {
499 	struct mlx5_reg_host_endianess he_in;
500 	struct mlx5_reg_host_endianess he_out;
501 	int err;
502 
503 	if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH &&
504 	    !MLX5_CAP_GEN(dev, roce))
505 		return 0;
506 
507 	memset(&he_in, 0, sizeof(he_in));
508 	he_in.he = MLX5_SET_HOST_ENDIANNESS;
509 	err = mlx5_core_access_reg(dev, &he_in,  sizeof(he_in),
510 					&he_out, sizeof(he_out),
511 					MLX5_REG_HOST_ENDIANNESS, 0, 1);
512 	return err;
513 }
514 
515 static int mlx5_core_enable_hca(struct mlx5_core_dev *dev)
516 {
517 	u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0};
518 	u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0};
519 
520 	MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
521 	return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out));
522 }
523 
524 static int mlx5_core_disable_hca(struct mlx5_core_dev *dev)
525 {
526 	u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0};
527 	u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0};
528 
529 	MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
530 	return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
531 }
532 
533 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
534 {
535 	u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0};
536 	u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0};
537 	u32 sup_issi;
538 	int err;
539 
540 	MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
541 
542 	err = mlx5_cmd_exec(dev, query_in, sizeof(query_in), query_out, sizeof(query_out));
543 	if (err) {
544 		u32 syndrome;
545 		u8 status;
546 
547 		mlx5_cmd_mbox_status(query_out, &status, &syndrome);
548 		if (status == MLX5_CMD_STAT_BAD_OP_ERR) {
549 			mlx5_core_dbg(dev, "Only ISSI 0 is supported\n");
550 			return 0;
551 		}
552 
553 		mlx5_core_err(dev, "failed to query ISSI\n");
554 		return err;
555 	}
556 
557 	sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
558 
559 	if (sup_issi & (1 << 1)) {
560 		u32 set_in[MLX5_ST_SZ_DW(set_issi_in)]	 = {0};
561 		u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0};
562 
563 		MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
564 		MLX5_SET(set_issi_in, set_in, current_issi, 1);
565 
566 		err = mlx5_cmd_exec(dev, set_in, sizeof(set_in), set_out, sizeof(set_out));
567 		if (err) {
568 			mlx5_core_err(dev, "failed to set ISSI=1 err(%d)\n", err);
569 			return err;
570 		}
571 
572 		dev->issi = 1;
573 
574 		return 0;
575 	} else if (sup_issi & (1 << 0)) {
576 		return 0;
577 	}
578 
579 	return -ENOTSUPP;
580 }
581 
582 
583 int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn, int *irqn)
584 {
585 	struct mlx5_eq_table *table = &dev->priv.eq_table;
586 	struct mlx5_eq *eq;
587 	int err = -ENOENT;
588 
589 	spin_lock(&table->lock);
590 	list_for_each_entry(eq, &table->comp_eqs_list, list) {
591 		if (eq->index == vector) {
592 			*eqn = eq->eqn;
593 			*irqn = eq->irqn;
594 			err = 0;
595 			break;
596 		}
597 	}
598 	spin_unlock(&table->lock);
599 
600 	return err;
601 }
602 EXPORT_SYMBOL(mlx5_vector2eqn);
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 	struct mlx5_eq *eq;
626 	int ncomp_vec;
627 	int nent;
628 	int err;
629 	int i;
630 
631 	INIT_LIST_HEAD(&table->comp_eqs_list);
632 	ncomp_vec = table->num_comp_vectors;
633 	nent = MLX5_COMP_EQ_SIZE;
634 	for (i = 0; i < ncomp_vec; i++) {
635 		eq = kzalloc(sizeof(*eq), GFP_KERNEL);
636 
637 		err = mlx5_create_map_eq(dev, eq,
638 					 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
639 					 &dev->priv.uuari.uars[0]);
640 		if (err) {
641 			kfree(eq);
642 			goto clean;
643 		}
644 		mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
645 		eq->index = i;
646 		spin_lock(&table->lock);
647 		list_add_tail(&eq->list, &table->comp_eqs_list);
648 		spin_unlock(&table->lock);
649 	}
650 
651 	return 0;
652 
653 clean:
654 	free_comp_eqs(dev);
655 	return err;
656 }
657 
658 static int map_bf_area(struct mlx5_core_dev *dev)
659 {
660 	resource_size_t bf_start = pci_resource_start(dev->pdev, 0);
661 	resource_size_t bf_len = pci_resource_len(dev->pdev, 0);
662 
663 	dev->priv.bf_mapping = io_mapping_create_wc(bf_start, bf_len);
664 
665 	return dev->priv.bf_mapping ? 0 : -ENOMEM;
666 }
667 
668 static void unmap_bf_area(struct mlx5_core_dev *dev)
669 {
670 	if (dev->priv.bf_mapping)
671 		io_mapping_free(dev->priv.bf_mapping);
672 }
673 
674 static inline int fw_initializing(struct mlx5_core_dev *dev)
