xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/main.c (revision f7f0adfe64de08803990dc4cbecd2849c04e314a)
1 /*
2  * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 
33 #include <linux/highmem.h>
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/errno.h>
37 #include <linux/pci.h>
38 #include <linux/dma-mapping.h>
39 #include <linux/slab.h>
40 #include <linux/interrupt.h>
41 #include <linux/delay.h>
42 #include <linux/mlx5/driver.h>
43 #include <linux/mlx5/cq.h>
44 #include <linux/mlx5/qp.h>
45 #include <linux/debugfs.h>
46 #include <linux/kmod.h>
47 #include <linux/mlx5/mlx5_ifc.h>
48 #include <linux/mlx5/vport.h>
49 #include <linux/version.h>
50 #include <net/devlink.h>
51 #include "mlx5_core.h"
52 #include "lib/eq.h"
53 #include "fs_core.h"
54 #include "lib/mpfs.h"
55 #include "eswitch.h"
56 #include "devlink.h"
57 #include "fw_reset.h"
58 #include "lib/mlx5.h"
59 #include "lib/tout.h"
60 #include "fpga/core.h"
61 #include "en_accel/ipsec.h"
62 #include "lib/clock.h"
63 #include "lib/vxlan.h"
64 #include "lib/geneve.h"
65 #include "lib/devcom.h"
66 #include "lib/pci_vsc.h"
67 #include "diag/fw_tracer.h"
68 #include "ecpf.h"
69 #include "lib/hv_vhca.h"
70 #include "diag/rsc_dump.h"
71 #include "sf/vhca_event.h"
72 #include "sf/dev/dev.h"
73 #include "sf/sf.h"
74 #include "mlx5_irq.h"
75 #include "hwmon.h"
76 #include "lag/lag.h"
77 
78 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
79 MODULE_DESCRIPTION("Mellanox 5th generation network adapters (ConnectX series) core driver");
80 MODULE_LICENSE("Dual BSD/GPL");
81 
82 unsigned int mlx5_core_debug_mask;
83 module_param_named(debug_mask, mlx5_core_debug_mask, uint, 0644);
84 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
85 
86 static unsigned int prof_sel = MLX5_DEFAULT_PROF;
87 module_param_named(prof_sel, prof_sel, uint, 0444);
88 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
89 
90 static u32 sw_owner_id[4];
91 #define MAX_SW_VHCA_ID (BIT(__mlx5_bit_sz(cmd_hca_cap_2, sw_vhca_id)) - 1)
92 static DEFINE_IDA(sw_vhca_ida);
93 
94 enum {
95 	MLX5_ATOMIC_REQ_MODE_BE = 0x0,
96 	MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
97 };
98 
99 #define LOG_MAX_SUPPORTED_QPS 0xff
100 
101 static struct mlx5_profile profile[] = {
102 	[0] = {
103 		.mask           = 0,
104 		.num_cmd_caches = MLX5_NUM_COMMAND_CACHES,
105 	},
106 	[1] = {
107 		.mask		= MLX5_PROF_MASK_QP_SIZE,
108 		.log_max_qp	= 12,
109 		.num_cmd_caches = MLX5_NUM_COMMAND_CACHES,
110 
111 	},
112 	[2] = {
113 		.mask		= MLX5_PROF_MASK_QP_SIZE |
114 				  MLX5_PROF_MASK_MR_CACHE,
115 		.log_max_qp	= LOG_MAX_SUPPORTED_QPS,
116 		.num_cmd_caches = MLX5_NUM_COMMAND_CACHES,
117 		.mr_cache[0]	= {
118 			.size	= 500,
119 			.limit	= 250
120 		},
121 		.mr_cache[1]	= {
122 			.size	= 500,
123 			.limit	= 250
124 		},
125 		.mr_cache[2]	= {
126 			.size	= 500,
127 			.limit	= 250
128 		},
129 		.mr_cache[3]	= {
130 			.size	= 500,
131 			.limit	= 250
132 		},
133 		.mr_cache[4]	= {
134 			.size	= 500,
135 			.limit	= 250
136 		},
137 		.mr_cache[5]	= {
138 			.size	= 500,
139 			.limit	= 250
140 		},
141 		.mr_cache[6]	= {
142 			.size	= 500,
143 			.limit	= 250
144 		},
145 		.mr_cache[7]	= {
146 			.size	= 500,
147 			.limit	= 250
148 		},
149 		.mr_cache[8]	= {
150 			.size	= 500,
151 			.limit	= 250
152 		},
153 		.mr_cache[9]	= {
154 			.size	= 500,
155 			.limit	= 250
156 		},
157 		.mr_cache[10]	= {
158 			.size	= 500,
159 			.limit	= 250
160 		},
161 		.mr_cache[11]	= {
162 			.size	= 500,
163 			.limit	= 250
164 		},
165 		.mr_cache[12]	= {
166 			.size	= 64,
167 			.limit	= 32
168 		},
169 		.mr_cache[13]	= {
170 			.size	= 32,
171 			.limit	= 16
172 		},
173 		.mr_cache[14]	= {
174 			.size	= 16,
175 			.limit	= 8
176 		},
177 		.mr_cache[15]	= {
178 			.size	= 8,
179 			.limit	= 4
180 		},
181 	},
182 	[3] = {
183 		.mask		= MLX5_PROF_MASK_QP_SIZE,
184 		.log_max_qp	= LOG_MAX_SUPPORTED_QPS,
185 		.num_cmd_caches = 0,
186 	},
187 };
188 
189 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili,
190 			u32 warn_time_mili, const char *init_state)
191 {
192 	unsigned long warn = jiffies + msecs_to_jiffies(warn_time_mili);
193 	unsigned long end = jiffies + msecs_to_jiffies(max_wait_mili);
194 	u32 fw_initializing;
195 
196 	do {
197 		fw_initializing = ioread32be(&dev->iseg->initializing);
198 		if (!(fw_initializing >> 31))
199 			break;
200 		if (time_after(jiffies, end)) {
201 			mlx5_core_err(dev, "Firmware over %u MS in %s state, aborting\n",
202 				      max_wait_mili, init_state);
203 			return -ETIMEDOUT;
204 		}
205 		if (test_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state)) {
206 			mlx5_core_warn(dev, "device is being removed, stop waiting for FW %s\n",
207 				       init_state);
208 			return -ENODEV;
209 		}
210 		if (warn_time_mili && time_after(jiffies, warn)) {
211 			mlx5_core_warn(dev, "Waiting for FW %s, timeout abort in %ds (0x%x)\n",
212 				       init_state, jiffies_to_msecs(end - warn) / 1000,
213 				       fw_initializing);
214 			warn = jiffies + msecs_to_jiffies(warn_time_mili);
215 		}
216 		msleep(mlx5_tout_ms(dev, FW_PRE_INIT_WAIT));
217 	} while (true);
218 
219 	return 0;
220 }
221 
222 static void mlx5_set_driver_version(struct mlx5_core_dev *dev)
223 {
224 	int driver_ver_sz = MLX5_FLD_SZ_BYTES(set_driver_version_in,
225 					      driver_version);
226 	u8 in[MLX5_ST_SZ_BYTES(set_driver_version_in)] = {};
227 	char *string;
228 
229 	if (!MLX5_CAP_GEN(dev, driver_version))
230 		return;
231 
232 	string = MLX5_ADDR_OF(set_driver_version_in, in, driver_version);
233 
234 	snprintf(string, driver_ver_sz, "Linux,%s,%u.%u.%u",
235 		 KBUILD_MODNAME, LINUX_VERSION_MAJOR,
236 		 LINUX_VERSION_PATCHLEVEL, LINUX_VERSION_SUBLEVEL);
237 
238 	/*Send the command*/
239 	MLX5_SET(set_driver_version_in, in, opcode,
240 		 MLX5_CMD_OP_SET_DRIVER_VERSION);
241 
242 	mlx5_cmd_exec_in(dev, set_driver_version, in);
243 }
244 
245 static int set_dma_caps(struct pci_dev *pdev)
246 {
247 	int err;
248 
249 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
250 	if (err) {
251 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
252 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
253 		if (err) {
254 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
255 			return err;
256 		}
257 	}
258 
259 	dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
260 	return err;
261 }
262 
263 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev)
264 {
265 	struct pci_dev *pdev = dev->pdev;
266 	int err = 0;
267 
268 	mutex_lock(&dev->pci_status_mutex);
269 	if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) {
270 		err = pci_enable_device(pdev);
271 		if (!err)
272 			dev->pci_status = MLX5_PCI_STATUS_ENABLED;
273 	}
274 	mutex_unlock(&dev->pci_status_mutex);
275 
276 	return err;
277 }
278 
279 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev)
280 {
281 	struct pci_dev *pdev = dev->pdev;
282 
283 	mutex_lock(&dev->pci_status_mutex);
284 	if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) {
285 		pci_disable_device(pdev);
286 		dev->pci_status = MLX5_PCI_STATUS_DISABLED;
287 	}
288 	mutex_unlock(&dev->pci_status_mutex);
289 }
290 
291 static int request_bar(struct pci_dev *pdev)
292 {
293 	int err = 0;
294 
295 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
296 		dev_err(&pdev->dev, "Missing registers BAR, aborting\n");
297 		return -ENODEV;
298 	}
299 
300 	err = pci_request_regions(pdev, KBUILD_MODNAME);
301 	if (err)
302 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
303 
304 	return err;
305 }
306 
307 static void release_bar(struct pci_dev *pdev)
308 {
309 	pci_release_regions(pdev);
310 }
311 
312 struct mlx5_reg_host_endianness {
313 	u8	he;
314 	u8      rsvd[15];
315 };
316 
317 static u16 to_fw_pkey_sz(struct mlx5_core_dev *dev, u32 size)
318 {
319 	switch (size) {
320 	case 128:
321 		return 0;
322 	case 256:
323 		return 1;
324 	case 512:
325 		return 2;
326 	case 1024:
327 		return 3;
328 	case 2048:
329 		return 4;
330 	case 4096:
331 		return 5;
332 	default:
333 		mlx5_core_warn(dev, "invalid pkey table size %d\n", size);
334 		return 0;
335 	}
336 }
337 
338 void mlx5_core_uplink_netdev_set(struct mlx5_core_dev *dev, struct net_device *netdev)
339 {
340 	mutex_lock(&dev->mlx5e_res.uplink_netdev_lock);
341 	dev->mlx5e_res.uplink_netdev = netdev;
342 	mlx5_blocking_notifier_call_chain(dev, MLX5_DRIVER_EVENT_UPLINK_NETDEV,
343 					  netdev);
344 	mutex_unlock(&dev->mlx5e_res.uplink_netdev_lock);
345 }
346 
347 void mlx5_core_uplink_netdev_event_replay(struct mlx5_core_dev *dev)
348 {
349 	mutex_lock(&dev->mlx5e_res.uplink_netdev_lock);
350 	mlx5_blocking_notifier_call_chain(dev, MLX5_DRIVER_EVENT_UPLINK_NETDEV,
351 					  dev->mlx5e_res.uplink_netdev);
352 	mutex_unlock(&dev->mlx5e_res.uplink_netdev_lock);
353 }
354 EXPORT_SYMBOL(mlx5_core_uplink_netdev_event_replay);
355 
356 void mlx5_core_mp_event_replay(struct mlx5_core_dev *dev, u32 event, void *data)
357 {
358 	mlx5_blocking_notifier_call_chain(dev, event, data);
359 }
360 EXPORT_SYMBOL(mlx5_core_mp_event_replay);
361 
362 int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type,
363 			    enum mlx5_cap_mode cap_mode)
364 {
365 	u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
366 	int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
367 	void *out, *hca_caps;
368 	u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
369 	int err;
370 
371 	if (WARN_ON(!dev->caps.hca[cap_type]))
372 		/* this cap_type must be added to mlx5_hca_caps_alloc() */
373 		return -EINVAL;
374 
375 	memset(in, 0, sizeof(in));
376 	out = kzalloc(out_sz, GFP_KERNEL);
377 	if (!out)
378 		return -ENOMEM;
379 
380 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
381 	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
382 	err = mlx5_cmd_exec_inout(dev, query_hca_cap, in, out);
383 	if (err) {
384 		mlx5_core_warn(dev,
385 			       "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
386 			       cap_type, cap_mode, err);
387 		goto query_ex;
388 	}
389 
390 	hca_caps =  MLX5_ADDR_OF(query_hca_cap_out, out, capability);
391 
392 	switch (cap_mode) {
393 	case HCA_CAP_OPMOD_GET_MAX:
