xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/main.c (revision 9410645520e9b820069761f3450ef6661418e279)
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 
wait_fw_init(struct mlx5_core_dev * dev,u32 max_wait_mili,u32 warn_time_mili,const char * init_state)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 
mlx5_set_driver_version(struct mlx5_core_dev * dev)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 
set_dma_caps(struct pci_dev * pdev)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 
mlx5_pci_enable_device(struct mlx5_core_dev * dev)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 
mlx5_pci_disable_device(struct mlx5_core_dev * dev)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 
request_bar(struct pci_dev * pdev)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 
release_bar(struct pci_dev * pdev)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 
to_fw_pkey_sz(struct mlx5_core_dev * dev,u32 size)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 
mlx5_core_uplink_netdev_set(struct mlx5_core_dev * dev,struct net_device * netdev)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 
mlx5_core_uplink_netdev_event_replay(struct mlx5_core_dev * dev)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 
mlx5_core_mp_event_replay(struct mlx5_core_dev * dev,u32 event,void * data)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 
mlx5_core_get_caps_mode(struct mlx5_core_dev * dev,enum mlx5_cap_type cap_type,enum mlx5_cap_mode cap_mode)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 	memset(in, 0, sizeof(in));
372 	out = kzalloc(out_sz, GFP_KERNEL);
373 	if (!out)
374 		return -ENOMEM;
375 
376 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
377 	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
378 	err = mlx5_cmd_exec_inout(dev, query_hca_cap, in, out);
379 	if (err) {
380 		mlx5_core_warn(dev,
381 			       "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
382 			       cap_type, cap_mode, err);
383 		goto query_ex;
384 	}
385 
386 	hca_caps =  MLX5_ADDR_OF(query_hca_cap_out, out, capability);
387 
388 	switch (cap_mode) {
389 	case HCA_CAP_OPMOD_GET_MAX:
390 		memcpy(dev->caps.hca[cap_type]->max, hca_caps,
391 		       MLX5_UN_SZ_BYTES(hca_cap_union));
392 		break;
393 	case HCA_CAP_OPMOD_GET_CUR:
394 		memcpy(dev->caps.hca[cap_type]->cur, hca_caps,
395 		       MLX5_UN_SZ_BYTES(hca_cap_union));
396 		break;
397 	default:
398 		mlx5_core_warn(dev,
399 			       "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
400 			       cap_type, cap_mode);
401 		err = -EINVAL;
402 		break;
403 	}
404 query_ex:
405 	kfree(out);
406 	return err;
407 }
408 
mlx5_core_get_caps(struct mlx5_core_dev * dev,enum mlx5_cap_type cap_type)409 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
410 {
411 	int ret;
412 
413 	ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
414 	if (ret)
415 		return ret;
416 	return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
417 }
418 
set_caps(struct mlx5_core_dev * dev,void * in,int opmod)419 static int set_caps(struct mlx5_core_dev *dev, void *in, int opmod)
420 {
421 	MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
422 	MLX5_SET(set_hca_cap_in, in, op_mod, opmod << 1);
423 	return mlx5_cmd_exec_in(dev, set_hca_cap, in);
424 }
425 
handle_hca_cap_atomic(struct mlx5_core_dev * dev,void * set_ctx)426 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev, void *set_ctx)
427 {
428 	void *set_hca_cap;
429 	int req_endianness;
430 	int err;
431 
432 	if (!MLX5_CAP_GEN(dev, atomic))
433 		return 0;
434 
435 	err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
436 	if (err)
437 		return err;
438 
439 	req_endianness =
440 		MLX5_CAP_ATOMIC(dev,
441 				supported_atomic_req_8B_endianness_mode_1);
442 
443 	if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
444 		return 0;
445 
446 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
447 
448 	/* Set requestor to host endianness */
449 	MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianness_mode,
450 		 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
451 
452 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ATOMIC);
453 }
454 
handle_hca_cap_odp(struct mlx5_core_dev * dev,void * set_ctx)455 static int handle_hca_cap_odp(struct mlx5_core_dev *dev, void *set_ctx)
456 {
457 	void *set_hca_cap;
458 	bool do_set = false;
459 	int err;
460 
461 	if (!IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) ||
462 	    !MLX5_CAP_GEN(dev, pg))
463 		return 0;
464 
465 	err = mlx5_core_get_caps(dev, MLX5_CAP_ODP);
466 	if (err)
467 		return err;
468 
469 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
470 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_ODP]->cur,
471 	       MLX5_ST_SZ_BYTES(odp_cap));
472 
473 #define ODP_CAP_SET_MAX(dev, field)                                            \
474 	do {                                                                   \
475 		u32 _res = MLX5_CAP_ODP_MAX(dev, field);                       \
476 		if (_res) {                                                    \
477 			do_set = true;                                         \
478 			MLX5_SET(odp_cap, set_hca_cap, field, _res);           \
479 		}                                                              \
480 	} while (0)
481 
482 	ODP_CAP_SET_MAX(dev, ud_odp_caps.srq_receive);
483 	ODP_CAP_SET_MAX(dev, rc_odp_caps.srq_receive);
484 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.srq_receive);
485 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.send);
486 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.receive);
487 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.write);
488 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.read);
489 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.atomic);
490 	ODP_CAP_SET_MAX(dev, dc_odp_caps.srq_receive);
491 	ODP_CAP_SET_MAX(dev, dc_odp_caps.send);
492 	ODP_CAP_SET_MAX(dev, dc_odp_caps.receive);
493 	ODP_CAP_SET_MAX(dev, dc_odp_caps.write);
494 	ODP_CAP_SET_MAX(dev, dc_odp_caps.read);
495 	ODP_CAP_SET_MAX(dev, dc_odp_caps.atomic);
496 
497 	if (!do_set)
498 		return 0;
499 
500 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ODP);
501 }
502 
max_uc_list_get_devlink_param(struct mlx5_core_dev * dev)503 static int max_uc_list_get_devlink_param(struct mlx5_core_dev *dev)
504 {
505 	struct devlink *devlink = priv_to_devlink(dev);
506 	union devlink_param_value val;
507 	int err;
508 
509 	err = devl_param_driverinit_value_get(devlink,
510 					      DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
511 					      &val);
512 	if (!err)
513 		return val.vu32;
514 	mlx5_core_dbg(dev, "Failed to get param. err = %d\n", err);
515 	return err;
516 }
517 
mlx5_is_roce_on(struct mlx5_core_dev * dev)518 bool mlx5_is_roce_on(struct mlx5_core_dev *dev)
519 {
520 	struct devlink *devlink = priv_to_devlink(dev);
521 	union devlink_param_value val;
522 	int err;
523 
524 	err = devl_param_driverinit_value_get(devlink,
525 					      DEVLINK_PARAM_GENERIC_ID_ENABLE_ROCE,
526 					      &val);
527 
528 	if (!err)
529 		return val.vbool;
530 
531 	mlx5_core_dbg(dev, "Failed to get param. err = %d\n", err);
532 	return MLX5_CAP_GEN(dev, roce);
533 }
534 EXPORT_SYMBOL(mlx5_is_roce_on);
535 
handle_hca_cap_2(struct mlx5_core_dev * dev,void * set_ctx)536 static int handle_hca_cap_2(struct mlx5_core_dev *dev, void *set_ctx)
537 {
538 	void *set_hca_cap;
539 	int err;
540 
541 	if (!MLX5_CAP_GEN_MAX(dev, hca_cap_2))
542 		return 0;
543 
544 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL_2);
545 	if (err)
546 		return err;
547 
548 	if (!MLX5_CAP_GEN_2_MAX(dev, sw_vhca_id_valid) ||
549 	    !(dev->priv.sw_vhca_id > 0))
550 		return 0;
551 
552 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
553 				   capability);
554 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_GENERAL_2]->cur,
555 	       MLX5_ST_SZ_BYTES(cmd_hca_cap_2));
556 	MLX5_SET(cmd_hca_cap_2, set_hca_cap, sw_vhca_id_valid, 1);
557 
558 	return set_caps(dev, set_ctx, MLX5_CAP_GENERAL_2);
559 }
560 
handle_hca_cap(struct mlx5_core_dev * dev,void * set_ctx)561 static int handle_hca_cap(struct mlx5_core_dev *dev, void *set_ctx)
562 {
563 	struct mlx5_profile *prof = &dev->profile;
564 	void *set_hca_cap;
565 	int max_uc_list;
566 	int err;
567 
568 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
569 	if (err)
570 		return err;
571 
572 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
573 				   capability);
574 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_GENERAL]->cur,
575 	       MLX5_ST_SZ_BYTES(cmd_hca_cap));
576 
577 	mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
578 		      mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
579 		      128);
580 	/* we limit the size of the pkey table to 128 entries for now */
581 	MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
582 		 to_fw_pkey_sz(dev, 128));
583 
584 	/* Check log_max_qp from HCA caps to set in current profile */
585 	if (prof->log_max_qp == LOG_MAX_SUPPORTED_QPS) {
586 		prof->log_max_qp = min_t(u8, 18, MLX5_CAP_GEN_MAX(dev, log_max_qp));
587 	} else if (MLX5_CAP_GEN_MAX(dev, log_max_qp) < prof->log_max_qp) {
588 		mlx5_core_warn(dev, "log_max_qp value in current profile is %d, changing it to HCA capability limit (%d)\n",
589 			       prof->log_max_qp,
590 			       MLX5_CAP_GEN_MAX(dev, log_max_qp));
591 		prof->log_max_qp = MLX5_CAP_GEN_MAX(dev, log_max_qp);
592 	}
593 	if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
594 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
595 			 prof->log_max_qp);
596 
597 	/* disable cmdif checksum */
598 	MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
599 
600 	/* Enable 4K UAR only when HCA supports it and page size is bigger
601 	 * than 4K.