675 {
676 	return ioread32be(&dev->iseg->initializing) >> 31;
677 }
678 
679 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili,
680 			u32 warn_time_mili)
681 {
682 	int warn = jiffies + msecs_to_jiffies(warn_time_mili);
683 	int end = jiffies + msecs_to_jiffies(max_wait_mili);
684 	int err = 0;
685 
686 	MPASS(max_wait_mili > warn_time_mili);
687 
688 	while (fw_initializing(dev) == 1) {
689 		if (time_after(jiffies, end)) {
690 			err = -EBUSY;
691 			break;
692 		}
693 		if (warn_time_mili && time_after(jiffies, warn)) {
694 			mlx5_core_warn(dev,
695 			    "Waiting for FW initialization, timeout abort in %u s\n",
696 			    (unsigned int)(jiffies_to_msecs(end - warn) / 1000));
697 			warn = jiffies + msecs_to_jiffies(warn_time_mili);
698 		}
699 		msleep(FW_INIT_WAIT_MS);
700 	}
701 
702 	if (err != 0)
703 		mlx5_core_dbg(dev, "Full initializing bit dword = 0x%x\n",
704 		    ioread32be(&dev->iseg->initializing));
705 
706 	return err;
707 }
708 
709 static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
710 {
711 	struct mlx5_device_context *dev_ctx;
712 	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
713 
714 	dev_ctx = kzalloc(sizeof(*dev_ctx), GFP_KERNEL);
715 	if (!dev_ctx)
716 		return;
717 
718 	dev_ctx->intf    = intf;
719 	CURVNET_SET_QUIET(vnet0);
720 	dev_ctx->context = intf->add(dev);
721 	CURVNET_RESTORE();
722 
723 	if (dev_ctx->context) {
724 		spin_lock_irq(&priv->ctx_lock);
725 		list_add_tail(&dev_ctx->list, &priv->ctx_list);
726 		spin_unlock_irq(&priv->ctx_lock);
727 	} else {
728 		kfree(dev_ctx);
729 	}
730 }
731 
732 static void mlx5_remove_device(struct mlx5_interface *intf, struct mlx5_priv *priv)
733 {
734 	struct mlx5_device_context *dev_ctx;
735 	struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv);
736 
737 	list_for_each_entry(dev_ctx, &priv->ctx_list, list)
738 		if (dev_ctx->intf == intf) {
739 			spin_lock_irq(&priv->ctx_lock);
740 			list_del(&dev_ctx->list);
741 			spin_unlock_irq(&priv->ctx_lock);
742 
743 			intf->remove(dev, dev_ctx->context);
744 			kfree(dev_ctx);
745 			return;
746 		}
747 }
748 
749 int
750 mlx5_register_device(struct mlx5_core_dev *dev)
751 {
752 	struct mlx5_priv *priv = &dev->priv;
753 	struct mlx5_interface *intf;
754 
755 	mutex_lock(&intf_mutex);
756 	list_add_tail(&priv->dev_list, &dev_list);
757 	list_for_each_entry(intf, &intf_list, list)
758 		mlx5_add_device(intf, priv);
759 	mutex_unlock(&intf_mutex);
760 
761 	return 0;
762 }
763 
764 void
765 mlx5_unregister_device(struct mlx5_core_dev *dev)
766 {
767 	struct mlx5_priv *priv = &dev->priv;
768 	struct mlx5_interface *intf;
769 
770 	mutex_lock(&intf_mutex);
771 	list_for_each_entry(intf, &intf_list, list)
772 		mlx5_remove_device(intf, priv);
773 	list_del(&priv->dev_list);
774 	mutex_unlock(&intf_mutex);
775 }
776 
777 int mlx5_register_interface(struct mlx5_interface *intf)
778 {
779 	struct mlx5_priv *priv;
780 
781 	if (!intf->add || !intf->remove)
782 		return -EINVAL;
783 
784 	mutex_lock(&intf_mutex);
785 	list_add_tail(&intf->list, &intf_list);
786 	list_for_each_entry(priv, &dev_list, dev_list)
787 		mlx5_add_device(intf, priv);
788 	mutex_unlock(&intf_mutex);
789 
790 	return 0;
791 }
792 EXPORT_SYMBOL(mlx5_register_interface);
793 
794 void mlx5_unregister_interface(struct mlx5_interface *intf)
795 {
796 	struct mlx5_priv *priv;
797 
798 	mutex_lock(&intf_mutex);
799 	list_for_each_entry(priv, &dev_list, dev_list)
800 		mlx5_remove_device(intf, priv);
801 	list_del(&intf->list);
802 	mutex_unlock(&intf_mutex);
803 }
804 EXPORT_SYMBOL(mlx5_unregister_interface);
805 
806 void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol)
807 {
808 	struct mlx5_priv *priv = &mdev->priv;
809 	struct mlx5_device_context *dev_ctx;
810 	unsigned long flags;
811 	void *result = NULL;
812 
813 	spin_lock_irqsave(&priv->ctx_lock, flags);
814 
815 	list_for_each_entry(dev_ctx, &mdev->priv.ctx_list, list)
816 		if ((dev_ctx->intf->protocol == protocol) &&
817 		    dev_ctx->intf->get_dev) {
818 			result = dev_ctx->intf->get_dev(dev_ctx->context);
819 			break;
820 		}
821 
822 	spin_unlock_irqrestore(&priv->ctx_lock, flags);
823 
824 	return result;
825 }
826 EXPORT_SYMBOL(mlx5_get_protocol_dev);
827 
828 static int mlx5_auto_fw_update;
829 SYSCTL_INT(_hw_mlx5, OID_AUTO, auto_fw_update, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
830     &mlx5_auto_fw_update, 0,
831     "Allow automatic firmware update on driver start");