394 		memcpy(dev->caps.hca[cap_type]->max, hca_caps,
395 		       MLX5_UN_SZ_BYTES(hca_cap_union));
396 		break;
397 	case HCA_CAP_OPMOD_GET_CUR:
398 		memcpy(dev->caps.hca[cap_type]->cur, hca_caps,
399 		       MLX5_UN_SZ_BYTES(hca_cap_union));
400 		break;
401 	default:
402 		mlx5_core_warn(dev,
403 			       "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
404 			       cap_type, cap_mode);
405 		err = -EINVAL;
406 		break;
407 	}
408 query_ex:
409 	kfree(out);
410 	return err;
411 }
412 
413 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
414 {
415 	int ret;
416 
417 	ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
418 	if (ret)
419 		return ret;
420 	return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
421 }
422 
423 static int set_caps(struct mlx5_core_dev *dev, void *in, int opmod)
424 {
425 	MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
426 	MLX5_SET(set_hca_cap_in, in, op_mod, opmod << 1);
427 	return mlx5_cmd_exec_in(dev, set_hca_cap, in);
428 }
429 
430 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev, void *set_ctx)
431 {
432 	void *set_hca_cap;
433 	int req_endianness;
434 	int err;
435 
436 	if (!MLX5_CAP_GEN(dev, atomic))
437 		return 0;
438 
439 	err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
440 	if (err)
441 		return err;
442 
443 	req_endianness =
444 		MLX5_CAP_ATOMIC(dev,
445 				supported_atomic_req_8B_endianness_mode_1);
446 
447 	if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
448 		return 0;
449 
450 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
451 
452 	/* Set requestor to host endianness */
453 	MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianness_mode,
454 		 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
455 
456 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ATOMIC);
457 }
458 
459 static int handle_hca_cap_odp(struct mlx5_core_dev *dev, void *set_ctx)
460 {
461 	bool do_set = false, mem_page_fault = false;
462 	void *set_hca_cap;
463 	int err;
464 
465 	if (!IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) ||
466 	    !MLX5_CAP_GEN(dev, pg))
467 		return 0;
468 
469 	err = mlx5_core_get_caps(dev, MLX5_CAP_ODP);
470 	if (err)
471 		return err;
472 
473 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
474 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_ODP]->cur,
475 	       MLX5_ST_SZ_BYTES(odp_cap));
476 
477 	/* For best performance, enable memory scheme ODP only when
478 	 * it has page prefetch enabled.
479 	 */
480 	if (MLX5_CAP_ODP_MAX(dev, mem_page_fault) &&
481 	    MLX5_CAP_ODP_MAX(dev, memory_page_fault_scheme_cap.page_prefetch)) {
482 		mem_page_fault = true;
483 		do_set = true;
484 		MLX5_SET(odp_cap, set_hca_cap, mem_page_fault, mem_page_fault);
485 		goto set;
486 	}
487 
488 #define ODP_CAP_SET_MAX(dev, field)                                            \
489 	do {                                                                   \
490 		u32 _res = MLX5_CAP_ODP_MAX(dev, field);                       \
491 		if (_res) {                                                    \
492 			do_set = true;                                         \
493 			MLX5_SET(odp_cap, set_hca_cap, field, _res);           \
494 		}                                                              \
495 	} while (0)
496 
497 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.ud_odp_caps.srq_receive);
498 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.rc_odp_caps.srq_receive);
499 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.srq_receive);
500 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.send);
501 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.receive);
502 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.write);
503 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.read);
504 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.atomic);
505 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.srq_receive);
506 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.send);
507 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.receive);
508 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.write);
509 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.read);
510 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.atomic);
511 
512 set:
513 	if (do_set)
514 		err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ODP);
515 
516 	mlx5_core_dbg(dev, "Using ODP %s scheme\n",
517 		      mem_page_fault ? "memory" : "transport");
518 	return err;
519 }
520 
521 static int max_uc_list_get_devlink_param(struct mlx5_core_dev *dev)
522 {
523 	struct devlink *devlink = priv_to_devlink(dev);
524 	union devlink_param_value val;
525 	int err;
526 
527 	err = devl_param_driverinit_value_get(devlink,
528 					      DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
529 					      &val);
530 	if (!err)
531 		return val.vu32;
532 	mlx5_core_dbg(dev, "Failed to get param. err = %d\n", err);
533 	return err;
534 }
535 
536 bool mlx5_is_roce_on(struct mlx5_core_dev *dev)
537 {
538 	struct devlink *devlink = priv_to_devlink(dev);
539 	union devlink_param_value val;
540 	int err;
541 
542 	err = devl_param_driverinit_value_get(devlink,
543 					      DEVLINK_PARAM_GENERIC_ID_ENABLE_ROCE,
544 					      &val);
545 
546 	if (!err)
547 		return val.vbool;
548 
549 	mlx5_core_dbg(dev, "Failed to get param. err = %d\n", err);
550 	return MLX5_CAP_GEN(dev, roce);
551 }
552 EXPORT_SYMBOL(mlx5_is_roce_on);
553 
554 static int handle_hca_cap_2(struct mlx5_core_dev *dev, void *set_ctx)
555 {
556 	void *set_hca_cap;
557 	int err;
558 
559 	if (!MLX5_CAP_GEN_MAX(dev, hca_cap_2))
560 		return 0;
561 
562 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL_2);
563 	if (err)
564 		return err;
565 
566 	if (!MLX5_CAP_GEN_2_MAX(dev, sw_vhca_id_valid) ||
567 	    !(dev->priv.sw_vhca_id > 0))
568 		return 0;
569 
570 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
571 				   capability);
572 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_GENERAL_2]->cur,
573 	       MLX5_ST_SZ_BYTES(cmd_hca_cap_2));
574 	MLX5_SET(cmd_hca_cap_2, set_hca_cap, sw_vhca_id_valid, 1);
575 
576 	return set_caps(dev, set_ctx, MLX5_CAP_GENERAL_2);
577 }
578 
579 static int handle_hca_cap(struct mlx5_core_dev *dev, void *set_ctx)
580 {
581 	struct mlx5_profile *prof = &dev->profile;
582 	void *set_hca_cap;
583 	int max_uc_list;
584 	int err;
585 
586 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
587 	if (err)
588 		return err;
589 
590 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
591 				   capability);
592 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_GENERAL]->cur,
593 	       MLX5_ST_SZ_BYTES(cmd_hca_cap));
594 
595 	mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
596 		      mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
597 		      128);
598 	/* we limit the size of the pkey table to 128 entries for now */
599 	MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
600 		 to_fw_pkey_sz(dev, 128));
601 
602 	/* Check log_max_qp from HCA caps to set in current profile */
603 	if (prof->log_max_qp == LOG_MAX_SUPPORTED_QPS) {
604 		prof->log_max_qp = min_t(u8, 18, MLX5_CAP_GEN_MAX(dev, log_max_qp));
605 	} else if (MLX5_CAP_GEN_MAX(dev, log_max_qp) < prof->log_max_qp) {
606 		mlx5_core_warn(dev, "log_max_qp value in current profile is %d, changing it to HCA capability limit (%d)\n",
607 			       prof->log_max_qp,
608 			       MLX5_CAP_GEN_MAX(dev, log_max_qp));
609 		prof->log_max_qp = MLX5_CAP_GEN_MAX(dev, log_max_qp);
610 	}
611 	if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
612 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
613 			 prof->log_max_qp);
614 
615 	/* disable cmdif checksum */
616 	MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
617 
618 	/* Enable 4K UAR only when HCA supports it and page size is bigger
619 	 * than 4K.
620 	 */
621 	if (MLX5_CAP_GEN_MAX(dev, uar_4k) && PAGE_SIZE > 4096)
622 		MLX5_SET(cmd_hca_cap, set_hca_cap, uar_4k, 1);
623 
624 	MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
625 
626 	if (MLX5_CAP_GEN_MAX(dev, cache_line_128byte))
627 		MLX5_SET(cmd_hca_cap,
628 			 set_hca_cap,
629 			 cache_line_128byte,
630 			 cache_line_size() >= 128 ? 1 : 0);
631 
632 	if (MLX5_CAP_GEN_MAX(dev, dct))
633 		MLX5_SET(cmd_hca_cap, set_hca_cap, dct, 1);
634 
635 	if (MLX5_CAP_GEN_MAX(dev, pci_sync_for_fw_update_event))
636 		MLX5_SET(cmd_hca_cap, set_hca_cap, pci_sync_for_fw_update_event, 1);
637 	if (MLX5_CAP_GEN_MAX(dev, pci_sync_for_fw_update_with_driver_unload))
638 		MLX5_SET(cmd_hca_cap, set_hca_cap,
639 			 pci_sync_for_fw_update_with_driver_unload, 1);
640 	if (MLX5_CAP_GEN_MAX(dev, pcie_reset_using_hotreset_method))
641 		MLX5_SET(cmd_hca_cap, set_hca_cap,
642 			 pcie_reset_using_hotreset_method, 1);
643 
644 	if (MLX5_CAP_GEN_MAX(dev, num_vhca_ports))
645 		MLX5_SET(cmd_hca_cap,
646 			 set_hca_cap,
647 			 num_vhca_ports,
648 			 MLX5_CAP_GEN_MAX(dev, num_vhca_ports));
649 
650 	if (MLX5_CAP_GEN_MAX(dev, release_all_pages))
651 		MLX5_SET(cmd_hca_cap, set_hca_cap, release_all_pages, 1);
652 
653 	if (MLX5_CAP_GEN_MAX(dev, mkey_by_name))
654 		MLX5_SET(cmd_hca_cap, set_hca_cap, mkey_by_name, 1);
655 
656 	mlx5_vhca_state_cap_handle(dev, set_hca_cap);
657 
658 	if (MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix))
659 		MLX5_SET(cmd_hca_cap, set_hca_cap, num_total_dynamic_vf_msix,
660 			 MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix));
661 
662 	if (MLX5_CAP_GEN(dev, roce_rw_supported) && MLX5_CAP_GEN_MAX(dev, roce))
663 		MLX5_SET(cmd_hca_cap, set_hca_cap, roce,
664 			 mlx5_is_roce_on(dev));
665 
666 	max_uc_list = max_uc_list_get_devlink_param(dev);
667 	if (max_uc_list > 0)
668 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_current_uc_list,
669 			 ilog2(max_uc_list));
670 
671 	/* enable absolute native port num */
672 	if (MLX5_CAP_GEN_MAX(dev, abs_native_port_num))
673 		MLX5_SET(cmd_hca_cap, set_hca_cap, abs_native_port_num, 1);
674 
675 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
676 }
677 
678 /* Cached MLX5_CAP_GEN(dev, roce) can be out of sync this early in the
679  * boot process.