602 	 */
603 	if (MLX5_CAP_GEN_MAX(dev, uar_4k) && PAGE_SIZE > 4096)
604 		MLX5_SET(cmd_hca_cap, set_hca_cap, uar_4k, 1);
605 
606 	MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
607 
608 	if (MLX5_CAP_GEN_MAX(dev, cache_line_128byte))
609 		MLX5_SET(cmd_hca_cap,
610 			 set_hca_cap,
611 			 cache_line_128byte,
612 			 cache_line_size() >= 128 ? 1 : 0);
613 
614 	if (MLX5_CAP_GEN_MAX(dev, dct))
615 		MLX5_SET(cmd_hca_cap, set_hca_cap, dct, 1);
616 
617 	if (MLX5_CAP_GEN_MAX(dev, pci_sync_for_fw_update_event))
618 		MLX5_SET(cmd_hca_cap, set_hca_cap, pci_sync_for_fw_update_event, 1);
619 	if (MLX5_CAP_GEN_MAX(dev, pci_sync_for_fw_update_with_driver_unload))
620 		MLX5_SET(cmd_hca_cap, set_hca_cap,
621 			 pci_sync_for_fw_update_with_driver_unload, 1);
622 	if (MLX5_CAP_GEN_MAX(dev, pcie_reset_using_hotreset_method))
623 		MLX5_SET(cmd_hca_cap, set_hca_cap,
624 			 pcie_reset_using_hotreset_method, 1);
625 
626 	if (MLX5_CAP_GEN_MAX(dev, num_vhca_ports))
627 		MLX5_SET(cmd_hca_cap,
628 			 set_hca_cap,
629 			 num_vhca_ports,
630 			 MLX5_CAP_GEN_MAX(dev, num_vhca_ports));
631 
632 	if (MLX5_CAP_GEN_MAX(dev, release_all_pages))
633 		MLX5_SET(cmd_hca_cap, set_hca_cap, release_all_pages, 1);
634 
635 	if (MLX5_CAP_GEN_MAX(dev, mkey_by_name))
636 		MLX5_SET(cmd_hca_cap, set_hca_cap, mkey_by_name, 1);
637 
638 	mlx5_vhca_state_cap_handle(dev, set_hca_cap);
639 
640 	if (MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix))
641 		MLX5_SET(cmd_hca_cap, set_hca_cap, num_total_dynamic_vf_msix,
642 			 MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix));
643 
644 	if (MLX5_CAP_GEN(dev, roce_rw_supported) && MLX5_CAP_GEN_MAX(dev, roce))
645 		MLX5_SET(cmd_hca_cap, set_hca_cap, roce,
646 			 mlx5_is_roce_on(dev));
647 
648 	max_uc_list = max_uc_list_get_devlink_param(dev);
649 	if (max_uc_list > 0)
650 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_current_uc_list,
651 			 ilog2(max_uc_list));
652 
653 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
654 }
655 
656 /* Cached MLX5_CAP_GEN(dev, roce) can be out of sync this early in the
657  * boot process.
658  * In case RoCE cap is writable in FW and user/devlink requested to change the
659  * cap, we are yet to query the final state of the above cap.
660  * Hence, the need for this function.
661  *
662  * Returns
663  * True:
664  * 1) RoCE cap is read only in FW and already disabled
665  * OR:
666  * 2) RoCE cap is writable in FW and user/devlink requested it off.
667  *
668  * In any other case, return False.
669  */
is_roce_fw_disabled(struct mlx5_core_dev * dev)670 static bool is_roce_fw_disabled(struct mlx5_core_dev *dev)
671 {
672 	return (MLX5_CAP_GEN(dev, roce_rw_supported) && !mlx5_is_roce_on(dev)) ||
673 		(!MLX5_CAP_GEN(dev, roce_rw_supported) && !MLX5_CAP_GEN(dev, roce));
674 }
675 
handle_hca_cap_roce(struct mlx5_core_dev * dev,void * set_ctx)676 static int handle_hca_cap_roce(struct mlx5_core_dev *dev, void *set_ctx)
677 {
678 	void *set_hca_cap;
679 	int err;
680 
681 	if (is_roce_fw_disabled(dev))
682 		return 0;
683 
684 	err = mlx5_core_get_caps(dev, MLX5_CAP_ROCE);
685 	if (err)
686 		return err;
687 
688 	if (MLX5_CAP_ROCE(dev, sw_r_roce_src_udp_port) ||
689 	    !MLX5_CAP_ROCE_MAX(dev, sw_r_roce_src_udp_port))
690 		return 0;
691 
692 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
693 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_ROCE]->cur,
694 	       MLX5_ST_SZ_BYTES(roce_cap));
695 	MLX5_SET(roce_cap, set_hca_cap, sw_r_roce_src_udp_port, 1);
696 
697 	if (MLX5_CAP_ROCE_MAX(dev, qp_ooo_transmit_default))
698 		MLX5_SET(roce_cap, set_hca_cap, qp_ooo_transmit_default, 1);
699 
700 	err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ROCE);
701 	return err;
702 }
703 
handle_hca_cap_port_selection(struct mlx5_core_dev * dev,void * set_ctx)704 static int handle_hca_cap_port_selection(struct mlx5_core_dev *dev,
705 					 void *set_ctx)
706 {
707 	void *set_hca_cap;
708 	int err;
709 
710 	if (!MLX5_CAP_GEN(dev, port_selection_cap))
711 		return 0;
712 
713 	err = mlx5_core_get_caps(dev, MLX5_CAP_PORT_SELECTION);
714 	if (err)
715 		return err;
716 
717 	if (MLX5_CAP_PORT_SELECTION(dev, port_select_flow_table_bypass) ||
718 	    !MLX5_CAP_PORT_SELECTION_MAX(dev, port_select_flow_table_bypass))
719 		return 0;
720 
721 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
722 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_PORT_SELECTION]->cur,
723 	       MLX5_ST_SZ_BYTES(port_selection_cap));
724 	MLX5_SET(port_selection_cap, set_hca_cap, port_select_flow_table_bypass, 1);
725 
726 	err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_PORT_SELECTION);
727 
728 	return err;
729 }
730 
set_hca_cap(struct mlx5_core_dev * dev)731 static int set_hca_cap(struct mlx5_core_dev *dev)
732 {
733 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
734 	void *set_ctx;
735 	int err;
736 
737 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
738 	if (!set_ctx)
739 		return -ENOMEM;
740 
741 	err = handle_hca_cap(dev, set_ctx);
742 	if (err) {
743 		mlx5_core_err(dev, "handle_hca_cap failed\n");
744 		goto out;
745 	}
746 
747 	memset(set_ctx, 0, set_sz);
748 	err = handle_hca_cap_atomic(dev, set_ctx);
749 	if (err) {
750 		mlx5_core_err(dev, "handle_hca_cap_atomic failed\n");
751 		goto out;
752 	}
753 
754 	memset(set_ctx, 0, set_sz);
755 	err = handle_hca_cap_odp(dev, set_ctx);
756 	if (err) {
757 		mlx5_core_err(dev, "handle_hca_cap_odp failed\n");
758 		goto out;
759 	}
760 
761 	memset(set_ctx, 0, set_sz);
762 	err = handle_hca_cap_roce(dev, set_ctx);
763 	if (err) {
764 		mlx5_core_err(dev, "handle_hca_cap_roce failed\n");
765 		goto out;
766 	}
767 
768 	memset(set_ctx, 0, set_sz);
769 	err = handle_hca_cap_2(dev, set_ctx);
770 	if (err) {
771 		mlx5_core_err(dev, "handle_hca_cap_2 failed\n");
772 		goto out;
773 	}
774 
775 	memset(set_ctx, 0, set_sz);
776 	err = handle_hca_cap_port_selection(dev, set_ctx);