832 static int
833 mlx5_firmware_update(struct mlx5_core_dev *dev)
834 {
835 	const struct firmware *fw;
836 	int err;
837 
838 	TUNABLE_INT_FETCH("hw.mlx5.auto_fw_update", &mlx5_auto_fw_update);
839 	if (!mlx5_auto_fw_update)
840 		return (0);
841 	fw = firmware_get("mlx5fw_mfa");
842 	if (fw) {
843 		err = mlx5_firmware_flash(dev, fw);
844 		firmware_put(fw, FIRMWARE_UNLOAD);
845 	}
846 	else
847 		return (-ENOENT);
848 
849 	return err;
850 }
851 
852 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
853 {
854 	struct pci_dev *pdev = dev->pdev;
855 	int err = 0;
856 
857 	pci_set_drvdata(dev->pdev, dev);
858 	strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN);
859 	priv->name[MLX5_MAX_NAME_LEN - 1] = 0;
860 
861 	mutex_init(&priv->pgdir_mutex);
862 	INIT_LIST_HEAD(&priv->pgdir_list);
863 	spin_lock_init(&priv->mkey_lock);
864 
865 	priv->numa_node = NUMA_NO_NODE;
866 
867 	err = mlx5_pci_enable_device(dev);
868 	if (err) {
869 		mlx5_core_err(dev, "Cannot enable PCI device, aborting\n");
870 		goto err_dbg;
871 	}
872 
873 	err = request_bar(pdev);
874 	if (err) {
875 		mlx5_core_err(dev, "error requesting BARs, aborting\n");
876 		goto err_disable;
877 	}
878 
879 	pci_set_master(pdev);
880 
881 	err = set_dma_caps(pdev);
882 	if (err) {
883 		mlx5_core_err(dev, "Failed setting DMA capabilities mask, aborting\n");
884 		goto err_clr_master;
885 	}
886 
887 	dev->iseg_base = pci_resource_start(dev->pdev, 0);
888 	dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
889 	if (!dev->iseg) {
890 		err = -ENOMEM;
891 		mlx5_core_err(dev, "Failed mapping initialization segment, aborting\n");
892 		goto err_clr_master;
893 	}
894 
895 	return 0;
896 
897 err_clr_master:
898 	release_bar(dev->pdev);
899 err_disable:
900 	mlx5_pci_disable_device(dev);
901 err_dbg:
902 	return err;
903 }
904 
905 static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
906 {
907 	iounmap(dev->iseg);
908 	release_bar(dev->pdev);
909 	mlx5_pci_disable_device(dev);
910 }
911 
912 static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv)
913 {
914 	int err;
915 
916 	err = mlx5_vsc_find_cap(dev);
917 	if (err)
918 		mlx5_core_err(dev, "Unable to find vendor specific capabilities\n");
919 
920 	err = mlx5_query_hca_caps(dev);
921 	if (err) {
922 		mlx5_core_err(dev, "query hca failed\n");
923 		goto out;
924 	}
925 
926 	err = mlx5_query_board_id(dev);
927 	if (err) {
928 		mlx5_core_err(dev, "query board id failed\n");
929 		goto out;
930 	}
931 
932 	err = mlx5_eq_init(dev);
933 	if (err) {
934 		mlx5_core_err(dev, "failed to initialize eq\n");
935 		goto out;
936 	}
937 
938 	MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock);
939 
940 	err = mlx5_init_cq_table(dev);
941 	if (err) {
942 		mlx5_core_err(dev, "failed to initialize cq table\n");
943 		goto err_eq_cleanup;
944 	}
945 
946 	mlx5_init_qp_table(dev);
947 	mlx5_init_srq_table(dev);
948 	mlx5_init_mr_table(dev);
949 
950 	mlx5_init_reserved_gids(dev);
951 	mlx5_fpga_init(dev);
952 
953 #ifdef RATELIMIT
954 	err = mlx5_init_rl_table(dev);
955 	if (err) {
956 		mlx5_core_err(dev, "Failed to init rate limiting\n");
957 		goto err_tables_cleanup;
958 	}
959 #endif
960 	return 0;
961 
962 #ifdef RATELIMIT
963 err_tables_cleanup:
964 	mlx5_cleanup_mr_table(dev);
965 	mlx5_cleanup_srq_table(dev);
966 	mlx5_cleanup_qp_table(dev);
967 	mlx5_cleanup_cq_table(dev);
968 #endif
969 
970 err_eq_cleanup:
971 	mlx5_eq_cleanup(dev);
972 
973 out:
974 	return err;
975 }
976 
977 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
978 {
979 #ifdef RATELIMIT
980 	mlx5_cleanup_rl_table(dev);
981 #endif
982 	mlx5_fpga_cleanup(dev);
983 	mlx5_cleanup_reserved_gids(dev);
984 	mlx5_cleanup_mr_table(dev);
985 	mlx5_cleanup_srq_table(dev);
986 	mlx5_cleanup_qp_table(dev);
987 	mlx5_cleanup_cq_table(dev);
988 	mlx5_eq_cleanup(dev);
989 }
990 
991 static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
992 			 bool boot)
993 {
994 	int err;
995 
996 	mutex_lock(&dev->intf_state_mutex);
997 	if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
998 		mlx5_core_warn(dev, "interface is up, NOP\n");
999 		goto out;
1000 	}
1001 
1002 	mlx5_core_dbg(dev, "firmware version: %d.%d.%d\n",
1003 	    fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev));
1004 
1005 	/*
1006 	 * On load removing any previous indication of internal error,
1007 	 * device is up
1008 	 */
1009 	dev->state = MLX5_DEVICE_STATE_UP;