680  * In case RoCE cap is writable in FW and user/devlink requested to change the
681  * cap, we are yet to query the final state of the above cap.
682  * Hence, the need for this function.
683  *
684  * Returns
685  * True:
686  * 1) RoCE cap is read only in FW and already disabled
687  * OR:
688  * 2) RoCE cap is writable in FW and user/devlink requested it off.
689  *
690  * In any other case, return False.
691  */
692 static bool is_roce_fw_disabled(struct mlx5_core_dev *dev)
693 {
694 	return (MLX5_CAP_GEN(dev, roce_rw_supported) && !mlx5_is_roce_on(dev)) ||
695 		(!MLX5_CAP_GEN(dev, roce_rw_supported) && !MLX5_CAP_GEN(dev, roce));
696 }
697 
698 static int handle_hca_cap_roce(struct mlx5_core_dev *dev, void *set_ctx)
699 {
700 	void *set_hca_cap;
701 	int err;
702 
703 	if (is_roce_fw_disabled(dev))
704 		return 0;
705 
706 	err = mlx5_core_get_caps(dev, MLX5_CAP_ROCE);
707 	if (err)
708 		return err;
709 
710 	if (MLX5_CAP_ROCE(dev, sw_r_roce_src_udp_port) ||
711 	    !MLX5_CAP_ROCE_MAX(dev, sw_r_roce_src_udp_port))
712 		return 0;
713 
714 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
715 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_ROCE]->cur,
716 	       MLX5_ST_SZ_BYTES(roce_cap));
717 	MLX5_SET(roce_cap, set_hca_cap, sw_r_roce_src_udp_port, 1);
718 
719 	if (MLX5_CAP_ROCE_MAX(dev, qp_ooo_transmit_default))
720 		MLX5_SET(roce_cap, set_hca_cap, qp_ooo_transmit_default, 1);
721 
722 	err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ROCE);
723 	return err;
724 }
725 
726 static int handle_hca_cap_port_selection(struct mlx5_core_dev *dev,
727 					 void *set_ctx)
728 {
729 	void *set_hca_cap;
730 	int err;
731 
732 	if (!MLX5_CAP_GEN(dev, port_selection_cap))
733 		return 0;
734 
735 	err = mlx5_core_get_caps(dev, MLX5_CAP_PORT_SELECTION);
736 	if (err)
737 		return err;
738 
739 	if (MLX5_CAP_PORT_SELECTION(dev, port_select_flow_table_bypass) ||
740 	    !MLX5_CAP_PORT_SELECTION_MAX(dev, port_select_flow_table_bypass))
741 		return 0;
742 
743 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
744 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_PORT_SELECTION]->cur,
745 	       MLX5_ST_SZ_BYTES(port_selection_cap));
746 	MLX5_SET(port_selection_cap, set_hca_cap, port_select_flow_table_bypass, 1);
747 
748 	err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_PORT_SELECTION);
749 
750 	return err;
751 }
752 
753 static int set_hca_cap(struct mlx5_core_dev *dev)
754 {
755 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
756 	void *set_ctx;
757 	int err;
758 
759 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
760 	if (!set_ctx)
761 		return -ENOMEM;
762 
763 	err = handle_hca_cap(dev, set_ctx);
764 	if (err) {
765 		mlx5_core_err(dev, "handle_hca_cap failed\n");
766 		goto out;
767 	}
768 
769 	memset(set_ctx, 0, set_sz);
770 	err = handle_hca_cap_atomic(dev, set_ctx);
771 	if (err) {
772 		mlx5_core_err(dev, "handle_hca_cap_atomic failed\n");
773 		goto out;
774 	}
775 
776 	memset(set_ctx, 0, set_sz);
777 	err = handle_hca_cap_odp(dev, set_ctx);
778 	if (err) {
779 		mlx5_core_err(dev, "handle_hca_cap_odp failed\n");
780 		goto out;
781 	}
782 
783 	memset(set_ctx, 0, set_sz);
784 	err = handle_hca_cap_roce(dev, set_ctx);
785 	if (err) {
786 		mlx5_core_err(dev, "handle_hca_cap_roce failed\n");
787 		goto out;
788 	}
789 
790 	memset(set_ctx, 0, set_sz);
791 	err = handle_hca_cap_2(dev, set_ctx);
792 	if (err) {
793 		mlx5_core_err(dev, "handle_hca_cap_2 failed\n");
794 		goto out;
795 	}
796 
797 	memset(set_ctx, 0, set_sz);
798 	err = handle_hca_cap_port_selection(dev, set_ctx);
799 	if (err) {
800 		mlx5_core_err(dev, "handle_hca_cap_port_selection failed\n");
801 		goto out;
802 	}
803 
804 out:
805 	kfree(set_ctx);
806 	return err;
807 }
808 
809 static int set_hca_ctrl(struct mlx5_core_dev *dev)
810 {
811 	struct mlx5_reg_host_endianness he_in;
812 	struct mlx5_reg_host_endianness he_out;
813 	int err;
814 
815 	if (!mlx5_core_is_pf(dev))
816 		return 0;
817 
818 	memset(&he_in, 0, sizeof(he_in));
819 	he_in.he = MLX5_SET_HOST_ENDIANNESS;
820 	err = mlx5_core_access_reg(dev, &he_in,  sizeof(he_in),
821 					&he_out, sizeof(he_out),
822 					MLX5_REG_HOST_ENDIANNESS, 0, 1);
823 	return err;
824 }
825 
826 static int mlx5_core_set_hca_defaults(struct mlx5_core_dev *dev)
827 {
828 	int ret = 0;
829 
830 	/* Disable local_lb by default */
831 	if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH)
832 		ret = mlx5_nic_vport_update_local_lb(dev, false);
833 
834 	return ret;
835 }
836 
837 int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
838 {
839 	u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {};
840 
841 	MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
842 	MLX5_SET(enable_hca_in, in, function_id, func_id);
843 	MLX5_SET(enable_hca_in, in, embedded_cpu_function,
844 		 dev->caps.embedded_cpu);
845 	return mlx5_cmd_exec_in(dev, enable_hca, in);
846 }
847 
848 int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
849 {
850 	u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {};
851 
852 	MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
853 	MLX5_SET(disable_hca_in, in, function_id, func_id);
854 	MLX5_SET(enable_hca_in, in, embedded_cpu_function,
855 		 dev->caps.embedded_cpu);
856 	return mlx5_cmd_exec_in(dev, disable_hca, in);
857 }
858 
859 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
860 {
861 	u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {};
862 	u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {};
863 	u32 sup_issi;
864 	int err;
865 
866 	MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
867 	err = mlx5_cmd_exec_inout(dev, query_issi, query_in, query_out);
868 	if (err) {
869 		u32 syndrome = MLX5_GET(query_issi_out, query_out, syndrome);
870 		u8 status = MLX5_GET(query_issi_out, query_out, status);
871 
872 		if (!status || syndrome == MLX5_DRIVER_SYND) {
873 			mlx5_core_err(dev, "Failed to query ISSI err(%d) status(%d) synd(%d)\n",
874 				      err, status, syndrome);
875 			return err;
876 		}
877 
878 		mlx5_core_warn(dev, "Query ISSI is not supported by FW, ISSI is 0\n");
879 		dev->issi = 0;
880 		return 0;
881 	}
882 
883 	sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
884 
885 	if (sup_issi & (1 << 1)) {
886 		u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {};
887 
888 		MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
889 		MLX5_SET(set_issi_in, set_in, current_issi, 1);
890 		err = mlx5_cmd_exec_in(dev, set_issi, set_in);
891 		if (err) {
892 			mlx5_core_err(dev, "Failed to set ISSI to 1 err(%d)\n",
893 				      err);
894 			return err;
895 		}
896 
897 		dev->issi = 1;
898 
899 		return 0;
900 	} else if (sup_issi & (1 << 0) || !sup_issi) {
901 		return 0;
902 	}
903 
904 	return -EOPNOTSUPP;
905 }
906 
907 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct pci_dev *pdev,
908 			 const struct pci_device_id *id)
909 {
910 	int err = 0;
911 
912 	mutex_init(&dev->pci_status_mutex);
913 	pci_set_drvdata(dev->pdev, dev);
914 
915 	dev->bar_addr = pci_resource_start(pdev, 0);
916 
917 	err = mlx5_pci_enable_device(dev);
918 	if (err) {
919 		mlx5_core_err(dev, "Cannot enable PCI device, aborting\n");
920 		return err;
921 	}
922 
923 	err = request_bar(pdev);
924 	if (err) {
925 		mlx5_core_err(dev, "error requesting BARs, aborting\n");
926 		goto err_disable;
927 	}
928 
929 	pci_set_master(pdev);
930 
931 	err = set_dma_caps(pdev);
932 	if (err) {
933 		mlx5_core_err(dev, "Failed setting DMA capabilities mask, aborting\n");
934 		goto err_clr_master;
935 	}
936 
937 	if (pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP32) &&
938 	    pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP64) &&
939 	    pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP128))
940 		mlx5_core_dbg(dev, "Enabling pci atomics failed\n");
941 
942 	dev->iseg_base = dev->bar_addr;
943 	dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
944 	if (!dev->iseg) {
945 		err = -ENOMEM;
946 		mlx5_core_err(dev, "Failed mapping initialization segment, aborting\n");
947 		goto err_clr_master;
948 	}
949 
950 	mlx5_pci_vsc_init(dev);
951 
952 	pci_enable_ptm(pdev, NULL);
953 
954 	return 0;
955 
956 err_clr_master:
957 	release_bar(dev->pdev);
958 err_disable:
959 	mlx5_pci_disable_device(dev);
960 	return err;
961 }
962 
963 static void mlx5_pci_close(struct mlx5_core_dev *dev)
964 {
965 	/* health work might still be active, and it needs pci bar in
966 	 * order to know the NIC state. Therefore, drain the health WQ
967 	 * before removing the pci bars
968 	 */
969 	mlx5_drain_health_wq(dev);
970 	pci_disable_ptm(dev->pdev);
971 	iounmap(dev->iseg);
972 	release_bar(dev->pdev);
973 	mlx5_pci_disable_device(dev);
974 }
975 
976 static void mlx5_register_hca_devcom_comp(struct mlx5_core_dev *dev)
977 {
978 	/* This component is use to sync adding core_dev to lag_dev and to sync
979 	 * changes of mlx5_adev_devices between LAG layer and other layers.