777 	if (err) {
778 		mlx5_core_err(dev, "handle_hca_cap_port_selection failed\n");
779 		goto out;
780 	}
781 
782 out:
783 	kfree(set_ctx);
784 	return err;
785 }
786 
set_hca_ctrl(struct mlx5_core_dev * dev)787 static int set_hca_ctrl(struct mlx5_core_dev *dev)
788 {
789 	struct mlx5_reg_host_endianness he_in;
790 	struct mlx5_reg_host_endianness he_out;
791 	int err;
792 
793 	if (!mlx5_core_is_pf(dev))
794 		return 0;
795 
796 	memset(&he_in, 0, sizeof(he_in));
797 	he_in.he = MLX5_SET_HOST_ENDIANNESS;
798 	err = mlx5_core_access_reg(dev, &he_in,  sizeof(he_in),
799 					&he_out, sizeof(he_out),
800 					MLX5_REG_HOST_ENDIANNESS, 0, 1);
801 	return err;
802 }
803 
mlx5_core_set_hca_defaults(struct mlx5_core_dev * dev)804 static int mlx5_core_set_hca_defaults(struct mlx5_core_dev *dev)
805 {
806 	int ret = 0;
807 
808 	/* Disable local_lb by default */
809 	if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH)
810 		ret = mlx5_nic_vport_update_local_lb(dev, false);
811 
812 	return ret;
813 }
814 
mlx5_core_enable_hca(struct mlx5_core_dev * dev,u16 func_id)815 int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
816 {
817 	u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {};
818 
819 	MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
820 	MLX5_SET(enable_hca_in, in, function_id, func_id);
821 	MLX5_SET(enable_hca_in, in, embedded_cpu_function,
822 		 dev->caps.embedded_cpu);
823 	return mlx5_cmd_exec_in(dev, enable_hca, in);
824 }
825 
mlx5_core_disable_hca(struct mlx5_core_dev * dev,u16 func_id)826 int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
827 {
828 	u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {};
829 
830 	MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
831 	MLX5_SET(disable_hca_in, in, function_id, func_id);
832 	MLX5_SET(enable_hca_in, in, embedded_cpu_function,
833 		 dev->caps.embedded_cpu);
834 	return mlx5_cmd_exec_in(dev, disable_hca, in);
835 }
836 
mlx5_core_set_issi(struct mlx5_core_dev * dev)837 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
838 {
839 	u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {};
840 	u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {};
841 	u32 sup_issi;
842 	int err;
843 
844 	MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
845 	err = mlx5_cmd_exec_inout(dev, query_issi, query_in, query_out);
846 	if (err) {
847 		u32 syndrome = MLX5_GET(query_issi_out, query_out, syndrome);
848 		u8 status = MLX5_GET(query_issi_out, query_out, status);
849 
850 		if (!status || syndrome == MLX5_DRIVER_SYND) {
851 			mlx5_core_err(dev, "Failed to query ISSI err(%d) status(%d) synd(%d)\n",
852 				      err, status, syndrome);
853 			return err;
854 		}
855 
856 		mlx5_core_warn(dev, "Query ISSI is not supported by FW, ISSI is 0\n");
857 		dev->issi = 0;
858 		return 0;
859 	}
860 
861 	sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
862 
863 	if (sup_issi & (1 << 1)) {
864 		u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {};
865 
866 		MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
867 		MLX5_SET(set_issi_in, set_in, current_issi, 1);
868 		err = mlx5_cmd_exec_in(dev, set_issi, set_in);
869 		if (err) {
870 			mlx5_core_err(dev, "Failed to set ISSI to 1 err(%d)\n",
871 				      err);
872 			return err;
873 		}
874 
875 		dev->issi = 1;
876 
877 		return 0;
878 	} else if (sup_issi & (1 << 0) || !sup_issi) {
879 		return 0;
880 	}
881 
882 	return -EOPNOTSUPP;
883 }
884 
mlx5_pci_init(struct mlx5_core_dev * dev,struct pci_dev * pdev,const struct pci_device_id * id)885 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct pci_dev *pdev,
886 			 const struct pci_device_id *id)
887 {
888 	int err = 0;
889 
890 	mutex_init(&dev->pci_status_mutex);
891 	pci_set_drvdata(dev->pdev, dev);
892 
893 	dev->bar_addr = pci_resource_start(pdev, 0);
894 
895 	err = mlx5_pci_enable_device(dev);
896 	if (err) {
897 		mlx5_core_err(dev, "Cannot enable PCI device, aborting\n");
898 		return err;
899 	}
900 
901 	err = request_bar(pdev);
902 	if (err) {
903 		mlx5_core_err(dev, "error requesting BARs, aborting\n");
904 		goto err_disable;
905 	}
906 
907 	pci_set_master(pdev);
908 
909 	err = set_dma_caps(pdev);
910 	if (err) {
911 		mlx5_core_err(dev, "Failed setting DMA capabilities mask, aborting\n");
912 		goto err_clr_master;
913 	}
914 
915 	if (pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP32) &&
916 	    pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP64) &&
917 	    pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP128))
918 		mlx5_core_dbg(dev, "Enabling pci atomics failed\n");
919 
920 	dev->iseg_base = dev->bar_addr;
921 	dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
922 	if (!dev->iseg) {
923 		err = -ENOMEM;
924 		mlx5_core_err(dev, "Failed mapping initialization segment, aborting\n");
925 		goto err_clr_master;
926 	}
927 
928 	mlx5_pci_vsc_init(dev);
929 
930 	err = pci_enable_ptm(pdev, NULL);
931 	if (err)
932 		mlx5_core_info(dev, "PTM is not supported by PCIe\n");
933 
934 	return 0;
935 
936 err_clr_master:
937 	release_bar(dev->pdev);
938 err_disable:
939 	mlx5_pci_disable_device(dev);
940 	return err;
941 }
942 
mlx5_pci_close(struct mlx5_core_dev * dev)943 static void mlx5_pci_close(struct mlx5_core_dev *dev)
944 {
945 	/* health work might still be active, and it needs pci bar in
946 	 * order to know the NIC state. Therefore, drain the health WQ
947 	 * before removing the pci bars
948 	 */
949 	mlx5_drain_health_wq(dev);
950 	pci_disable_ptm(dev->pdev);
951 	iounmap(dev->iseg);
952 	release_bar(dev->pdev);
953 	mlx5_pci_disable_device(dev);
954 }
955 
mlx5_register_hca_devcom_comp(struct mlx5_core_dev * dev)956 static void mlx5_register_hca_devcom_comp(struct mlx5_core_dev *dev)
957 {
958 	/* This component is use to sync adding core_dev to lag_dev and to sync
959 	 * changes of mlx5_adev_devices between LAG layer and other layers.