1010 
1011 	/* wait for firmware to accept initialization segments configurations
1012 	*/
1013 	err = wait_fw_init(dev, FW_PRE_INIT_TIMEOUT_MILI,
1014 	    FW_INIT_WARN_MESSAGE_INTERVAL);
1015 	if (err) {
1016 		dev_err(&dev->pdev->dev,
1017 		    "Firmware over %d MS in pre-initializing state, aborting\n",
1018 		    FW_PRE_INIT_TIMEOUT_MILI);
1019 		goto out_err;
1020 	}
1021 
1022 	err = mlx5_cmd_init(dev);
1023 	if (err) {
1024 		mlx5_core_err(dev,
1025 		    "Failed initializing command interface, aborting\n");
1026 		goto out_err;
1027 	}
1028 
1029 	err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI, 0);
1030 	if (err) {
1031 		mlx5_core_err(dev,
1032 		    "Firmware over %d MS in initializing state, aborting\n",
1033 		    FW_INIT_TIMEOUT_MILI);
1034 		goto err_cmd_cleanup;
1035 	}
1036 
1037 	err = mlx5_core_enable_hca(dev);
1038 	if (err) {
1039 		mlx5_core_err(dev, "enable hca failed\n");
1040 		goto err_cmd_cleanup;
1041 	}
1042 
1043 	err = mlx5_core_set_issi(dev);
1044 	if (err) {
1045 		mlx5_core_err(dev, "failed to set issi\n");
1046 		goto err_disable_hca;
1047 	}
1048 
1049 	err = mlx5_pagealloc_start(dev);
1050 	if (err) {
1051 		mlx5_core_err(dev, "mlx5_pagealloc_start failed\n");
1052 		goto err_disable_hca;
1053 	}
1054 
1055 	err = mlx5_satisfy_startup_pages(dev, 1);
1056 	if (err) {
1057 		mlx5_core_err(dev, "failed to allocate boot pages\n");
1058 		goto err_pagealloc_stop;
1059 	}
1060 
1061 	err = set_hca_ctrl(dev);
1062 	if (err) {
1063 		mlx5_core_err(dev, "set_hca_ctrl failed\n");
1064 		goto reclaim_boot_pages;
1065 	}
1066 
1067 	err = handle_hca_cap(dev);
1068 	if (err) {
1069 		mlx5_core_err(dev, "handle_hca_cap failed\n");
1070 		goto reclaim_boot_pages;
1071 	}
1072 
1073 	err = handle_hca_cap_atomic(dev);
1074 	if (err) {
1075 		mlx5_core_err(dev, "handle_hca_cap_atomic failed\n");
1076 		goto reclaim_boot_pages;
1077 	}
1078 
1079 	err = mlx5_satisfy_startup_pages(dev, 0);
1080 	if (err) {
1081 		mlx5_core_err(dev, "failed to allocate init pages\n");
1082 		goto reclaim_boot_pages;
1083 	}
1084 
1085 	err = mlx5_cmd_init_hca(dev);
1086 	if (err) {
1087 		mlx5_core_err(dev, "init hca failed\n");
1088 		goto reclaim_boot_pages;
1089 	}
1090 
1091 	mlx5_start_health_poll(dev);
1092 
1093 	if (boot && mlx5_init_once(dev, priv)) {
1094 		mlx5_core_err(dev, "sw objs init failed\n");
1095 		goto err_stop_poll;
1096 	}
1097 
1098 	err = mlx5_enable_msix(dev);
1099 	if (err) {
1100 		mlx5_core_err(dev, "enable msix failed\n");
1101 		goto err_cleanup_once;
1102 	}
1103 
1104 	err = mlx5_alloc_uuars(dev, &priv->uuari);
1105 	if (err) {
1106 		mlx5_core_err(dev, "Failed allocating uar, aborting\n");
1107 		goto err_disable_msix;
1108 	}
1109 
1110 	err = mlx5_start_eqs(dev);
1111 	if (err) {
1112 		mlx5_core_err(dev, "Failed to start pages and async EQs\n");
1113 		goto err_free_uar;
1114 	}
1115 
1116 	err = alloc_comp_eqs(dev);
1117 	if (err) {
1118 		mlx5_core_err(dev, "Failed to alloc completion EQs\n");
1119 		goto err_stop_eqs;
1120 	}
1121 
1122 	if (map_bf_area(dev))
1123 		mlx5_core_err(dev, "Failed to map blue flame area\n");
1124 
1125 	err = mlx5_init_fs(dev);
1126 	if (err) {
1127 		mlx5_core_err(dev, "flow steering init %d\n", err);
1128 		goto err_free_comp_eqs;
1129 	}
1130 
1131 	err = mlx5_mpfs_init(dev);
1132 	if (err) {
1133 		mlx5_core_err(dev, "mpfs init failed %d\n", err);
1134 		goto err_fs;
1135 	}
1136 
1137 	err = mlx5_fpga_device_start(dev);
1138 	if (err) {
1139 		mlx5_core_err(dev, "fpga device start failed %d\n", err);
1140 		goto err_mpfs;
1141 	}
1142 
1143 	err = mlx5_register_device(dev);
1144 	if (err) {
1145 		mlx5_core_err(dev, "mlx5_register_device failed %d\n", err);
1146 		goto err_fpga;
1147 	}
1148 
1149 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1150 
1151 out:
1152 	mutex_unlock(&dev->intf_state_mutex);
1153 	return 0;
1154 
1155 err_fpga:
1156 	mlx5_fpga_device_stop(dev);
1157 
1158 err_mpfs:
1159 	mlx5_mpfs_destroy(dev);
1160 
1161 err_fs:
1162 	mlx5_cleanup_fs(dev);
1163 
1164 err_free_comp_eqs:
1165 	free_comp_eqs(dev);
1166 	unmap_bf_area(dev);
1167 
1168 err_stop_eqs:
1169 	mlx5_stop_eqs(dev);
1170 
1171 err_free_uar:
1172 	mlx5_free_uuars(dev, &priv->uuari);
1173 
1174 err_disable_msix:
1175 	mlx5_disable_msix(dev);
1176 
1177 err_cleanup_once:
1178 	if (boot)
1179 		mlx5_cleanup_once(dev);
1180 
1181 err_stop_poll:
1182 	mlx5_stop_health_poll(dev, boot);