980 	 */
981 	if (!mlx5_lag_is_supported(dev))
982 		return;
983 
984 	dev->priv.hca_devcom_comp =
985 		mlx5_devcom_register_component(dev->priv.devc, MLX5_DEVCOM_HCA_PORTS,
986 					       mlx5_query_nic_system_image_guid(dev),
987 					       NULL, dev);
988 	if (IS_ERR(dev->priv.hca_devcom_comp))
989 		mlx5_core_err(dev, "Failed to register devcom HCA component\n");
990 }
991 
992 static void mlx5_unregister_hca_devcom_comp(struct mlx5_core_dev *dev)
993 {
994 	mlx5_devcom_unregister_component(dev->priv.hca_devcom_comp);
995 }
996 
997 static int mlx5_init_once(struct mlx5_core_dev *dev)
998 {
999 	int err;
1000 
1001 	dev->priv.devc = mlx5_devcom_register_device(dev);
1002 	if (IS_ERR(dev->priv.devc))
1003 		mlx5_core_warn(dev, "failed to register devcom device %ld\n",
1004 			       PTR_ERR(dev->priv.devc));
1005 	mlx5_register_hca_devcom_comp(dev);
1006 
1007 	err = mlx5_query_board_id(dev);
1008 	if (err) {
1009 		mlx5_core_err(dev, "query board id failed\n");
1010 		goto err_devcom;
1011 	}
1012 
1013 	err = mlx5_irq_table_init(dev);
1014 	if (err) {
1015 		mlx5_core_err(dev, "failed to initialize irq table\n");
1016 		goto err_devcom;
1017 	}
1018 
1019 	err = mlx5_eq_table_init(dev);
1020 	if (err) {
1021 		mlx5_core_err(dev, "failed to initialize eq\n");
1022 		goto err_irq_cleanup;
1023 	}
1024 
1025 	err = mlx5_events_init(dev);
1026 	if (err) {
1027 		mlx5_core_err(dev, "failed to initialize events\n");
1028 		goto err_eq_cleanup;
1029 	}
1030 
1031 	err = mlx5_fw_reset_init(dev);
1032 	if (err) {
1033 		mlx5_core_err(dev, "failed to initialize fw reset events\n");
1034 		goto err_events_cleanup;
1035 	}
1036 
1037 	mlx5_cq_debugfs_init(dev);
1038 
1039 	mlx5_init_reserved_gids(dev);
1040 
1041 	mlx5_init_clock(dev);
1042 
1043 	dev->vxlan = mlx5_vxlan_create(dev);
1044 	dev->geneve = mlx5_geneve_create(dev);
1045 
1046 	err = mlx5_init_rl_table(dev);
1047 	if (err) {
1048 		mlx5_core_err(dev, "Failed to init rate limiting\n");
1049 		goto err_tables_cleanup;
1050 	}
1051 
1052 	err = mlx5_mpfs_init(dev);
1053 	if (err) {
1054 		mlx5_core_err(dev, "Failed to init l2 table %d\n", err);
1055 		goto err_rl_cleanup;
1056 	}
1057 
1058 	err = mlx5_sriov_init(dev);
1059 	if (err) {
1060 		mlx5_core_err(dev, "Failed to init sriov %d\n", err);
1061 		goto err_mpfs_cleanup;
1062 	}
1063 
1064 	err = mlx5_eswitch_init(dev);
1065 	if (err) {
1066 		mlx5_core_err(dev, "Failed to init eswitch %d\n", err);
1067 		goto err_sriov_cleanup;
1068 	}
1069 
1070 	err = mlx5_fpga_init(dev);
1071 	if (err) {
1072 		mlx5_core_err(dev, "Failed to init fpga device %d\n", err);
1073 		goto err_eswitch_cleanup;
1074 	}
1075 
1076 	err = mlx5_vhca_event_init(dev);
1077 	if (err) {
1078 		mlx5_core_err(dev, "Failed to init vhca event notifier %d\n", err);
1079 		goto err_fpga_cleanup;
1080 	}
1081 
1082 	err = mlx5_sf_hw_table_init(dev);
1083 	if (err) {
1084 		mlx5_core_err(dev, "Failed to init SF HW table %d\n", err);
1085 		goto err_sf_hw_table_cleanup;
1086 	}
1087 
1088 	err = mlx5_sf_table_init(dev);
1089 	if (err) {
1090 		mlx5_core_err(dev, "Failed to init SF table %d\n", err);
1091 		goto err_sf_table_cleanup;
1092 	}
1093 
1094 	err = mlx5_fs_core_alloc(dev);
1095 	if (err) {
1096 		mlx5_core_err(dev, "Failed to alloc flow steering\n");
1097 		goto err_fs;
1098 	}
1099 
1100 	dev->dm = mlx5_dm_create(dev);
1101 	if (IS_ERR(dev->dm))
1102 		mlx5_core_warn(dev, "Failed to init device memory %ld\n", PTR_ERR(dev->dm));
1103 
1104 	dev->tracer = mlx5_fw_tracer_create(dev);
1105 	dev->hv_vhca = mlx5_hv_vhca_create(dev);
1106 	dev->rsc_dump = mlx5_rsc_dump_create(dev);
1107 
1108 	return 0;
1109 
1110 err_fs:
1111 	mlx5_sf_table_cleanup(dev);
1112 err_sf_table_cleanup:
1113 	mlx5_sf_hw_table_cleanup(dev);
1114 err_sf_hw_table_cleanup:
1115 	mlx5_vhca_event_cleanup(dev);
1116 err_fpga_cleanup:
1117 	mlx5_fpga_cleanup(dev);
1118 err_eswitch_cleanup:
1119 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1120 err_sriov_cleanup:
1121 	mlx5_sriov_cleanup(dev);
1122 err_mpfs_cleanup:
1123 	mlx5_mpfs_cleanup(dev);
1124 err_rl_cleanup:
1125 	mlx5_cleanup_rl_table(dev);
1126 err_tables_cleanup:
1127 	mlx5_geneve_destroy(dev->geneve);
1128 	mlx5_vxlan_destroy(dev->vxlan);
1129 	mlx5_cleanup_clock(dev);
1130 	mlx5_cleanup_reserved_gids(dev);
1131 	mlx5_cq_debugfs_cleanup(dev);
1132 	mlx5_fw_reset_cleanup(dev);
1133 err_events_cleanup:
1134 	mlx5_events_cleanup(dev);
1135 err_eq_cleanup:
1136 	mlx5_eq_table_cleanup(dev);
1137 err_irq_cleanup:
1138 	mlx5_irq_table_cleanup(dev);
1139 err_devcom:
1140 	mlx5_unregister_hca_devcom_comp(dev);
1141 	mlx5_devcom_unregister_device(dev->priv.devc);
1142 
1143 	return err;
1144 }
1145 
1146 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
1147 {
1148 	mlx5_rsc_dump_destroy(dev);
1149 	mlx5_hv_vhca_destroy(dev->hv_vhca);
1150 	mlx5_fw_tracer_destroy(dev->tracer);
1151 	mlx5_dm_cleanup(dev);
1152 	mlx5_fs_core_free(dev);
1153 	mlx5_sf_table_cleanup(dev);
1154 	mlx5_sf_hw_table_cleanup(dev);
1155 	mlx5_vhca_event_cleanup(dev);
1156 	mlx5_fpga_cleanup(dev);
1157 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1158 	mlx5_sriov_cleanup(dev);
1159 	mlx5_mpfs_cleanup(dev);
1160 	mlx5_cleanup_rl_table(dev);
1161 	mlx5_geneve_destroy(dev->geneve);
1162 	mlx5_vxlan_destroy(dev->vxlan);
1163 	mlx5_cleanup_clock(dev);
1164 	mlx5_cleanup_reserved_gids(dev);
1165 	mlx5_cq_debugfs_cleanup(dev);
1166 	mlx5_fw_reset_cleanup(dev);
1167 	mlx5_events_cleanup(dev);
1168 	mlx5_eq_table_cleanup(dev);
1169 	mlx5_irq_table_cleanup(dev);
1170 	mlx5_unregister_hca_devcom_comp(dev);
1171 	mlx5_devcom_unregister_device(dev->priv.devc);
1172 }
1173 
1174 static int mlx5_function_enable(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1175 {
1176 	int err;
1177 
1178 	mlx5_core_info(dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1179 		       fw_rev_min(dev), fw_rev_sub(dev));
1180 
1181 	/* Only PFs hold the relevant PCIe information for this query */
1182 	if (mlx5_core_is_pf(dev))
1183 		pcie_print_link_status(dev->pdev);
1184 
1185 	/* wait for firmware to accept initialization segments configurations
1186 	 */
1187 	err = wait_fw_init(dev, timeout,
1188 			   mlx5_tout_ms(dev, FW_PRE_INIT_WARN_MESSAGE_INTERVAL),
1189 			   "pre-initializing");
1190 	if (err)
1191 		return err;
1192 
1193 	err = mlx5_cmd_enable(dev);
1194 	if (err) {
1195 		mlx5_core_err(dev, "Failed initializing command interface, aborting\n");
1196 		return err;
1197 	}
1198 
1199 	mlx5_tout_query_iseg(dev);
1200 
1201 	err = wait_fw_init(dev, mlx5_tout_ms(dev, FW_INIT), 0, "initializing");
1202 	if (err)
1203 		goto err_cmd_cleanup;
1204 
1205 	dev->caps.embedded_cpu = mlx5_read_embedded_cpu(dev);
1206 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_UP);
1207 
1208 	mlx5_start_health_poll(dev);
1209 
1210 	err = mlx5_core_enable_hca(dev, 0);
1211 	if (err) {
1212 		mlx5_core_err(dev, "enable hca failed\n");
1213 		goto stop_health_poll;
1214 	}
1215 
1216 	err = mlx5_core_set_issi(dev);
1217 	if (err) {
1218 		mlx5_core_err(dev, "failed to set issi\n");
1219 		goto err_disable_hca;
1220 	}
1221 
1222 	err = mlx5_satisfy_startup_pages(dev, 1);
1223 	if (err) {
1224 		mlx5_core_err(dev, "failed to allocate boot pages\n");