960 	 */
961 	if (!mlx5_lag_is_supported(dev))
962 		return;
963 
964 	dev->priv.hca_devcom_comp =
965 		mlx5_devcom_register_component(dev->priv.devc, MLX5_DEVCOM_HCA_PORTS,
966 					       mlx5_query_nic_system_image_guid(dev),
967 					       NULL, dev);
968 	if (IS_ERR(dev->priv.hca_devcom_comp))
969 		mlx5_core_err(dev, "Failed to register devcom HCA component\n");
970 }
971 
mlx5_unregister_hca_devcom_comp(struct mlx5_core_dev * dev)972 static void mlx5_unregister_hca_devcom_comp(struct mlx5_core_dev *dev)
973 {
974 	mlx5_devcom_unregister_component(dev->priv.hca_devcom_comp);
975 }
976 
mlx5_init_once(struct mlx5_core_dev * dev)977 static int mlx5_init_once(struct mlx5_core_dev *dev)
978 {
979 	int err;
980 
981 	dev->priv.devc = mlx5_devcom_register_device(dev);
982 	if (IS_ERR(dev->priv.devc))
983 		mlx5_core_warn(dev, "failed to register devcom device %ld\n",
984 			       PTR_ERR(dev->priv.devc));
985 	mlx5_register_hca_devcom_comp(dev);
986 
987 	err = mlx5_query_board_id(dev);
988 	if (err) {
989 		mlx5_core_err(dev, "query board id failed\n");
990 		goto err_devcom;
991 	}
992 
993 	err = mlx5_irq_table_init(dev);
994 	if (err) {
995 		mlx5_core_err(dev, "failed to initialize irq table\n");
996 		goto err_devcom;
997 	}
998 
999 	err = mlx5_eq_table_init(dev);
1000 	if (err) {
1001 		mlx5_core_err(dev, "failed to initialize eq\n");
1002 		goto err_irq_cleanup;
1003 	}
1004 
1005 	err = mlx5_events_init(dev);
1006 	if (err) {
1007 		mlx5_core_err(dev, "failed to initialize events\n");
1008 		goto err_eq_cleanup;
1009 	}
1010 
1011 	err = mlx5_fw_reset_init(dev);
1012 	if (err) {
1013 		mlx5_core_err(dev, "failed to initialize fw reset events\n");
1014 		goto err_events_cleanup;
1015 	}
1016 
1017 	mlx5_cq_debugfs_init(dev);
1018 
1019 	mlx5_init_reserved_gids(dev);
1020 
1021 	mlx5_init_clock(dev);
1022 
1023 	dev->vxlan = mlx5_vxlan_create(dev);
1024 	dev->geneve = mlx5_geneve_create(dev);
1025 
1026 	err = mlx5_init_rl_table(dev);
1027 	if (err) {
1028 		mlx5_core_err(dev, "Failed to init rate limiting\n");
1029 		goto err_tables_cleanup;
1030 	}
1031 
1032 	err = mlx5_mpfs_init(dev);
1033 	if (err) {
1034 		mlx5_core_err(dev, "Failed to init l2 table %d\n", err);
1035 		goto err_rl_cleanup;
1036 	}
1037 
1038 	err = mlx5_sriov_init(dev);
1039 	if (err) {
1040 		mlx5_core_err(dev, "Failed to init sriov %d\n", err);
1041 		goto err_mpfs_cleanup;
1042 	}
1043 
1044 	err = mlx5_eswitch_init(dev);
1045 	if (err) {
1046 		mlx5_core_err(dev, "Failed to init eswitch %d\n", err);
1047 		goto err_sriov_cleanup;
1048 	}
1049 
1050 	err = mlx5_fpga_init(dev);
1051 	if (err) {
1052 		mlx5_core_err(dev, "Failed to init fpga device %d\n", err);
1053 		goto err_eswitch_cleanup;
1054 	}
1055 
1056 	err = mlx5_vhca_event_init(dev);
1057 	if (err) {
1058 		mlx5_core_err(dev, "Failed to init vhca event notifier %d\n", err);
1059 		goto err_fpga_cleanup;
1060 	}
1061 
1062 	err = mlx5_sf_hw_table_init(dev);
1063 	if (err) {
1064 		mlx5_core_err(dev, "Failed to init SF HW table %d\n", err);
1065 		goto err_sf_hw_table_cleanup;
1066 	}
1067 
1068 	err = mlx5_sf_table_init(dev);
1069 	if (err) {
1070 		mlx5_core_err(dev, "Failed to init SF table %d\n", err);
1071 		goto err_sf_table_cleanup;
1072 	}
1073 
1074 	err = mlx5_fs_core_alloc(dev);
1075 	if (err) {
1076 		mlx5_core_err(dev, "Failed to alloc flow steering\n");
1077 		goto err_fs;
1078 	}
1079 
1080 	dev->dm = mlx5_dm_create(dev);
1081 	if (IS_ERR(dev->dm))
1082 		mlx5_core_warn(dev, "Failed to init device memory %ld\n", PTR_ERR(dev->dm));
1083 
1084 	dev->tracer = mlx5_fw_tracer_create(dev);
1085 	dev->hv_vhca = mlx5_hv_vhca_create(dev);
1086 	dev->rsc_dump = mlx5_rsc_dump_create(dev);
1087 
1088 	return 0;
1089 
1090 err_fs:
1091 	mlx5_sf_table_cleanup(dev);
1092 err_sf_table_cleanup:
1093 	mlx5_sf_hw_table_cleanup(dev);
1094 err_sf_hw_table_cleanup:
1095 	mlx5_vhca_event_cleanup(dev);
1096 err_fpga_cleanup:
1097 	mlx5_fpga_cleanup(dev);
1098 err_eswitch_cleanup:
1099 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1100 err_sriov_cleanup:
1101 	mlx5_sriov_cleanup(dev);
1102 err_mpfs_cleanup:
1103 	mlx5_mpfs_cleanup(dev);
1104 err_rl_cleanup:
1105 	mlx5_cleanup_rl_table(dev);
1106 err_tables_cleanup:
1107 	mlx5_geneve_destroy(dev->geneve);
1108 	mlx5_vxlan_destroy(dev->vxlan);
1109 	mlx5_cleanup_clock(dev);
1110 	mlx5_cleanup_reserved_gids(dev);
1111 	mlx5_cq_debugfs_cleanup(dev);
1112 	mlx5_fw_reset_cleanup(dev);
1113 err_events_cleanup:
1114 	mlx5_events_cleanup(dev);
1115 err_eq_cleanup:
1116 	mlx5_eq_table_cleanup(dev);
1117 err_irq_cleanup:
1118 	mlx5_irq_table_cleanup(dev);
1119 err_devcom:
1120 	mlx5_unregister_hca_devcom_comp(dev);
1121 	mlx5_devcom_unregister_device(dev->priv.devc);
1122 
1123 	return err;
1124 }
1125 
mlx5_cleanup_once(struct mlx5_core_dev * dev)1126 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
1127 {
1128 	mlx5_rsc_dump_destroy(dev);
1129 	mlx5_hv_vhca_destroy(dev->hv_vhca);
1130 	mlx5_fw_tracer_destroy(dev->tracer);
1131 	mlx5_dm_cleanup(dev);
1132 	mlx5_fs_core_free(dev);
1133 	mlx5_sf_table_cleanup(dev);
1134 	mlx5_sf_hw_table_cleanup(dev);
1135 	mlx5_vhca_event_cleanup(dev);
1136 	mlx5_fpga_cleanup(dev);
1137 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1138 	mlx5_sriov_cleanup(dev);
1139 	mlx5_mpfs_cleanup(dev);
1140 	mlx5_cleanup_rl_table(dev);
1141 	mlx5_geneve_destroy(dev->geneve);
1142 	mlx5_vxlan_destroy(dev->vxlan);
1143 	mlx5_cleanup_clock(dev);
1144 	mlx5_cleanup_reserved_gids(dev);
1145 	mlx5_cq_debugfs_cleanup(dev);
1146 	mlx5_fw_reset_cleanup(dev);
1147 	mlx5_events_cleanup(dev);
1148 	mlx5_eq_table_cleanup(dev);
1149 	mlx5_irq_table_cleanup(dev);
1150 	mlx5_unregister_hca_devcom_comp(dev);
1151 	mlx5_devcom_unregister_device(dev->priv.devc);
1152 }
1153 
mlx5_function_enable(struct mlx5_core_dev * dev,bool boot,u64 timeout)1154 static int mlx5_function_enable(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1155 {
1156 	int err;
1157 
1158 	mlx5_core_info(dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1159 		       fw_rev_min(dev), fw_rev_sub(dev));
1160 
1161 	/* Only PFs hold the relevant PCIe information for this query */
1162 	if (mlx5_core_is_pf(dev))
1163 		pcie_print_link_status(dev->pdev);
1164 
1165 	/* wait for firmware to accept initialization segments configurations
1166 	 */
1167 	err = wait_fw_init(dev, timeout,
1168 			   mlx5_tout_ms(dev, FW_PRE_INIT_WARN_MESSAGE_INTERVAL),
1169 			   "pre-initializing");
1170 	if (err)
1171 		return err;
1172 
1173 	err = mlx5_cmd_enable(dev);
1174 	if (err) {
1175 		mlx5_core_err(dev, "Failed initializing command interface, aborting\n");
1176 		return err;
1177 	}
1178 
1179 	mlx5_tout_query_iseg(dev);
1180 
1181 	err = wait_fw_init(dev, mlx5_tout_ms(dev, FW_INIT), 0, "initializing");
1182 	if (err)