1183 	if (mlx5_cmd_teardown_hca(dev)) {
1184 		mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n");
1185 		goto out_err;
1186 	}
1187 
1188 reclaim_boot_pages:
1189 	mlx5_reclaim_startup_pages(dev);
1190 
1191 err_pagealloc_stop:
1192 	mlx5_pagealloc_stop(dev);
1193 
1194 err_disable_hca:
1195 	mlx5_core_disable_hca(dev);
1196 
1197 err_cmd_cleanup:
1198 	mlx5_cmd_cleanup(dev);
1199 
1200 out_err:
1201 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1202 	mutex_unlock(&dev->intf_state_mutex);
1203 
1204 	return err;
1205 }
1206 
1207 static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv,
1208 			   bool cleanup)
1209 {
1210 	int err = 0;
1211 
1212 	if (cleanup)
1213 		mlx5_drain_health_recovery(dev);
1214 
1215 	mutex_lock(&dev->intf_state_mutex);
1216 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1217 		mlx5_core_warn(dev, "%s: interface is down, NOP\n", __func__);
1218                 if (cleanup)
1219                         mlx5_cleanup_once(dev);
1220 		goto out;
1221 	}
1222 
1223 	mlx5_unregister_device(dev);
1224 
1225 	mlx5_fpga_device_stop(dev);
1226 	mlx5_mpfs_destroy(dev);
1227 	mlx5_cleanup_fs(dev);
1228 	unmap_bf_area(dev);
1229 	mlx5_wait_for_reclaim_vfs_pages(dev);
1230 	free_comp_eqs(dev);
1231 	mlx5_stop_eqs(dev);
1232 	mlx5_free_uuars(dev, &priv->uuari);
1233 	mlx5_disable_msix(dev);
1234         if (cleanup)
1235                 mlx5_cleanup_once(dev);
1236 	mlx5_stop_health_poll(dev, cleanup);
1237 	err = mlx5_cmd_teardown_hca(dev);
1238 	if (err) {
1239 		mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n");
1240 		goto out;
1241 	}
1242 	mlx5_pagealloc_stop(dev);
1243 	mlx5_reclaim_startup_pages(dev);
1244 	mlx5_core_disable_hca(dev);
1245 	mlx5_cmd_cleanup(dev);
1246 
1247 out:
1248 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1249 	mutex_unlock(&dev->intf_state_mutex);
1250 	return err;
1251 }
1252 
1253 void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event,
1254 		     unsigned long param)
1255 {
1256 	struct mlx5_priv *priv = &dev->priv;
1257 	struct mlx5_device_context *dev_ctx;
1258 	unsigned long flags;
1259 
1260 	spin_lock_irqsave(&priv->ctx_lock, flags);
1261 
1262 	list_for_each_entry(dev_ctx, &priv->ctx_list, list)
1263 		if (dev_ctx->intf->event)
1264 			dev_ctx->intf->event(dev, dev_ctx->context, event, param);
1265 
1266 	spin_unlock_irqrestore(&priv->ctx_lock, flags);
1267 }
1268 
1269 struct mlx5_core_event_handler {
1270 	void (*event)(struct mlx5_core_dev *dev,
1271 		      enum mlx5_dev_event event,
1272 		      void *data);
1273 };
1274 
1275 #define	MLX5_STATS_DESC(a, b, c, d, e, ...) d, e,
1276 
1277 #define	MLX5_PORT_MODULE_ERROR_STATS(m)				\
1278 m(+1, u64, power_budget_exceeded, "power_budget", "Module Power Budget Exceeded") \
1279 m(+1, u64, long_range, "long_range", "Module Long Range for non MLNX cable/module") \
1280 m(+1, u64, bus_stuck, "bus_stuck", "Module Bus stuck(I2C or data shorted)") \
1281 m(+1, u64, no_eeprom, "no_eeprom", "No EEPROM/retry timeout") \
1282 m(+1, u64, enforce_part_number, "enforce_part_number", "Module Enforce part number list") \
1283 m(+1, u64, unknown_id, "unknown_id", "Module Unknown identifier") \
1284 m(+1, u64, high_temp, "high_temp", "Module High Temperature") \
1285 m(+1, u64, cable_shorted, "cable_shorted", "Module Cable is shorted")
1286 
1287 static const char *mlx5_pme_err_desc[] = {
1288 	MLX5_PORT_MODULE_ERROR_STATS(MLX5_STATS_DESC)
1289 };
1290 
1291 static int init_one(struct pci_dev *pdev,
1292 		    const struct pci_device_id *id)
1293 {
1294 	struct mlx5_core_dev *dev;
1295 	struct mlx5_priv *priv;
1296 	device_t bsddev = pdev->dev.bsddev;
1297 	int i,err;
1298 	struct sysctl_oid *pme_sysctl_node;
1299 	struct sysctl_oid *pme_err_sysctl_node;
1300 
1301 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
1302 	priv = &dev->priv;
1303 	if (id)
1304 		priv->pci_dev_data = id->driver_data;
1305 
1306 	if (mlx5_prof_sel < 0 || mlx5_prof_sel >= ARRAY_SIZE(profiles)) {
1307 		device_printf(bsddev,
1308 		    "WARN: selected profile out of range, selecting default (%d)\n",
1309 		    MLX5_DEFAULT_PROF);
1310 		mlx5_prof_sel = MLX5_DEFAULT_PROF;
1311 	}
1312 	dev->profile = &profiles[mlx5_prof_sel];
1313 	dev->pdev = pdev;
1314 	dev->event = mlx5_core_event;
1315 
1316 	/* Set desc */
1317 	device_set_desc(bsddev, mlx5_version);
1318 
1319 	sysctl_ctx_init(&dev->sysctl_ctx);
1320 	SYSCTL_ADD_INT(&dev->sysctl_ctx,