1225 		goto err_disable_hca;
1226 	}
1227 
1228 	err = mlx5_tout_query_dtor(dev);
1229 	if (err) {
1230 		mlx5_core_err(dev, "failed to read dtor\n");
1231 		goto reclaim_boot_pages;
1232 	}
1233 
1234 	return 0;
1235 
1236 reclaim_boot_pages:
1237 	mlx5_reclaim_startup_pages(dev);
1238 err_disable_hca:
1239 	mlx5_core_disable_hca(dev, 0);
1240 stop_health_poll:
1241 	mlx5_stop_health_poll(dev, boot);
1242 err_cmd_cleanup:
1243 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1244 	mlx5_cmd_disable(dev);
1245 
1246 	return err;
1247 }
1248 
1249 static void mlx5_function_disable(struct mlx5_core_dev *dev, bool boot)
1250 {
1251 	mlx5_reclaim_startup_pages(dev);
1252 	mlx5_core_disable_hca(dev, 0);
1253 	mlx5_stop_health_poll(dev, boot);
1254 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1255 	mlx5_cmd_disable(dev);
1256 }
1257 
1258 static int mlx5_function_open(struct mlx5_core_dev *dev)
1259 {
1260 	int err;
1261 
1262 	err = set_hca_ctrl(dev);
1263 	if (err) {
1264 		mlx5_core_err(dev, "set_hca_ctrl failed\n");
1265 		return err;
1266 	}
1267 
1268 	err = set_hca_cap(dev);
1269 	if (err) {
1270 		mlx5_core_err(dev, "set_hca_cap failed\n");
1271 		return err;
1272 	}
1273 
1274 	err = mlx5_satisfy_startup_pages(dev, 0);
1275 	if (err) {
1276 		mlx5_core_err(dev, "failed to allocate init pages\n");
1277 		return err;
1278 	}
1279 
1280 	err = mlx5_cmd_init_hca(dev, sw_owner_id);
1281 	if (err) {
1282 		mlx5_core_err(dev, "init hca failed\n");
1283 		return err;
1284 	}
1285 
1286 	mlx5_set_driver_version(dev);
1287 
1288 	err = mlx5_query_hca_caps(dev);
1289 	if (err) {
1290 		mlx5_core_err(dev, "query hca failed\n");
1291 		return err;
1292 	}
1293 	mlx5_start_health_fw_log_up(dev);
1294 	return 0;
1295 }
1296 
1297 static int mlx5_function_close(struct mlx5_core_dev *dev)
1298 {
1299 	int err;
1300 
1301 	err = mlx5_cmd_teardown_hca(dev);
1302 	if (err) {
1303 		mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n");
1304 		return err;
1305 	}
1306 
1307 	return 0;
1308 }
1309 
1310 static int mlx5_function_setup(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1311 {
1312 	int err;
1313 
1314 	err = mlx5_function_enable(dev, boot, timeout);
1315 	if (err)
1316 		return err;
1317 
1318 	err = mlx5_function_open(dev);
1319 	if (err)
1320 		mlx5_function_disable(dev, boot);
1321 	return err;
1322 }
1323 
1324 static int mlx5_function_teardown(struct mlx5_core_dev *dev, bool boot)
1325 {
1326 	int err = mlx5_function_close(dev);
1327 
1328 	if (!err)
1329 		mlx5_function_disable(dev, boot);
1330 	else
1331 		mlx5_stop_health_poll(dev, boot);
1332 
1333 	return err;
1334 }
1335 
1336 static int mlx5_load(struct mlx5_core_dev *dev)
1337 {
1338 	int err;
1339 
1340 	dev->priv.uar = mlx5_get_uars_page(dev);
1341 	if (IS_ERR(dev->priv.uar)) {
1342 		mlx5_core_err(dev, "Failed allocating uar, aborting\n");
1343 		err = PTR_ERR(dev->priv.uar);
1344 		return err;
1345 	}
1346 
1347 	mlx5_events_start(dev);
1348 	mlx5_pagealloc_start(dev);
1349 
1350 	err = mlx5_irq_table_create(dev);
1351 	if (err) {
1352 		mlx5_core_err(dev, "Failed to alloc IRQs\n");
1353 		goto err_irq_table;
1354 	}
1355 
1356 	err = mlx5_eq_table_create(dev);
1357 	if (err) {
1358 		mlx5_core_err(dev, "Failed to create EQs\n");
1359 		goto err_eq_table;
1360 	}
1361 
1362 	err = mlx5_fw_tracer_init(dev->tracer);
1363 	if (err) {
1364 		mlx5_core_err(dev, "Failed to init FW tracer %d\n", err);
1365 		mlx5_fw_tracer_destroy(dev->tracer);
1366 		dev->tracer = NULL;
1367 	}
1368 
1369 	mlx5_fw_reset_events_start(dev);
1370 	mlx5_hv_vhca_init(dev->hv_vhca);
1371 
1372 	err = mlx5_rsc_dump_init(dev);
1373 	if (err) {
1374 		mlx5_core_err(dev, "Failed to init Resource dump %d\n", err);
1375 		mlx5_rsc_dump_destroy(dev);
1376 		dev->rsc_dump = NULL;
1377 	}
1378 
1379 	err = mlx5_fpga_device_start(dev);
1380 	if (err) {
1381 		mlx5_core_err(dev, "fpga device start failed %d\n", err);
1382 		goto err_fpga_start;
1383 	}
1384 
1385 	err = mlx5_fs_core_init(dev);
1386 	if (err) {
1387 		mlx5_core_err(dev, "Failed to init flow steering\n");
1388 		goto err_fs;
1389 	}
1390 
1391 	err = mlx5_core_set_hca_defaults(dev);
1392 	if (err) {
1393 		mlx5_core_err(dev, "Failed to set hca defaults\n");
1394 		goto err_set_hca;
1395 	}
1396 
1397 	mlx5_vhca_event_start(dev);
1398 
1399 	err = mlx5_sf_hw_table_create(dev);
1400 	if (err) {
1401 		mlx5_core_err(dev, "sf table create failed %d\n", err);
1402 		goto err_vhca;
1403 	}
1404 
1405 	err = mlx5_ec_init(dev);
1406 	if (err) {
1407 		mlx5_core_err(dev, "Failed to init embedded CPU\n");
1408 		goto err_ec;
1409 	}
1410 
1411 	mlx5_lag_add_mdev(dev);
1412 	err = mlx5_sriov_attach(dev);
1413 	if (err) {
1414 		mlx5_core_err(dev, "sriov init failed %d\n", err);
1415 		goto err_sriov;
1416 	}
1417 
1418 	mlx5_sf_dev_table_create(dev);
1419 
1420 	err = mlx5_devlink_traps_register(priv_to_devlink(dev));
1421 	if (err)
1422 		goto err_traps_reg;
1423 
1424 	return 0;
1425 
1426 err_traps_reg:
1427 	mlx5_sf_dev_table_destroy(dev);
1428 	mlx5_sriov_detach(dev);
1429 err_sriov:
1430 	mlx5_lag_remove_mdev(dev);
1431 	mlx5_ec_cleanup(dev);
1432 err_ec:
1433 	mlx5_sf_hw_table_destroy(dev);
1434 err_vhca:
1435 	mlx5_vhca_event_stop(dev);
1436 err_set_hca:
1437 	mlx5_fs_core_cleanup(dev);
1438 err_fs:
1439 	mlx5_fpga_device_stop(dev);
1440 err_fpga_start:
1441 	mlx5_rsc_dump_cleanup(dev);
1442 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1443 	mlx5_fw_reset_events_stop(dev);
1444 	mlx5_fw_tracer_cleanup(dev->tracer);
1445 	mlx5_eq_table_destroy(dev);
1446 err_eq_table:
1447 	mlx5_irq_table_destroy(dev);
1448 err_irq_table:
1449 	mlx5_pagealloc_stop(dev);
1450 	mlx5_events_stop(dev);
1451 	mlx5_put_uars_page(dev, dev->priv.uar);
1452 	return err;
1453 }
1454 
1455 static void mlx5_unload(struct mlx5_core_dev *dev)
1456 {
1457 	mlx5_eswitch_disable(dev->priv.eswitch);
1458 	mlx5_devlink_traps_unregister(priv_to_devlink(dev));
1459 	mlx5_sf_dev_table_destroy(dev);
1460 	mlx5_sriov_detach(dev);
1461 	mlx5_lag_remove_mdev(dev);
1462 	mlx5_ec_cleanup(dev);
1463 	mlx5_sf_hw_table_destroy(dev);
1464 	mlx5_vhca_event_stop(dev);
1465 	mlx5_fs_core_cleanup(dev);
1466 	mlx5_fpga_device_stop(dev);
1467 	mlx5_rsc_dump_cleanup(dev);
1468 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1469 	mlx5_fw_reset_events_stop(dev);
1470 	mlx5_fw_tracer_cleanup(dev->tracer);
1471 	mlx5_eq_table_destroy(dev);
1472 	mlx5_irq_table_destroy(dev);
1473 	mlx5_pagealloc_stop(dev);
1474 	mlx5_events_stop(dev);
1475 	mlx5_put_uars_page(dev, dev->priv.uar);
1476 }
1477 
1478 int mlx5_init_one_devl_locked(struct mlx5_core_dev *dev)
1479 {
1480 	bool light_probe = mlx5_dev_is_lightweight(dev);
1481 	int err = 0;
1482 
1483 	mutex_lock(&dev->intf_state_mutex);
1484 	dev->state = MLX5_DEVICE_STATE_UP;
1485 
1486 	err = mlx5_function_setup(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1487 	if (err)
1488 		goto err_function;
1489 
1490 	err = mlx5_init_once(dev);
1491 	if (err) {
1492 		mlx5_core_err(dev, "sw objs init failed\n");
1493 		goto function_teardown;
1494 	}
1495 
1496 	/* In case of light_probe, mlx5_devlink is already registered.
1497 	 * Hence, don't register devlink again.