1183 		goto err_cmd_cleanup;
1184 
1185 	dev->caps.embedded_cpu = mlx5_read_embedded_cpu(dev);
1186 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_UP);
1187 
1188 	mlx5_start_health_poll(dev);
1189 
1190 	err = mlx5_core_enable_hca(dev, 0);
1191 	if (err) {
1192 		mlx5_core_err(dev, "enable hca failed\n");
1193 		goto stop_health_poll;
1194 	}
1195 
1196 	err = mlx5_core_set_issi(dev);
1197 	if (err) {
1198 		mlx5_core_err(dev, "failed to set issi\n");
1199 		goto err_disable_hca;
1200 	}
1201 
1202 	err = mlx5_satisfy_startup_pages(dev, 1);
1203 	if (err) {
1204 		mlx5_core_err(dev, "failed to allocate boot pages\n");
1205 		goto err_disable_hca;
1206 	}
1207 
1208 	err = mlx5_tout_query_dtor(dev);
1209 	if (err) {
1210 		mlx5_core_err(dev, "failed to read dtor\n");
1211 		goto reclaim_boot_pages;
1212 	}
1213 
1214 	return 0;
1215 
1216 reclaim_boot_pages:
1217 	mlx5_reclaim_startup_pages(dev);
1218 err_disable_hca:
1219 	mlx5_core_disable_hca(dev, 0);
1220 stop_health_poll:
1221 	mlx5_stop_health_poll(dev, boot);
1222 err_cmd_cleanup:
1223 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1224 	mlx5_cmd_disable(dev);
1225 
1226 	return err;
1227 }
1228 
mlx5_function_disable(struct mlx5_core_dev * dev,bool boot)1229 static void mlx5_function_disable(struct mlx5_core_dev *dev, bool boot)
1230 {
1231 	mlx5_reclaim_startup_pages(dev);
1232 	mlx5_core_disable_hca(dev, 0);
1233 	mlx5_stop_health_poll(dev, boot);
1234 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1235 	mlx5_cmd_disable(dev);
1236 }
1237 
mlx5_function_open(struct mlx5_core_dev * dev)1238 static int mlx5_function_open(struct mlx5_core_dev *dev)
1239 {
1240 	int err;
1241 
1242 	err = set_hca_ctrl(dev);
1243 	if (err) {
1244 		mlx5_core_err(dev, "set_hca_ctrl failed\n");
1245 		return err;
1246 	}
1247 
1248 	err = set_hca_cap(dev);
1249 	if (err) {
1250 		mlx5_core_err(dev, "set_hca_cap failed\n");
1251 		return err;
1252 	}
1253 
1254 	err = mlx5_satisfy_startup_pages(dev, 0);
1255 	if (err) {
1256 		mlx5_core_err(dev, "failed to allocate init pages\n");
1257 		return err;
1258 	}
1259 
1260 	err = mlx5_cmd_init_hca(dev, sw_owner_id);
1261 	if (err) {
1262 		mlx5_core_err(dev, "init hca failed\n");
1263 		return err;
1264 	}
1265 
1266 	mlx5_set_driver_version(dev);
1267 
1268 	err = mlx5_query_hca_caps(dev);
1269 	if (err) {
1270 		mlx5_core_err(dev, "query hca failed\n");
1271 		return err;
1272 	}
1273 	mlx5_start_health_fw_log_up(dev);
1274 	return 0;
1275 }
1276 
mlx5_function_close(struct mlx5_core_dev * dev)1277 static int mlx5_function_close(struct mlx5_core_dev *dev)
1278 {
1279 	int err;
1280 
1281 	err = mlx5_cmd_teardown_hca(dev);
1282 	if (err) {
1283 		mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n");
1284 		return err;
1285 	}
1286 
1287 	return 0;
1288 }
1289 
mlx5_function_setup(struct mlx5_core_dev * dev,bool boot,u64 timeout)1290 static int mlx5_function_setup(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1291 {
1292 	int err;
1293 
1294 	err = mlx5_function_enable(dev, boot, timeout);
1295 	if (err)
1296 		return err;
1297 
1298 	err = mlx5_function_open(dev);
1299 	if (err)
1300 		mlx5_function_disable(dev, boot);
1301 	return err;
1302 }
1303 
mlx5_function_teardown(struct mlx5_core_dev * dev,bool boot)1304 static int mlx5_function_teardown(struct mlx5_core_dev *dev, bool boot)
1305 {
1306 	int err = mlx5_function_close(dev);
1307 
1308 	if (!err)
1309 		mlx5_function_disable(dev, boot);
1310 	else
1311 		mlx5_stop_health_poll(dev, boot);
1312 
1313 	return err;
1314 }
1315 
mlx5_load(struct mlx5_core_dev * dev)1316 static int mlx5_load(struct mlx5_core_dev *dev)
1317 {
1318 	int err;
1319 
1320 	dev->priv.uar = mlx5_get_uars_page(dev);
1321 	if (IS_ERR(dev->priv.uar)) {
1322 		mlx5_core_err(dev, "Failed allocating uar, aborting\n");
1323 		err = PTR_ERR(dev->priv.uar);
1324 		return err;
1325 	}
1326 
1327 	mlx5_events_start(dev);
1328 	mlx5_pagealloc_start(dev);
1329 
1330 	err = mlx5_irq_table_create(dev);
1331 	if (err) {
1332 		mlx5_core_err(dev, "Failed to alloc IRQs\n");
1333 		goto err_irq_table;
1334 	}
1335 
1336 	err = mlx5_eq_table_create(dev);
1337 	if (err) {
1338 		mlx5_core_err(dev, "Failed to create EQs\n");
1339 		goto err_eq_table;
1340 	}
1341 
1342 	err = mlx5_fw_tracer_init(dev->tracer);
1343 	if (err) {
1344 		mlx5_core_err(dev, "Failed to init FW tracer %d\n", err);
1345 		mlx5_fw_tracer_destroy(dev->tracer);
1346 		dev->tracer = NULL;
1347 	}
1348 
1349 	mlx5_fw_reset_events_start(dev);
1350 	mlx5_hv_vhca_init(dev->hv_vhca);
1351 
1352 	err = mlx5_rsc_dump_init(dev);
1353 	if (err) {
1354 		mlx5_core_err(dev, "Failed to init Resource dump %d\n", err);
1355 		mlx5_rsc_dump_destroy(dev);
1356 		dev->rsc_dump = NULL;
1357 	}
1358 
1359 	err = mlx5_fpga_device_start(dev);
1360 	if (err) {
1361 		mlx5_core_err(dev, "fpga device start failed %d\n", err);
1362 		goto err_fpga_start;
1363 	}
1364 
1365 	err = mlx5_fs_core_init(dev);
1366 	if (err) {
1367 		mlx5_core_err(dev, "Failed to init flow steering\n");
1368 		goto err_fs;
1369 	}
1370 
1371 	err = mlx5_core_set_hca_defaults(dev);
1372 	if (err) {
1373 		mlx5_core_err(dev, "Failed to set hca defaults\n");
1374 		goto err_set_hca;
1375 	}
1376 
1377 	mlx5_vhca_event_start(dev);
1378 
1379 	err = mlx5_sf_hw_table_create(dev);
1380 	if (err) {
1381 		mlx5_core_err(dev, "sf table create failed %d\n", err);
1382 		goto err_vhca;
1383 	}
1384 
1385 	err = mlx5_ec_init(dev);
1386 	if (err) {
1387 		mlx5_core_err(dev, "Failed to init embedded CPU\n");
1388 		goto err_ec;
1389 	}
1390 
1391 	mlx5_lag_add_mdev(dev);
1392 	err = mlx5_sriov_attach(dev);
1393 	if (err) {
1394 		mlx5_core_err(dev, "sriov init failed %d\n", err);
1395 		goto err_sriov;
1396 	}
1397 
1398 	mlx5_sf_dev_table_create(dev);
1399 
1400 	err = mlx5_devlink_traps_register(priv_to_devlink(dev));
1401 	if (err)
1402 		goto err_traps_reg;
1403 
1404 	return 0;
1405 
1406 err_traps_reg:
1407 	mlx5_sf_dev_table_destroy(dev);
1408 	mlx5_sriov_detach(dev);
1409 err_sriov:
1410 	mlx5_lag_remove_mdev(dev);
1411 	mlx5_ec_cleanup(dev);
1412 err_ec:
1413 	mlx5_sf_hw_table_destroy(dev);
1414 err_vhca:
1415 	mlx5_vhca_event_stop(dev);
1416 err_set_hca:
1417 	mlx5_fs_core_cleanup(dev);
1418 err_fs:
1419 	mlx5_fpga_device_stop(dev);
1420 err_fpga_start:
1421 	mlx5_rsc_dump_cleanup(dev);
1422 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1423 	mlx5_fw_reset_events_stop(dev);
1424 	mlx5_fw_tracer_cleanup(dev->tracer);
1425 	mlx5_eq_table_destroy(dev);
1426 err_eq_table:
1427 	mlx5_irq_table_destroy(dev);
1428 err_irq_table:
1429 	mlx5_pagealloc_stop(dev);
1430 	mlx5_events_stop(dev);
1431 	mlx5_put_uars_page(dev, dev->priv.uar);
1432 	return err;
1433 }
1434 
mlx5_unload(struct mlx5_core_dev * dev)1435 static void mlx5_unload(struct mlx5_core_dev *dev)