1321 	    SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)),
1322 	    OID_AUTO, "msix_eqvec", CTLFLAG_RDTUN, &dev->msix_eqvec, 0,
1323 	    "Maximum number of MSIX event queue vectors, if set");
1324 	SYSCTL_ADD_INT(&dev->sysctl_ctx,
1325 	    SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)),
1326 	    OID_AUTO, "power_status", CTLFLAG_RD, &dev->pwr_status, 0,
1327 	    "0:Invalid 1:Sufficient 2:Insufficient");
1328 	SYSCTL_ADD_INT(&dev->sysctl_ctx,
1329 	    SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)),
1330 	    OID_AUTO, "power_value", CTLFLAG_RD, &dev->pwr_value, 0,
1331 	    "Current power value in Watts");
1332 
1333 	pme_sysctl_node = SYSCTL_ADD_NODE(&dev->sysctl_ctx,
1334 	    SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)),
1335 	    OID_AUTO, "pme_stats", CTLFLAG_RD, NULL,
1336 	    "Port module event statistics");
1337 	if (pme_sysctl_node == NULL) {
1338 		err = -ENOMEM;
1339 		goto clean_sysctl_ctx;
1340 	}
1341 	pme_err_sysctl_node = SYSCTL_ADD_NODE(&dev->sysctl_ctx,
1342 	    SYSCTL_CHILDREN(pme_sysctl_node),
1343 	    OID_AUTO, "errors", CTLFLAG_RD, NULL,
1344 	    "Port module event error statistics");
1345 	if (pme_err_sysctl_node == NULL) {
1346 		err = -ENOMEM;
1347 		goto clean_sysctl_ctx;
1348 	}
1349 	SYSCTL_ADD_U64(&dev->sysctl_ctx,
1350 	    SYSCTL_CHILDREN(pme_sysctl_node), OID_AUTO,
1351 	    "module_plug", CTLFLAG_RD | CTLFLAG_MPSAFE,
1352 	    &dev->priv.pme_stats.status_counters[MLX5_MODULE_STATUS_PLUGGED_ENABLED],
1353 	    0, "Number of time module plugged");
1354 	SYSCTL_ADD_U64(&dev->sysctl_ctx,
1355 	    SYSCTL_CHILDREN(pme_sysctl_node), OID_AUTO,
1356 	    "module_unplug", CTLFLAG_RD | CTLFLAG_MPSAFE,
1357 	    &dev->priv.pme_stats.status_counters[MLX5_MODULE_STATUS_UNPLUGGED],
1358 	    0, "Number of time module unplugged");
1359 	for (i = 0 ; i < MLX5_MODULE_EVENT_ERROR_NUM; i++) {
1360 		SYSCTL_ADD_U64(&dev->sysctl_ctx,
1361 		    SYSCTL_CHILDREN(pme_err_sysctl_node), OID_AUTO,
1362 		    mlx5_pme_err_desc[2 * i], CTLFLAG_RD | CTLFLAG_MPSAFE,
1363 		    &dev->priv.pme_stats.error_counters[i],
1364 		    0, mlx5_pme_err_desc[2 * i + 1]);
1365 	}
1366 
1367 
1368 	INIT_LIST_HEAD(&priv->ctx_list);
1369 	spin_lock_init(&priv->ctx_lock);
1370 	mutex_init(&dev->pci_status_mutex);
1371 	mutex_init(&dev->intf_state_mutex);
1372 	mtx_init(&dev->dump_lock, "mlx5dmp", NULL, MTX_DEF | MTX_NEW);
1373 	err = mlx5_pci_init(dev, priv);
1374 	if (err) {
1375 		mlx5_core_err(dev, "mlx5_pci_init failed %d\n", err);
1376 		goto clean_dev;
1377 	}
1378 
1379 	err = mlx5_health_init(dev);
1380 	if (err) {
1381 		mlx5_core_err(dev, "mlx5_health_init failed %d\n", err);
1382 		goto close_pci;
1383 	}
1384 
1385 	mlx5_pagealloc_init(dev);
1386 
1387 	err = mlx5_load_one(dev, priv, true);
1388 	if (err) {
1389 		mlx5_core_err(dev, "mlx5_load_one failed %d\n", err);
1390 		goto clean_health;
1391 	}
1392 
1393 	mlx5_fwdump_prep(dev);
1394 
1395 	mlx5_firmware_update(dev);
1396 
1397 	pci_save_state(bsddev);
1398 	return 0;
1399 
1400 clean_health:
1401 	mlx5_pagealloc_cleanup(dev);
1402 	mlx5_health_cleanup(dev);
1403 close_pci:
1404 	mlx5_pci_close(dev, priv);
1405 clean_dev:
1406 	mtx_destroy(&dev->dump_lock);
1407 clean_sysctl_ctx:
1408 	sysctl_ctx_free(&dev->sysctl_ctx);
1409 	kfree(dev);
1410 	return err;
1411 }
1412 
1413 static void remove_one(struct pci_dev *pdev)
1414 {
1415 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1416 	struct mlx5_priv *priv = &dev->priv;
1417 
1418 	if (mlx5_unload_one(dev, priv, true)) {
1419 		mlx5_core_err(dev, "mlx5_unload_one failed\n");
1420 		mlx5_health_cleanup(dev);
1421 		return;
1422 	}
1423 
1424 	mlx5_pagealloc_cleanup(dev);
1425 	mlx5_health_cleanup(dev);
1426 	mlx5_fwdump_clean(dev);
1427 	mlx5_pci_close(dev, priv);
1428 	mtx_destroy(&dev->dump_lock);
1429 	pci_set_drvdata(pdev, NULL);
1430 	sysctl_ctx_free(&dev->sysctl_ctx);
1431 	kfree(dev);
1432 }
1433 
1434 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
1435 					      pci_channel_state_t state)
1436 {
1437 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1438 	struct mlx5_priv *priv = &dev->priv;
1439 
1440 	mlx5_core_info(dev, "%s was called\n", __func__);
1441 	mlx5_enter_error_state(dev, false);
1442 	mlx5_unload_one(dev, priv, false);
1443 
1444 	if (state) {
1445 		mlx5_drain_health_wq(dev);
1446 		mlx5_pci_disable_device(dev);
1447 	}
1448 
1449 	return state == pci_channel_io_perm_failure ?