1498 	 */
1499 	if (!light_probe) {
1500 		err = mlx5_devlink_params_register(priv_to_devlink(dev));
1501 		if (err)
1502 			goto err_devlink_params_reg;
1503 	}
1504 
1505 	err = mlx5_load(dev);
1506 	if (err)
1507 		goto err_load;
1508 
1509 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1510 
1511 	err = mlx5_register_device(dev);
1512 	if (err)
1513 		goto err_register;
1514 
1515 	err = mlx5_crdump_enable(dev);
1516 	if (err)
1517 		mlx5_core_err(dev, "mlx5_crdump_enable failed with error code %d\n", err);
1518 
1519 	err = mlx5_hwmon_dev_register(dev);
1520 	if (err)
1521 		mlx5_core_err(dev, "mlx5_hwmon_dev_register failed with error code %d\n", err);
1522 
1523 	mutex_unlock(&dev->intf_state_mutex);
1524 	return 0;
1525 
1526 err_register:
1527 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1528 	mlx5_unload(dev);
1529 err_load:
1530 	if (!light_probe)
1531 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1532 err_devlink_params_reg:
1533 	mlx5_cleanup_once(dev);
1534 function_teardown:
1535 	mlx5_function_teardown(dev, true);
1536 err_function:
1537 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1538 	mutex_unlock(&dev->intf_state_mutex);
1539 	return err;
1540 }
1541 
1542 int mlx5_init_one(struct mlx5_core_dev *dev)
1543 {
1544 	struct devlink *devlink = priv_to_devlink(dev);
1545 	int err;
1546 
1547 	devl_lock(devlink);
1548 	devl_register(devlink);
1549 	err = mlx5_init_one_devl_locked(dev);
1550 	if (err)
1551 		devl_unregister(devlink);
1552 	devl_unlock(devlink);
1553 	return err;
1554 }
1555 
1556 void mlx5_uninit_one(struct mlx5_core_dev *dev)
1557 {
1558 	struct devlink *devlink = priv_to_devlink(dev);
1559 
1560 	devl_lock(devlink);
1561 	mutex_lock(&dev->intf_state_mutex);
1562 
1563 	mlx5_hwmon_dev_unregister(dev);
1564 	mlx5_crdump_disable(dev);
1565 	mlx5_unregister_device(dev);
1566 
1567 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1568 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1569 			       __func__);
1570 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1571 		mlx5_cleanup_once(dev);
1572 		goto out;
1573 	}
1574 
1575 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1576 	mlx5_unload(dev);
1577 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1578 	mlx5_cleanup_once(dev);
1579 	mlx5_function_teardown(dev, true);
1580 out:
1581 	mutex_unlock(&dev->intf_state_mutex);
1582 	devl_unregister(devlink);
1583 	devl_unlock(devlink);
1584 }
1585 
1586 int mlx5_load_one_devl_locked(struct mlx5_core_dev *dev, bool recovery)
1587 {
1588 	int err = 0;
1589 	u64 timeout;
1590 
1591 	devl_assert_locked(priv_to_devlink(dev));
1592 	mutex_lock(&dev->intf_state_mutex);
1593 	if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1594 		mlx5_core_warn(dev, "interface is up, NOP\n");
1595 		goto out;
1596 	}
1597 	/* remove any previous indication of internal error */
1598 	dev->state = MLX5_DEVICE_STATE_UP;
1599 
1600 	if (recovery)
1601 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_ON_RECOVERY_TIMEOUT);
1602 	else
1603 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT);
1604 	err = mlx5_function_setup(dev, false, timeout);
1605 	if (err)
1606 		goto err_function;
1607 
1608 	err = mlx5_load(dev);
1609 	if (err)
1610 		goto err_load;
1611 
1612 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1613 
1614 	err = mlx5_attach_device(dev);
1615 	if (err)
1616 		goto err_attach;
1617 
1618 	mutex_unlock(&dev->intf_state_mutex);
1619 	return 0;
1620 
1621 err_attach:
1622 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1623 	mlx5_unload(dev);
1624 err_load:
1625 	mlx5_function_teardown(dev, false);
1626 err_function:
1627 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1628 out:
1629 	mutex_unlock(&dev->intf_state_mutex);
1630 	return err;
1631 }
1632 
1633 int mlx5_load_one(struct mlx5_core_dev *dev, bool recovery)
1634 {
1635 	struct devlink *devlink = priv_to_devlink(dev);
1636 	int ret;
1637 
1638 	devl_lock(devlink);
1639 	ret = mlx5_load_one_devl_locked(dev, recovery);
1640 	devl_unlock(devlink);
1641 	return ret;
1642 }
1643 
1644 void mlx5_unload_one_devl_locked(struct mlx5_core_dev *dev, bool suspend)
1645 {
1646 	devl_assert_locked(priv_to_devlink(dev));
1647 	mutex_lock(&dev->intf_state_mutex);
1648 
1649 	mlx5_detach_device(dev, suspend);
1650 
1651 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1652 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1653 			       __func__);
1654 		goto out;
1655 	}
1656 
1657 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1658 	mlx5_unload(dev);
1659 	mlx5_function_teardown(dev, false);
1660 out:
1661 	mutex_unlock(&dev->intf_state_mutex);
1662 }
1663 
1664 void mlx5_unload_one(struct mlx5_core_dev *dev, bool suspend)
1665 {
1666 	struct devlink *devlink = priv_to_devlink(dev);
1667 
1668 	devl_lock(devlink);
1669 	mlx5_unload_one_devl_locked(dev, suspend);
1670 	devl_unlock(devlink);
1671 }
1672 
1673 /* In case of light probe, we don't need a full query of hca_caps, but only the bellow caps.
1674  * A full query of hca_caps will be done when the device will reload.
1675  */
1676 static int mlx5_query_hca_caps_light(struct mlx5_core_dev *dev)
1677 {
1678 	int err;
1679 
1680 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
1681 	if (err)
1682 		return err;
1683 
1684 	if (MLX5_CAP_GEN(dev, eth_net_offloads)) {
1685 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_ETHERNET_OFFLOADS,
1686 					      HCA_CAP_OPMOD_GET_CUR);
1687 		if (err)
1688 			return err;
1689 	}
1690 
1691 	if (MLX5_CAP_GEN(dev, nic_flow_table) ||
1692 	    MLX5_CAP_GEN(dev, ipoib_enhanced_offloads)) {
1693 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_FLOW_TABLE,
1694 					      HCA_CAP_OPMOD_GET_CUR);
1695 		if (err)
1696 			return err;
1697 	}
1698 
1699 	if (MLX5_CAP_GEN_64(dev, general_obj_types) &
1700 		MLX5_GENERAL_OBJ_TYPES_CAP_VIRTIO_NET_Q) {
1701 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_VDPA_EMULATION,
1702 					      HCA_CAP_OPMOD_GET_CUR);
1703 		if (err)
1704 			return err;
1705 	}
1706 
1707 	return 0;
1708 }
1709 
1710 int mlx5_init_one_light(struct mlx5_core_dev *dev)
1711 {
1712 	struct devlink *devlink = priv_to_devlink(dev);
1713 	int err;
1714 
1715 	devl_lock(devlink);
1716 	devl_register(devlink);
1717 	dev->state = MLX5_DEVICE_STATE_UP;
1718 	err = mlx5_function_enable(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1719 	if (err) {
1720 		mlx5_core_warn(dev, "mlx5_function_enable err=%d\n", err);
1721 		goto out;
1722 	}
1723 
1724 	err = mlx5_query_hca_caps_light(dev);
1725 	if (err) {
1726 		mlx5_core_warn(dev, "mlx5_query_hca_caps_light err=%d\n", err);
1727 		goto query_hca_caps_err;
1728 	}
1729 
1730 	err = mlx5_devlink_params_register(priv_to_devlink(dev));
1731 	if (err) {
1732 		mlx5_core_warn(dev, "mlx5_devlink_param_reg err = %d\n", err);
1733 		goto query_hca_caps_err;
1734 	}
1735 
1736 	devl_unlock(devlink);
1737 	return 0;
1738 
1739 query_hca_caps_err:
1740 	mlx5_function_disable(dev, true);
1741 out:
1742 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1743 	devl_unregister(devlink);
1744 	devl_unlock(devlink);
1745 	return err;
1746 }
1747 
1748 void mlx5_uninit_one_light(struct mlx5_core_dev *dev)
1749 {
1750 	struct devlink *devlink = priv_to_devlink(dev);
1751 
1752 	devl_lock(devlink);
1753 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1754 	devl_unregister(devlink);
1755 	devl_unlock(devlink);
1756 	if (dev->state != MLX5_DEVICE_STATE_UP)
1757 		return;
1758 	mlx5_function_disable(dev, true);
1759 }
1760 
1761 /* xxx_light() function are used in order to configure the device without full
1762  * init (light init). e.g.: There isn't a point in reload a device to light state.
1763  * Hence, mlx5_load_one_light() isn't needed.
1764  */
1765 
1766 void mlx5_unload_one_light(struct mlx5_core_dev *dev)
1767 {
1768 	if (dev->state != MLX5_DEVICE_STATE_UP)
1769 		return;
1770 	mlx5_function_disable(dev, false);
1771 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1772 }
1773 
1774 static const int types[] = {
1775 	MLX5_CAP_GENERAL,
1776 	MLX5_CAP_GENERAL_2,
1777 	MLX5_CAP_ETHERNET_OFFLOADS,
1778 	MLX5_CAP_IPOIB_ENHANCED_OFFLOADS,
1779 	MLX5_CAP_ODP,
1780 	MLX5_CAP_ATOMIC,
1781 	MLX5_CAP_ROCE,
1782 	MLX5_CAP_IPOIB_OFFLOADS,
1783 	MLX5_CAP_FLOW_TABLE,
1784 	MLX5_CAP_ESWITCH_FLOW_TABLE,
1785 	MLX5_CAP_ESWITCH,
1786 	MLX5_CAP_QOS,
1787 	MLX5_CAP_DEBUG,
1788 	MLX5_CAP_DEV_MEM,
1789 	MLX5_CAP_DEV_EVENT,
1790 	MLX5_CAP_TLS,
1791 	MLX5_CAP_VDPA_EMULATION,
1792 	MLX5_CAP_IPSEC,
1793 	MLX5_CAP_PORT_SELECTION,
1794 	MLX5_CAP_MACSEC,
1795 	MLX5_CAP_ADV_VIRTUALIZATION,
1796 	MLX5_CAP_CRYPTO,
1797 	MLX5_CAP_SHAMPO,
1798 };
1799 
1800 static void mlx5_hca_caps_free(struct mlx5_core_dev *dev)
1801 {
1802 	int type;
1803 	int i;
1804 
1805 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1806 		type = types[i];
1807 		kfree(dev->caps.hca[type]);
1808 	}
1809 }
1810 
1811 static int mlx5_hca_caps_alloc(struct mlx5_core_dev *dev)
1812 {
1813 	struct mlx5_hca_cap *cap;
1814 	int type;
1815 	int i;
1816 
1817 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1818 		cap = kzalloc(sizeof(*cap), GFP_KERNEL);
1819 		if (!cap)
1820 			goto err;
1821 		type = types[i];
1822 		dev->caps.hca[type] = cap;
1823 	}
1824 
1825 	return 0;
1826 
1827 err:
1828 	mlx5_hca_caps_free(dev);
1829 	return -ENOMEM;
1830 }
1831 
1832 static int vhca_id_show(struct seq_file *file, void *priv)
1833 {
1834 	struct mlx5_core_dev *dev = file->private;
1835 
1836 	seq_printf(file, "0x%x\n", MLX5_CAP_GEN(dev, vhca_id));
1837 	return 0;
1838 }
1839 
1840 DEFINE_SHOW_ATTRIBUTE(vhca_id);
1841 
1842 int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
1843 {
1844 	struct mlx5_priv *priv = &dev->priv;
1845 	int err;
1846 
1847 	memcpy(&dev->profile, &profile[profile_idx], sizeof(dev->profile));
1848 	lockdep_register_key(&dev->lock_key);
1849 	mutex_init(&dev->intf_state_mutex);
1850 	lockdep_set_class(&dev->intf_state_mutex, &dev->lock_key);
1851 	mutex_init(&dev->mlx5e_res.uplink_netdev_lock);
1852 	mutex_init(&dev->wc_state_lock);
1853 
1854 	mutex_init(&priv->bfregs.reg_head.lock);
1855 	mutex_init(&priv->bfregs.wc_head.lock);
1856 	INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1857 	INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1858 
1859 	mutex_init(&priv->alloc_mutex);
1860 	mutex_init(&priv->pgdir_mutex);
1861 	INIT_LIST_HEAD(&priv->pgdir_list);
1862 
1863 	priv->numa_node = dev_to_node(mlx5_core_dma_dev(dev));
1864 	priv->dbg.dbg_root = debugfs_create_dir(dev_name(dev->device),
1865 						mlx5_debugfs_root);
1866 	debugfs_create_file("vhca_id", 0400, priv->dbg.dbg_root, dev, &vhca_id_fops);
1867 	INIT_LIST_HEAD(&priv->traps);
1868 
1869 	err = mlx5_cmd_init(dev);
1870 	if (err) {
1871 		mlx5_core_err(dev, "Failed initializing cmdif SW structs, aborting\n");
1872 		goto err_cmd_init;
1873 	}
1874 
1875 	err = mlx5_tout_init(dev);
1876 	if (err) {
1877 		mlx5_core_err(dev, "Failed initializing timeouts, aborting\n");
1878 		goto err_timeout_init;
1879 	}
1880 
1881 	err = mlx5_health_init(dev);
1882 	if (err)
1883 		goto err_health_init;
1884 
1885 	err = mlx5_pagealloc_init(dev);
1886 	if (err)
1887 		goto err_pagealloc_init;
1888 
1889 	err = mlx5_adev_init(dev);
1890 	if (err)
1891 		goto err_adev_init;
1892 
1893 	err = mlx5_hca_caps_alloc(dev);
1894 	if (err)
1895 		goto err_hca_caps;
1896 
1897 	/* The conjunction of sw_vhca_id with sw_owner_id will be a global
1898 	 * unique id per function which uses mlx5_core.