1436 {
1437 	mlx5_eswitch_disable(dev->priv.eswitch);
1438 	mlx5_devlink_traps_unregister(priv_to_devlink(dev));
1439 	mlx5_sf_dev_table_destroy(dev);
1440 	mlx5_sriov_detach(dev);
1441 	mlx5_lag_remove_mdev(dev);
1442 	mlx5_ec_cleanup(dev);
1443 	mlx5_sf_hw_table_destroy(dev);
1444 	mlx5_vhca_event_stop(dev);
1445 	mlx5_fs_core_cleanup(dev);
1446 	mlx5_fpga_device_stop(dev);
1447 	mlx5_rsc_dump_cleanup(dev);
1448 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1449 	mlx5_fw_reset_events_stop(dev);
1450 	mlx5_fw_tracer_cleanup(dev->tracer);
1451 	mlx5_eq_table_destroy(dev);
1452 	mlx5_irq_table_destroy(dev);
1453 	mlx5_pagealloc_stop(dev);
1454 	mlx5_events_stop(dev);
1455 	mlx5_put_uars_page(dev, dev->priv.uar);
1456 }
1457 
mlx5_init_one_devl_locked(struct mlx5_core_dev * dev)1458 int mlx5_init_one_devl_locked(struct mlx5_core_dev *dev)
1459 {
1460 	bool light_probe = mlx5_dev_is_lightweight(dev);
1461 	int err = 0;
1462 
1463 	mutex_lock(&dev->intf_state_mutex);
1464 	dev->state = MLX5_DEVICE_STATE_UP;
1465 
1466 	err = mlx5_function_setup(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1467 	if (err)
1468 		goto err_function;
1469 
1470 	err = mlx5_init_once(dev);
1471 	if (err) {
1472 		mlx5_core_err(dev, "sw objs init failed\n");
1473 		goto function_teardown;
1474 	}
1475 
1476 	/* In case of light_probe, mlx5_devlink is already registered.
1477 	 * Hence, don't register devlink again.
1478 	 */
1479 	if (!light_probe) {
1480 		err = mlx5_devlink_params_register(priv_to_devlink(dev));
1481 		if (err)
1482 			goto err_devlink_params_reg;
1483 	}
1484 
1485 	err = mlx5_load(dev);
1486 	if (err)
1487 		goto err_load;
1488 
1489 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1490 
1491 	err = mlx5_register_device(dev);
1492 	if (err)
1493 		goto err_register;
1494 
1495 	err = mlx5_crdump_enable(dev);
1496 	if (err)
1497 		mlx5_core_err(dev, "mlx5_crdump_enable failed with error code %d\n", err);
1498 
1499 	err = mlx5_hwmon_dev_register(dev);
1500 	if (err)
1501 		mlx5_core_err(dev, "mlx5_hwmon_dev_register failed with error code %d\n", err);
1502 
1503 	mutex_unlock(&dev->intf_state_mutex);
1504 	return 0;
1505 
1506 err_register:
1507 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1508 	mlx5_unload(dev);
1509 err_load:
1510 	if (!light_probe)
1511 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1512 err_devlink_params_reg:
1513 	mlx5_cleanup_once(dev);
1514 function_teardown:
1515 	mlx5_function_teardown(dev, true);
1516 err_function:
1517 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1518 	mutex_unlock(&dev->intf_state_mutex);
1519 	return err;
1520 }
1521 
mlx5_init_one(struct mlx5_core_dev * dev)1522 int mlx5_init_one(struct mlx5_core_dev *dev)
1523 {
1524 	struct devlink *devlink = priv_to_devlink(dev);
1525 	int err;
1526 
1527 	devl_lock(devlink);
1528 	devl_register(devlink);
1529 	err = mlx5_init_one_devl_locked(dev);
1530 	if (err)
1531 		devl_unregister(devlink);
1532 	devl_unlock(devlink);
1533 	return err;
1534 }
1535 
mlx5_uninit_one(struct mlx5_core_dev * dev)1536 void mlx5_uninit_one(struct mlx5_core_dev *dev)
1537 {
1538 	struct devlink *devlink = priv_to_devlink(dev);
1539 
1540 	devl_lock(devlink);
1541 	mutex_lock(&dev->intf_state_mutex);
1542 
1543 	mlx5_hwmon_dev_unregister(dev);
1544 	mlx5_crdump_disable(dev);
1545 	mlx5_unregister_device(dev);
1546 
1547 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1548 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1549 			       __func__);
1550 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1551 		mlx5_cleanup_once(dev);
1552 		goto out;
1553 	}
1554 
1555 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1556 	mlx5_unload(dev);
1557 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1558 	mlx5_cleanup_once(dev);
1559 	mlx5_function_teardown(dev, true);
1560 out:
1561 	mutex_unlock(&dev->intf_state_mutex);
1562 	devl_unregister(devlink);
1563 	devl_unlock(devlink);
1564 }
1565 
mlx5_load_one_devl_locked(struct mlx5_core_dev * dev,bool recovery)1566 int mlx5_load_one_devl_locked(struct mlx5_core_dev *dev, bool recovery)
1567 {
1568 	int err = 0;
1569 	u64 timeout;
1570 
1571 	devl_assert_locked(priv_to_devlink(dev));
1572 	mutex_lock(&dev->intf_state_mutex);
1573 	if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1574 		mlx5_core_warn(dev, "interface is up, NOP\n");
1575 		goto out;
1576 	}
1577 	/* remove any previous indication of internal error */
1578 	dev->state = MLX5_DEVICE_STATE_UP;
1579 
1580 	if (recovery)
1581 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_ON_RECOVERY_TIMEOUT);
1582 	else
1583 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT);
1584 	err = mlx5_function_setup(dev, false, timeout);
1585 	if (err)
1586 		goto err_function;
1587 
1588 	err = mlx5_load(dev);
1589 	if (err)
1590 		goto err_load;
1591 
1592 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1593 
1594 	err = mlx5_attach_device(dev);
1595 	if (err)
1596 		goto err_attach;
1597 
1598 	mutex_unlock(&dev->intf_state_mutex);
1599 	return 0;
1600 
1601 err_attach:
1602 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1603 	mlx5_unload(dev);
1604 err_load:
1605 	mlx5_function_teardown(dev, false);
1606 err_function:
1607 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1608 out:
1609 	mutex_unlock(&dev->intf_state_mutex);
1610 	return err;
1611 }
1612 
mlx5_load_one(struct mlx5_core_dev * dev,bool recovery)1613 int mlx5_load_one(struct mlx5_core_dev *dev, bool recovery)
1614 {
1615 	struct devlink *devlink = priv_to_devlink(dev);
1616 	int ret;
1617 
1618 	devl_lock(devlink);
1619 	ret = mlx5_load_one_devl_locked(dev, recovery);
1620 	devl_unlock(devlink);
1621 	return ret;
1622 }
1623 
mlx5_unload_one_devl_locked(struct mlx5_core_dev * dev,bool suspend)1624 void mlx5_unload_one_devl_locked(struct mlx5_core_dev *dev, bool suspend)
1625 {
1626 	devl_assert_locked(priv_to_devlink(dev));
1627 	mutex_lock(&dev->intf_state_mutex);
1628 
1629 	mlx5_detach_device(dev, suspend);
1630 
1631 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1632 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1633 			       __func__);
1634 		goto out;
1635 	}
1636 
1637 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1638 	mlx5_unload(dev);
1639 	mlx5_function_teardown(dev, false);
1640 out:
1641 	mutex_unlock(&dev->intf_state_mutex);
1642 }
1643 
mlx5_unload_one(struct mlx5_core_dev * dev,bool suspend)1644 void mlx5_unload_one(struct mlx5_core_dev *dev, bool suspend)
1645 {
1646 	struct devlink *devlink = priv_to_devlink(dev);
1647 
1648 	devl_lock(devlink);
1649 	mlx5_unload_one_devl_locked(dev, suspend);
1650 	devl_unlock(devlink);
1651 }
1652 
1653 /* In case of light probe, we don't need a full query of hca_caps, but only the bellow caps.