1450 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1451 }
1452 
1453 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
1454 {
1455 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1456 	int err = 0;
1457 
1458 	mlx5_core_info(dev,"%s was called\n", __func__);
1459 
1460 	err = mlx5_pci_enable_device(dev);
1461 	if (err) {
1462 		mlx5_core_err(dev, "mlx5_pci_enable_device failed with error code: %d\n"
1463 			,err);
1464 		return PCI_ERS_RESULT_DISCONNECT;
1465 	}
1466 	pci_set_master(pdev);
1467 	pci_set_powerstate(pdev->dev.bsddev, PCI_POWERSTATE_D0);
1468 	pci_restore_state(pdev->dev.bsddev);
1469 	pci_save_state(pdev->dev.bsddev);
1470 
1471 	return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
1472 }
1473 
1474 /* wait for the device to show vital signs. For now we check
1475  * that we can read the device ID and that the health buffer
1476  * shows a non zero value which is different than 0xffffffff
1477  */
1478 static void wait_vital(struct pci_dev *pdev)
1479 {
1480 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1481 	struct mlx5_core_health *health = &dev->priv.health;
1482 	const int niter = 100;
1483 	u32 count;
1484 	u16 did;
1485 	int i;
1486 
1487 	/* Wait for firmware to be ready after reset */
1488 	msleep(1000);
1489 	for (i = 0; i < niter; i++) {
1490 		if (pci_read_config_word(pdev, 2, &did)) {
1491 			mlx5_core_warn(dev, "failed reading config word\n");
1492 			break;
1493 		}
1494 		if (did == pdev->device) {
1495 			mlx5_core_info(dev,
1496 			    "device ID correctly read after %d iterations\n", i);
1497 			break;
1498 		}
1499 		msleep(50);
1500 	}
1501 	if (i == niter)
1502 		mlx5_core_warn(dev, "could not read device ID\n");
1503 
1504 	for (i = 0; i < niter; i++) {
1505 		count = ioread32be(health->health_counter);
1506 		if (count && count != 0xffffffff) {
1507 			mlx5_core_info(dev,
1508 			"Counter value 0x%x after %d iterations\n", count, i);
1509 			break;
1510 		}
1511 		msleep(50);
1512 	}
1513 
1514 	if (i == niter)
1515 		mlx5_core_warn(dev, "could not read device ID\n");
1516 }
1517 
1518 static void mlx5_pci_resume(struct pci_dev *pdev)
1519 {
1520 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1521 	struct mlx5_priv *priv = &dev->priv;
1522 	int err;
1523 
1524 	mlx5_core_info(dev,"%s was called\n", __func__);
1525 
1526 	wait_vital(pdev);
1527 
1528 	err = mlx5_load_one(dev, priv, false);
1529 	if (err)
1530 		mlx5_core_err(dev,
1531 		    "mlx5_load_one failed with error code: %d\n" ,err);
1532 	else
1533 		mlx5_core_info(dev,"device recovered\n");
1534 }
1535 
1536 static const struct pci_error_handlers mlx5_err_handler = {
1537 	.error_detected = mlx5_pci_err_detected,
1538 	.slot_reset	= mlx5_pci_slot_reset,
1539 	.resume		= mlx5_pci_resume
1540 };
1541 
1542 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
1543 {
1544 	bool fast_teardown, force_teardown;
1545 	int err;
1546 
1547 	if (!mlx5_fast_unload_enabled) {
1548 		mlx5_core_dbg(dev, "fast unload is disabled by user\n");
1549 		return -EOPNOTSUPP;
1550 	}
1551 
1552 	fast_teardown = MLX5_CAP_GEN(dev, fast_teardown);
1553 	force_teardown = MLX5_CAP_GEN(dev, force_teardown);
1554 
1555 	mlx5_core_dbg(dev, "force teardown firmware support=%d\n", force_teardown);
1556 	mlx5_core_dbg(dev, "fast teardown firmware support=%d\n", fast_teardown);
1557 
1558 	if (!fast_teardown && !force_teardown)
1559 		return -EOPNOTSUPP;
1560 
1561 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1562 		mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
1563 		return -EAGAIN;
1564 	}
1565 
1566 	/* Panic tear down fw command will stop the PCI bus communication
1567 	 * with the HCA, so the health polll is no longer needed.