1899 	 * Those values are supplied to FW as part of the init HCA command to
1900 	 * be used by both driver and FW when it's applicable.
1901 	 */
1902 	dev->priv.sw_vhca_id = ida_alloc_range(&sw_vhca_ida, 1,
1903 					       MAX_SW_VHCA_ID,
1904 					       GFP_KERNEL);
1905 	if (dev->priv.sw_vhca_id < 0)
1906 		mlx5_core_err(dev, "failed to allocate sw_vhca_id, err=%d\n",
1907 			      dev->priv.sw_vhca_id);
1908 
1909 	return 0;
1910 
1911 err_hca_caps:
1912 	mlx5_adev_cleanup(dev);
1913 err_adev_init:
1914 	mlx5_pagealloc_cleanup(dev);
1915 err_pagealloc_init:
1916 	mlx5_health_cleanup(dev);
1917 err_health_init:
1918 	mlx5_tout_cleanup(dev);
1919 err_timeout_init:
1920 	mlx5_cmd_cleanup(dev);
1921 err_cmd_init:
1922 	debugfs_remove(dev->priv.dbg.dbg_root);
1923 	mutex_destroy(&priv->pgdir_mutex);
1924 	mutex_destroy(&priv->alloc_mutex);
1925 	mutex_destroy(&priv->bfregs.wc_head.lock);
1926 	mutex_destroy(&priv->bfregs.reg_head.lock);
1927 	mutex_destroy(&dev->intf_state_mutex);
1928 	lockdep_unregister_key(&dev->lock_key);
1929 	return err;
1930 }
1931 
1932 void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
1933 {
1934 	struct mlx5_priv *priv = &dev->priv;
1935 
1936 	if (priv->sw_vhca_id > 0)
1937 		ida_free(&sw_vhca_ida, dev->priv.sw_vhca_id);
1938 
1939 	mlx5_hca_caps_free(dev);
1940 	mlx5_adev_cleanup(dev);
1941 	mlx5_pagealloc_cleanup(dev);
1942 	mlx5_health_cleanup(dev);
1943 	mlx5_tout_cleanup(dev);
1944 	mlx5_cmd_cleanup(dev);
1945 	debugfs_remove_recursive(dev->priv.dbg.dbg_root);
1946 	mutex_destroy(&priv->pgdir_mutex);
1947 	mutex_destroy(&priv->alloc_mutex);
1948 	mutex_destroy(&priv->bfregs.wc_head.lock);
1949 	mutex_destroy(&priv->bfregs.reg_head.lock);
1950 	mutex_destroy(&dev->wc_state_lock);
1951 	mutex_destroy(&dev->mlx5e_res.uplink_netdev_lock);
1952 	mutex_destroy(&dev->intf_state_mutex);
1953 	lockdep_unregister_key(&dev->lock_key);
1954 }
1955 
1956 static int probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
1957 {
1958 	struct mlx5_core_dev *dev;
1959 	struct devlink *devlink;
1960 	int err;
1961 
1962 	devlink = mlx5_devlink_alloc(&pdev->dev);
1963 	if (!devlink) {
1964 		dev_err(&pdev->dev, "devlink alloc failed\n");
1965 		return -ENOMEM;
1966 	}
1967 
1968 	dev = devlink_priv(devlink);
1969 	dev->device = &pdev->dev;
1970 	dev->pdev = pdev;
1971 
1972 	dev->coredev_type = id->driver_data & MLX5_PCI_DEV_IS_VF ?
1973 			 MLX5_COREDEV_VF : MLX5_COREDEV_PF;
1974 
1975 	dev->priv.adev_idx = mlx5_adev_idx_alloc();
1976 	if (dev->priv.adev_idx < 0) {
1977 		err = dev->priv.adev_idx;
1978 		goto adev_init_err;
1979 	}
1980 
1981 	err = mlx5_mdev_init(dev, prof_sel);
1982 	if (err)
1983 		goto mdev_init_err;
1984 
1985 	err = mlx5_pci_init(dev, pdev, id);
1986 	if (err) {
1987 		mlx5_core_err(dev, "mlx5_pci_init failed with error code %d\n",
1988 			      err);
1989 		goto pci_init_err;
1990 	}
1991 
1992 	err = mlx5_init_one(dev);
1993 	if (err) {
1994 		mlx5_core_err(dev, "mlx5_init_one failed with error code %d\n",
1995 			      err);
1996 		goto err_init_one;
1997 	}
1998 
1999 	pci_save_state(pdev);
2000 	return 0;
2001 
2002 err_init_one:
2003 	mlx5_pci_close(dev);
2004 pci_init_err:
2005 	mlx5_mdev_uninit(dev);
2006 mdev_init_err:
2007 	mlx5_adev_idx_free(dev->priv.adev_idx);
2008 adev_init_err:
2009 	mlx5_devlink_free(devlink);
2010 
2011 	return err;
2012 }
2013 
2014 static void remove_one(struct pci_dev *pdev)
2015 {
2016 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2017 	struct devlink *devlink = priv_to_devlink(dev);
2018 
2019 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2020 	mlx5_drain_fw_reset(dev);
2021 	mlx5_drain_health_wq(dev);
2022 	mlx5_sriov_disable(pdev, false);
2023 	mlx5_uninit_one(dev);
2024 	mlx5_pci_close(dev);
2025 	mlx5_mdev_uninit(dev);
2026 	mlx5_adev_idx_free(dev->priv.adev_idx);
2027 	mlx5_devlink_free(devlink);
2028 }
2029 
2030 #define mlx5_pci_trace(dev, fmt, ...) ({ \
2031 	struct mlx5_core_dev *__dev = (dev); \
2032 	mlx5_core_info(__dev, "%s Device state = %d health sensors: %d pci_status: %d. " fmt, \
2033 		       __func__, __dev->state, mlx5_health_check_fatal_sensors(__dev), \
2034 		       __dev->pci_status, ##__VA_ARGS__); \
2035 })
2036 
2037 static const char *result2str(enum pci_ers_result result)
2038 {
2039 	return  result == PCI_ERS_RESULT_NEED_RESET ? "need reset" :
2040 		result == PCI_ERS_RESULT_DISCONNECT ? "disconnect" :
2041 		result == PCI_ERS_RESULT_RECOVERED  ? "recovered" :
2042 		"unknown";
2043 }
2044 
2045 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
2046 					      pci_channel_state_t state)
2047 {
2048 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2049 	enum pci_ers_result res;
2050 
2051 	mlx5_pci_trace(dev, "Enter, pci channel state = %d\n", state);
2052 
2053 	mlx5_enter_error_state(dev, false);
2054 	mlx5_error_sw_reset(dev);
2055 	mlx5_unload_one(dev, false);
2056 	mlx5_drain_health_wq(dev);
2057 	mlx5_pci_disable_device(dev);
2058 
2059 	res = state == pci_channel_io_perm_failure ?
2060 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2061 
2062 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, result = %d, %s\n",
2063 		       __func__, dev->state, dev->pci_status, res, result2str(res));
2064 	return res;
2065 }
2066 
2067 /* wait for the device to show vital signs by waiting
2068  * for the health counter to start counting.