1654  * A full query of hca_caps will be done when the device will reload.
1655  */
mlx5_query_hca_caps_light(struct mlx5_core_dev * dev)1656 static int mlx5_query_hca_caps_light(struct mlx5_core_dev *dev)
1657 {
1658 	int err;
1659 
1660 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
1661 	if (err)
1662 		return err;
1663 
1664 	if (MLX5_CAP_GEN(dev, eth_net_offloads)) {
1665 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_ETHERNET_OFFLOADS,
1666 					      HCA_CAP_OPMOD_GET_CUR);
1667 		if (err)
1668 			return err;
1669 	}
1670 
1671 	if (MLX5_CAP_GEN(dev, nic_flow_table) ||
1672 	    MLX5_CAP_GEN(dev, ipoib_enhanced_offloads)) {
1673 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_FLOW_TABLE,
1674 					      HCA_CAP_OPMOD_GET_CUR);
1675 		if (err)
1676 			return err;
1677 	}
1678 
1679 	if (MLX5_CAP_GEN_64(dev, general_obj_types) &
1680 		MLX5_GENERAL_OBJ_TYPES_CAP_VIRTIO_NET_Q) {
1681 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_VDPA_EMULATION,
1682 					      HCA_CAP_OPMOD_GET_CUR);
1683 		if (err)
1684 			return err;
1685 	}
1686 
1687 	return 0;
1688 }
1689 
mlx5_init_one_light(struct mlx5_core_dev * dev)1690 int mlx5_init_one_light(struct mlx5_core_dev *dev)
1691 {
1692 	struct devlink *devlink = priv_to_devlink(dev);
1693 	int err;
1694 
1695 	devl_lock(devlink);
1696 	devl_register(devlink);
1697 	dev->state = MLX5_DEVICE_STATE_UP;
1698 	err = mlx5_function_enable(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1699 	if (err) {
1700 		mlx5_core_warn(dev, "mlx5_function_enable err=%d\n", err);
1701 		goto out;
1702 	}
1703 
1704 	err = mlx5_query_hca_caps_light(dev);
1705 	if (err) {
1706 		mlx5_core_warn(dev, "mlx5_query_hca_caps_light err=%d\n", err);
1707 		goto query_hca_caps_err;
1708 	}
1709 
1710 	err = mlx5_devlink_params_register(priv_to_devlink(dev));
1711 	if (err) {
1712 		mlx5_core_warn(dev, "mlx5_devlink_param_reg err = %d\n", err);
1713 		goto query_hca_caps_err;
1714 	}
1715 
1716 	devl_unlock(devlink);
1717 	return 0;
1718 
1719 query_hca_caps_err:
1720 	mlx5_function_disable(dev, true);
1721 out:
1722 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1723 	devl_unregister(devlink);
1724 	devl_unlock(devlink);
1725 	return err;
1726 }
1727 
mlx5_uninit_one_light(struct mlx5_core_dev * dev)1728 void mlx5_uninit_one_light(struct mlx5_core_dev *dev)
1729 {
1730 	struct devlink *devlink = priv_to_devlink(dev);
1731 
1732 	devl_lock(devlink);
1733 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1734 	devl_unregister(devlink);
1735 	devl_unlock(devlink);
1736 	if (dev->state != MLX5_DEVICE_STATE_UP)
1737 		return;
1738 	mlx5_function_disable(dev, true);
1739 }
1740 
1741 /* xxx_light() function are used in order to configure the device without full
1742  * init (light init). e.g.: There isn't a point in reload a device to light state.
1743  * Hence, mlx5_load_one_light() isn't needed.
1744  */
1745 
mlx5_unload_one_light(struct mlx5_core_dev * dev)1746 void mlx5_unload_one_light(struct mlx5_core_dev *dev)
1747 {
1748 	if (dev->state != MLX5_DEVICE_STATE_UP)
1749 		return;
1750 	mlx5_function_disable(dev, false);
1751 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1752 }
1753 
1754 static const int types[] = {
1755 	MLX5_CAP_GENERAL,
1756 	MLX5_CAP_GENERAL_2,
1757 	MLX5_CAP_ETHERNET_OFFLOADS,
1758 	MLX5_CAP_IPOIB_ENHANCED_OFFLOADS,
1759 	MLX5_CAP_ODP,
1760 	MLX5_CAP_ATOMIC,
1761 	MLX5_CAP_ROCE,
1762 	MLX5_CAP_IPOIB_OFFLOADS,
1763 	MLX5_CAP_FLOW_TABLE,
1764 	MLX5_CAP_ESWITCH_FLOW_TABLE,
1765 	MLX5_CAP_ESWITCH,
1766 	MLX5_CAP_QOS,
1767 	MLX5_CAP_DEBUG,
1768 	MLX5_CAP_DEV_MEM,
1769 	MLX5_CAP_DEV_EVENT,
1770 	MLX5_CAP_TLS,
1771 	MLX5_CAP_VDPA_EMULATION,
1772 	MLX5_CAP_IPSEC,
1773 	MLX5_CAP_PORT_SELECTION,
1774 	MLX5_CAP_MACSEC,
1775 	MLX5_CAP_ADV_VIRTUALIZATION,
1776 	MLX5_CAP_CRYPTO,
1777 };
1778 
mlx5_hca_caps_free(struct mlx5_core_dev * dev)1779 static void mlx5_hca_caps_free(struct mlx5_core_dev *dev)
1780 {
1781 	int type;
1782 	int i;
1783 
1784 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1785 		type = types[i];
1786 		kfree(dev->caps.hca[type]);
1787 	}
1788 }
1789 
mlx5_hca_caps_alloc(struct mlx5_core_dev * dev)1790 static int mlx5_hca_caps_alloc(struct mlx5_core_dev *dev)
1791 {
1792 	struct mlx5_hca_cap *cap;
1793 	int type;
1794 	int i;
1795 
1796 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1797 		cap = kzalloc(sizeof(*cap), GFP_KERNEL);
1798 		if (!cap)
1799 			goto err;
1800 		type = types[i];
1801 		dev->caps.hca[type] = cap;
1802 	}
1803 
1804 	return 0;
1805 
1806 err:
1807 	mlx5_hca_caps_free(dev);
1808 	return -ENOMEM;
1809 }
1810 
vhca_id_show(struct seq_file * file,void * priv)1811 static int vhca_id_show(struct seq_file *file, void *priv)
1812 {
1813 	struct mlx5_core_dev *dev = file->private;
1814 
1815 	seq_printf(file, "0x%x\n", MLX5_CAP_GEN(dev, vhca_id));
1816 	return 0;
1817 }
1818 
1819 DEFINE_SHOW_ATTRIBUTE(vhca_id);
1820 
mlx5_mdev_init(struct mlx5_core_dev * dev,int profile_idx)1821 int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
1822 {
1823 	struct mlx5_priv *priv = &dev->priv;
1824 	int err;
1825 
1826 	memcpy(&dev->profile, &profile[profile_idx], sizeof(dev->profile));
1827 	lockdep_register_key(&dev->lock_key);
1828 	mutex_init(&dev->intf_state_mutex);
1829 	lockdep_set_class(&dev->intf_state_mutex, &dev->lock_key);
1830 	mutex_init(&dev->mlx5e_res.uplink_netdev_lock);
1831 	mutex_init(&dev->wc_state_lock);
1832 
1833 	mutex_init(&priv->bfregs.reg_head.lock);