1568 	 */
1569 	mlx5_drain_health_wq(dev);
1570 	mlx5_stop_health_poll(dev, false);
1571 
1572 	err = mlx5_cmd_fast_teardown_hca(dev);
1573 	if (!err)
1574 		goto done;
1575 
1576 	err = mlx5_cmd_force_teardown_hca(dev);
1577 	if (!err)
1578 		goto done;
1579 
1580 	mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", err);
1581 	mlx5_start_health_poll(dev);
1582 	return err;
1583 done:
1584 	mlx5_enter_error_state(dev, true);
1585 	return 0;
1586 }
1587 
1588 static void mlx5_disable_interrupts(struct mlx5_core_dev *mdev)
1589 {
1590 	int nvec = mdev->priv.eq_table.num_comp_vectors + MLX5_EQ_VEC_COMP_BASE;
1591 	int x;
1592 
1593 	mdev->priv.disable_irqs = 1;
1594 
1595 	/* wait for all IRQ handlers to finish processing */
1596 	for (x = 0; x != nvec; x++)
1597 		synchronize_irq(mdev->priv.msix_arr[x].vector);
1598 }
1599 
1600 static void shutdown_one(struct pci_dev *pdev)
1601 {
1602 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1603 	struct mlx5_priv *priv = &dev->priv;
1604 	int err;
1605 
1606 	/* enter polling mode */
1607 	mlx5_cmd_use_polling(dev);
1608 
1609 	set_bit(MLX5_INTERFACE_STATE_TEARDOWN, &dev->intf_state);
1610 
1611 	/* disable all interrupts */
1612 	mlx5_disable_interrupts(dev);
1613 
1614 	err = mlx5_try_fast_unload(dev);
1615 	if (err)
1616 	        mlx5_unload_one(dev, priv, false);
1617 	mlx5_pci_disable_device(dev);
1618 }
1619 
1620 static const struct pci_device_id mlx5_core_pci_table[] = {
1621 	{ PCI_VDEVICE(MELLANOX, 4113) }, /* Connect-IB */
1622 	{ PCI_VDEVICE(MELLANOX, 4114) }, /* Connect-IB VF */
1623 	{ PCI_VDEVICE(MELLANOX, 4115) }, /* ConnectX-4 */
1624 	{ PCI_VDEVICE(MELLANOX, 4116) }, /* ConnectX-4 VF */
1625 	{ PCI_VDEVICE(MELLANOX, 4117) }, /* ConnectX-4LX */
1626 	{ PCI_VDEVICE(MELLANOX, 4118) }, /* ConnectX-4LX VF */
1627 	{ PCI_VDEVICE(MELLANOX, 4119) }, /* ConnectX-5 */
1628 	{ PCI_VDEVICE(MELLANOX, 4120) }, /* ConnectX-5 VF */
1629 	{ PCI_VDEVICE(MELLANOX, 4121) },
1630 	{ PCI_VDEVICE(MELLANOX, 4122) },
1631 	{ PCI_VDEVICE(MELLANOX, 4123) },
1632 	{ PCI_VDEVICE(MELLANOX, 4124) },
1633 	{ PCI_VDEVICE(MELLANOX, 4125) },
1634 	{ PCI_VDEVICE(MELLANOX, 4126) },
1635 	{ PCI_VDEVICE(MELLANOX, 4127) },
1636 	{ PCI_VDEVICE(MELLANOX, 4128) },
1637 	{ PCI_VDEVICE(MELLANOX, 4129) },
1638 	{ PCI_VDEVICE(MELLANOX, 4130) },
1639 	{ PCI_VDEVICE(MELLANOX, 4131) },
1640 	{ PCI_VDEVICE(MELLANOX, 4132) },
1641 	{ PCI_VDEVICE(MELLANOX, 4133) },
1642 	{ PCI_VDEVICE(MELLANOX, 4134) },
1643 	{ PCI_VDEVICE(MELLANOX, 4135) },
1644 	{ PCI_VDEVICE(MELLANOX, 4136) },
1645 	{ PCI_VDEVICE(MELLANOX, 4137) },
1646 	{ PCI_VDEVICE(MELLANOX, 4138) },
1647 	{ PCI_VDEVICE(MELLANOX, 4139) },
1648 	{ PCI_VDEVICE(MELLANOX, 4140) },
1649 	{ PCI_VDEVICE(MELLANOX, 4141) },
1650 	{ PCI_VDEVICE(MELLANOX, 4142) },
1651 	{ PCI_VDEVICE(MELLANOX, 4143) },
1652 	{ PCI_VDEVICE(MELLANOX, 4144) },
1653 	{ 0, }
1654 };
1655 
1656 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
1657 
1658 void mlx5_disable_device(struct mlx5_core_dev *dev)
1659 {
1660 	mlx5_pci_err_detected(dev->pdev, 0);
1661 }
1662 
1663 void mlx5_recover_device(struct mlx5_core_dev *dev)
1664 {
1665 	mlx5_pci_disable_device(dev);
1666 	if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED)
1667 		mlx5_pci_resume(dev->pdev);
1668 }
1669 
1670 struct pci_driver mlx5_core_driver = {
1671 	.name           = DRIVER_NAME,
1672 	.id_table       = mlx5_core_pci_table,
1673 	.shutdown	= shutdown_one,
1674 	.probe          = init_one,
1675 	.remove         = remove_one,
1676 	.err_handler	= &mlx5_err_handler
1677 };
1678 
1679 static int __init init(void)
1680 {
1681 	int err;
1682 
1683 	err = pci_register_driver(&mlx5_core_driver);
1684 	if (err)
1685 		goto err_debug;
1686 
1687 	err = mlx5_ctl_init();
1688 	if (err)
1689 		goto err_ctl;
1690 
1691  	return 0;
1692 
1693 err_ctl:
1694 	pci_unregister_driver(&mlx5_core_driver);
1695 
1696 err_debug:
1697 	return err;
1698 }
1699 
1700 static void __exit cleanup(void)
1701 {
1702 	mlx5_ctl_fini();
1703 	pci_unregister_driver(&mlx5_core_driver);
1704 }
1705 
1706 module_init(init);
1707 module_exit(cleanup);
1708