2069  */
2070 static int wait_vital(struct pci_dev *pdev)
2071 {
2072 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2073 	struct mlx5_core_health *health = &dev->priv.health;
2074 	const int niter = 100;
2075 	u32 last_count = 0;
2076 	u32 count;
2077 	int i;
2078 
2079 	for (i = 0; i < niter; i++) {
2080 		count = ioread32be(health->health_counter);
2081 		if (count && count != 0xffffffff) {
2082 			if (last_count && last_count != count) {
2083 				mlx5_core_info(dev,
2084 					       "wait vital counter value 0x%x after %d iterations\n",
2085 					       count, i);
2086 				return 0;
2087 			}
2088 			last_count = count;
2089 		}
2090 		msleep(50);
2091 	}
2092 
2093 	return -ETIMEDOUT;
2094 }
2095 
2096 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
2097 {
2098 	enum pci_ers_result res = PCI_ERS_RESULT_DISCONNECT;
2099 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2100 	int err;
2101 
2102 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Enter\n",
2103 		       __func__, dev->state, dev->pci_status);
2104 
2105 	err = mlx5_pci_enable_device(dev);
2106 	if (err) {
2107 		mlx5_core_err(dev, "%s: mlx5_pci_enable_device failed with error code: %d\n",
2108 			      __func__, err);
2109 		goto out;
2110 	}
2111 
2112 	pci_set_master(pdev);
2113 	pci_restore_state(pdev);
2114 	pci_save_state(pdev);
2115 
2116 	err = wait_vital(pdev);
2117 	if (err) {
2118 		mlx5_core_err(dev, "%s: wait vital failed with error code: %d\n",
2119 			      __func__, err);
2120 		goto out;
2121 	}
2122 
2123 	res = PCI_ERS_RESULT_RECOVERED;
2124 out:
2125 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, err = %d, result = %d, %s\n",
2126 		       __func__, dev->state, dev->pci_status, err, res, result2str(res));
2127 	return res;
2128 }
2129 
2130 static void mlx5_pci_resume(struct pci_dev *pdev)
2131 {
2132 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2133 	int err;
2134 
2135 	mlx5_pci_trace(dev, "Enter, loading driver..\n");
2136 
2137 	err = mlx5_load_one(dev, false);
2138 
2139 	if (!err)
2140 		devlink_health_reporter_state_update(dev->priv.health.fw_fatal_reporter,
2141 						     DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
2142 
2143 	mlx5_pci_trace(dev, "Done, err = %d, device %s\n", err,
2144 		       !err ? "recovered" : "Failed");
2145 }
2146 
2147 static const struct pci_error_handlers mlx5_err_handler = {
2148 	.error_detected = mlx5_pci_err_detected,
2149 	.slot_reset	= mlx5_pci_slot_reset,
2150 	.resume		= mlx5_pci_resume
2151 };
2152 
2153 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
2154 {
2155 	bool fast_teardown = false, force_teardown = false;
2156 	int ret = 1;
2157 
2158 	fast_teardown = MLX5_CAP_GEN(dev, fast_teardown);
2159 	force_teardown = MLX5_CAP_GEN(dev, force_teardown);
2160 
2161 	mlx5_core_dbg(dev, "force teardown firmware support=%d\n", force_teardown);
2162 	mlx5_core_dbg(dev, "fast teardown firmware support=%d\n", fast_teardown);
2163 
2164 	if (!fast_teardown && !force_teardown)
2165 		return -EOPNOTSUPP;
2166 
2167 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
2168 		mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
2169 		return -EAGAIN;
2170 	}
2171 
2172 	/* Panic tear down fw command will stop the PCI bus communication
2173 	 * with the HCA, so the health poll is no longer needed.
2174 	 */
2175 	mlx5_stop_health_poll(dev, false);
2176 
2177 	ret = mlx5_cmd_fast_teardown_hca(dev);
2178 	if (!ret)
2179 		goto succeed;
2180 
2181 	ret = mlx5_cmd_force_teardown_hca(dev);
2182 	if (!ret)
2183 		goto succeed;
2184 
2185 	mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", ret);
2186 	mlx5_start_health_poll(dev);
2187 	return ret;
2188 
2189 succeed:
2190 	mlx5_enter_error_state(dev, true);
2191 
2192 	/* Some platforms requiring freeing the IRQ's in the shutdown
2193 	 * flow. If they aren't freed they can't be allocated after
2194 	 * kexec. There is no need to cleanup the mlx5_core software
2195 	 * contexts.
2196 	 */
2197 	mlx5_core_eq_free_irqs(dev);
2198 
2199 	return 0;
2200 }
2201 
2202 static void shutdown(struct pci_dev *pdev)
2203 {
2204 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2205 	int err;
2206 
2207 	mlx5_core_info(dev, "Shutdown was called\n");
2208 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2209 	mlx5_drain_health_wq(dev);
2210 	err = mlx5_try_fast_unload(dev);
2211 	if (err)
2212 		mlx5_unload_one(dev, false);
2213 	mlx5_pci_disable_device(dev);
2214 }
2215 
2216 static int mlx5_suspend(struct pci_dev *pdev, pm_message_t state)
2217 {
2218 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2219 
2220 	mlx5_unload_one(dev, true);
2221 
2222 	return 0;
2223 }
2224 
2225 static int mlx5_resume(struct pci_dev *pdev)
2226 {
2227 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2228 
2229 	return mlx5_load_one(dev, false);
2230 }
2231 
2232 static const struct pci_device_id mlx5_core_pci_table[] = {
2233 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTIB) },
2234 	{ PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF},	/* Connect-IB VF */
2235 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4) },
2236 	{ PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4 VF */
2237 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4_LX) },
2238 	{ PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4LX VF */
2239 	{ PCI_VDEVICE(MELLANOX, 0x1017) },			/* ConnectX-5, PCIe 3.0 */
2240 	{ PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 VF */
2241 	{ PCI_VDEVICE(MELLANOX, 0x1019) },			/* ConnectX-5 Ex */
2242 	{ PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 Ex VF */
2243 	{ PCI_VDEVICE(MELLANOX, 0x101b) },			/* ConnectX-6 */
2244 	{ PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF},	/* ConnectX-6 VF */
2245 	{ PCI_VDEVICE(MELLANOX, 0x101d) },			/* ConnectX-6 Dx */
2246 	{ PCI_VDEVICE(MELLANOX, 0x101e), MLX5_PCI_DEV_IS_VF},	/* ConnectX Family mlx5Gen Virtual Function */
2247 	{ PCI_VDEVICE(MELLANOX, 0x101f) },			/* ConnectX-6 LX */
2248 	{ PCI_VDEVICE(MELLANOX, 0x1021) },			/* ConnectX-7 */
2249 	{ PCI_VDEVICE(MELLANOX, 0x1023) },			/* ConnectX-8 */
2250 	{ PCI_VDEVICE(MELLANOX, 0x1025) },			/* ConnectX-9 */
2251 	{ PCI_VDEVICE(MELLANOX, 0xa2d2) },			/* BlueField integrated ConnectX-5 network controller */
2252 	{ PCI_VDEVICE(MELLANOX, 0xa2d3), MLX5_PCI_DEV_IS_VF},	/* BlueField integrated ConnectX-5 network controller VF */
2253 	{ PCI_VDEVICE(MELLANOX, 0xa2d6) },			/* BlueField-2 integrated ConnectX-6 Dx network controller */
2254 	{ PCI_VDEVICE(MELLANOX, 0xa2dc) },			/* BlueField-3 integrated ConnectX-7 network controller */
2255 	{ PCI_VDEVICE(MELLANOX, 0xa2df) },			/* BlueField-4 integrated ConnectX-8 network controller */
2256 	{ 0, }
2257 };
2258 
2259 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
2260 
2261 void mlx5_disable_device(struct mlx5_core_dev *dev)
2262 {
2263 	mlx5_error_sw_reset(dev);
2264 	mlx5_unload_one_devl_locked(dev, false);
2265 }
2266 
2267 int mlx5_recover_device(struct mlx5_core_dev *dev)
2268 {
2269 	if (!mlx5_core_is_sf(dev)) {
2270 		mlx5_pci_disable_device(dev);
2271 		if (mlx5_pci_slot_reset(dev->pdev) != PCI_ERS_RESULT_RECOVERED)
2272 			return -EIO;
2273 	}
2274 
2275 	return mlx5_load_one_devl_locked(dev, true);
2276 }
2277 
2278 static struct pci_driver mlx5_core_driver = {
2279 	.name           = KBUILD_MODNAME,
2280 	.id_table       = mlx5_core_pci_table,
2281 	.probe          = probe_one,
2282 	.remove         = remove_one,
2283 	.suspend        = mlx5_suspend,
2284 	.resume         = mlx5_resume,
2285 	.shutdown	= shutdown,
2286 	.err_handler	= &mlx5_err_handler,
2287 	.sriov_configure   = mlx5_core_sriov_configure,
2288 	.sriov_get_vf_total_msix = mlx5_sriov_get_vf_total_msix,
2289 	.sriov_set_msix_vec_count = mlx5_core_sriov_set_msix_vec_count,
2290 };
2291 
2292 /**
2293  * mlx5_vf_get_core_dev - Get the mlx5 core device from a given VF PCI device if
2294  *                     mlx5_core is its driver.
2295  * @pdev: The associated PCI device.
2296  *
2297  * Upon return the interface state lock stay held to let caller uses it safely.
2298  * Caller must ensure to use the returned mlx5 device for a narrow window
2299  * and put it back with mlx5_vf_put_core_dev() immediately once usage was over.
2300  *
2301  * Return: Pointer to the associated mlx5_core_dev or NULL.
2302  */
2303 struct mlx5_core_dev *mlx5_vf_get_core_dev(struct pci_dev *pdev)
2304 {
2305 	struct mlx5_core_dev *mdev;
2306 
2307 	mdev = pci_iov_get_pf_drvdata(pdev, &mlx5_core_driver);
2308 	if (IS_ERR(mdev))
2309 		return NULL;
2310 
2311 	mutex_lock(&mdev->intf_state_mutex);
2312 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &mdev->intf_state)) {
2313 		mutex_unlock(&mdev->intf_state_mutex);
2314 		return NULL;
2315 	}
2316 
2317 	return mdev;
2318 }
2319 EXPORT_SYMBOL(mlx5_vf_get_core_dev);
2320 
2321 /**
2322  * mlx5_vf_put_core_dev - Put the mlx5 core device back.
2323  * @mdev: The mlx5 core device.
2324  *
2325  * Upon return the interface state lock is unlocked and caller should not
2326  * access the mdev any more.
2327  */
2328 void mlx5_vf_put_core_dev(struct mlx5_core_dev *mdev)
2329 {
2330 	mutex_unlock(&mdev->intf_state_mutex);
2331 }
2332 EXPORT_SYMBOL(mlx5_vf_put_core_dev);
2333 
2334 static void mlx5_core_verify_params(void)
2335 {
2336 	if (prof_sel >= ARRAY_SIZE(profile)) {
2337 		pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
2338 			prof_sel,
2339 			ARRAY_SIZE(profile) - 1,
2340 			MLX5_DEFAULT_PROF);
2341 		prof_sel = MLX5_DEFAULT_PROF;
2342 	}
2343 }
2344 
2345 static int __init mlx5_init(void)
2346 {
2347 	int err;
2348 
2349 	WARN_ONCE(strcmp(MLX5_ADEV_NAME, KBUILD_MODNAME),
2350 		  "mlx5_core name not in sync with kernel module name");
2351 
2352 	get_random_bytes(&sw_owner_id, sizeof(sw_owner_id));
2353 
2354 	mlx5_core_verify_params();
2355 	mlx5_register_debugfs();
2356 
2357 	err = mlx5e_init();
2358 	if (err)
2359 		goto err_debug;
2360 
2361 	err = mlx5_sf_driver_register();
2362 	if (err)
2363 		goto err_sf;
2364 
2365 	err = pci_register_driver(&mlx5_core_driver);
2366 	if (err)
2367 		goto err_pci;
2368 
2369 	return 0;
2370 
2371 err_pci:
2372 	mlx5_sf_driver_unregister();
2373 err_sf:
2374 	mlx5e_cleanup();
2375 err_debug:
2376 	mlx5_unregister_debugfs();
2377 	return err;
2378 }
2379 
2380 static void __exit mlx5_cleanup(void)
2381 {
2382 	pci_unregister_driver(&mlx5_core_driver);
2383 	mlx5_sf_driver_unregister();
2384 	mlx5e_cleanup();
2385 	mlx5_unregister_debugfs();
2386 }
2387 
2388 module_init(mlx5_init);
2389 module_exit(mlx5_cleanup);
2390