1834 	mutex_init(&priv->bfregs.wc_head.lock);
1835 	INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1836 	INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1837 
1838 	mutex_init(&priv->alloc_mutex);
1839 	mutex_init(&priv->pgdir_mutex);
1840 	INIT_LIST_HEAD(&priv->pgdir_list);
1841 
1842 	priv->numa_node = dev_to_node(mlx5_core_dma_dev(dev));
1843 	priv->dbg.dbg_root = debugfs_create_dir(dev_name(dev->device),
1844 						mlx5_debugfs_root);
1845 	debugfs_create_file("vhca_id", 0400, priv->dbg.dbg_root, dev, &vhca_id_fops);
1846 	INIT_LIST_HEAD(&priv->traps);
1847 
1848 	err = mlx5_cmd_init(dev);
1849 	if (err) {
1850 		mlx5_core_err(dev, "Failed initializing cmdif SW structs, aborting\n");
1851 		goto err_cmd_init;
1852 	}
1853 
1854 	err = mlx5_tout_init(dev);
1855 	if (err) {
1856 		mlx5_core_err(dev, "Failed initializing timeouts, aborting\n");
1857 		goto err_timeout_init;
1858 	}
1859 
1860 	err = mlx5_health_init(dev);
1861 	if (err)
1862 		goto err_health_init;
1863 
1864 	err = mlx5_pagealloc_init(dev);
1865 	if (err)
1866 		goto err_pagealloc_init;
1867 
1868 	err = mlx5_adev_init(dev);
1869 	if (err)
1870 		goto err_adev_init;
1871 
1872 	err = mlx5_hca_caps_alloc(dev);
1873 	if (err)
1874 		goto err_hca_caps;
1875 
1876 	/* The conjunction of sw_vhca_id with sw_owner_id will be a global
1877 	 * unique id per function which uses mlx5_core.
1878 	 * Those values are supplied to FW as part of the init HCA command to
1879 	 * be used by both driver and FW when it's applicable.
1880 	 */
1881 	dev->priv.sw_vhca_id = ida_alloc_range(&sw_vhca_ida, 1,
1882 					       MAX_SW_VHCA_ID,
1883 					       GFP_KERNEL);
1884 	if (dev->priv.sw_vhca_id < 0)
1885 		mlx5_core_err(dev, "failed to allocate sw_vhca_id, err=%d\n",
1886 			      dev->priv.sw_vhca_id);
1887 
1888 	return 0;
1889 
1890 err_hca_caps:
1891 	mlx5_adev_cleanup(dev);
1892 err_adev_init:
1893 	mlx5_pagealloc_cleanup(dev);
1894 err_pagealloc_init:
1895 	mlx5_health_cleanup(dev);
1896 err_health_init:
1897 	mlx5_tout_cleanup(dev);
1898 err_timeout_init:
1899 	mlx5_cmd_cleanup(dev);
1900 err_cmd_init:
1901 	debugfs_remove(dev->priv.dbg.dbg_root);
1902 	mutex_destroy(&priv->pgdir_mutex);
1903 	mutex_destroy(&priv->alloc_mutex);
1904 	mutex_destroy(&priv->bfregs.wc_head.lock);
1905 	mutex_destroy(&priv->bfregs.reg_head.lock);
1906 	mutex_destroy(&dev->intf_state_mutex);
1907 	lockdep_unregister_key(&dev->lock_key);
1908 	return err;
1909 }
1910 
mlx5_mdev_uninit(struct mlx5_core_dev * dev)1911 void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
1912 {
1913 	struct mlx5_priv *priv = &dev->priv;
1914 
1915 	if (priv->sw_vhca_id > 0)
1916 		ida_free(&sw_vhca_ida, dev->priv.sw_vhca_id);
1917 
1918 	mlx5_hca_caps_free(dev);
1919 	mlx5_adev_cleanup(dev);
1920 	mlx5_pagealloc_cleanup(dev);
1921 	mlx5_health_cleanup(dev);
1922 	mlx5_tout_cleanup(dev);
1923 	mlx5_cmd_cleanup(dev);
1924 	debugfs_remove_recursive(dev->priv.dbg.dbg_root);
1925 	mutex_destroy(&priv->pgdir_mutex);
1926 	mutex_destroy(&priv->alloc_mutex);
1927 	mutex_destroy(&priv->bfregs.wc_head.lock);
1928 	mutex_destroy(&priv->bfregs.reg_head.lock);
1929 	mutex_destroy(&dev->wc_state_lock);
1930 	mutex_destroy(&dev->mlx5e_res.uplink_netdev_lock);
1931 	mutex_destroy(&dev->intf_state_mutex);
1932 	lockdep_unregister_key(&dev->lock_key);
1933 }
1934 
probe_one(struct pci_dev * pdev,const struct pci_device_id * id)1935 static int probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
1936 {
1937 	struct mlx5_core_dev *dev;
1938 	struct devlink *devlink;
1939 	int err;
1940 
1941 	devlink = mlx5_devlink_alloc(&pdev->dev);
1942 	if (!devlink) {
1943 		dev_err(&pdev->dev, "devlink alloc failed\n");
1944 		return -ENOMEM;
1945 	}
1946 
1947 	dev = devlink_priv(devlink);
1948 	dev->device = &pdev->dev;
1949 	dev->pdev = pdev;
1950 
1951 	dev->coredev_type = id->driver_data & MLX5_PCI_DEV_IS_VF ?
1952 			 MLX5_COREDEV_VF : MLX5_COREDEV_PF;
1953 
1954 	dev->priv.adev_idx = mlx5_adev_idx_alloc();
1955 	if (dev->priv.adev_idx < 0) {
1956 		err = dev->priv.adev_idx;
1957 		goto adev_init_err;
1958 	}
1959 
1960 	err = mlx5_mdev_init(dev, prof_sel);
1961 	if (err)
1962 		goto mdev_init_err;
1963 
1964 	err = mlx5_pci_init(dev, pdev, id);
1965 	if (err) {
1966 		mlx5_core_err(dev, "mlx5_pci_init failed with error code %d\n",
1967 			      err);
1968 		goto pci_init_err;
1969 	}
1970 
1971 	err = mlx5_init_one(dev);
1972 	if (err) {
1973 		mlx5_core_err(dev, "mlx5_init_one failed with error code %d\n",
1974 			      err);
1975 		goto err_init_one;
1976 	}
1977 
1978 	pci_save_state(pdev);
1979 	return 0;
1980 
1981 err_init_one:
1982 	mlx5_pci_close(dev);
1983 pci_init_err:
1984 	mlx5_mdev_uninit(dev);
1985 mdev_init_err:
1986 	mlx5_adev_idx_free(dev->priv.adev_idx);
1987 adev_init_err:
1988 	mlx5_devlink_free(devlink);
1989 
1990 	return err;
1991 }
1992 
remove_one(struct pci_dev * pdev)1993 static void remove_one(struct pci_dev *pdev)
1994 {
1995 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1996 	struct devlink *devlink = priv_to_devlink(dev);
1997 
1998 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
1999 	mlx5_drain_fw_reset(dev);
2000 	mlx5_drain_health_wq(dev);
2001 	mlx5_sriov_disable(pdev, false);
2002 	mlx5_uninit_one(dev);
2003 	mlx5_pci_close(dev);
2004 	mlx5_mdev_uninit(dev);
2005 	mlx5_adev_idx_free(dev->priv.adev_idx);
2006 	mlx5_devlink_free(devlink);
2007 }
2008 
2009 #define mlx5_pci_trace(dev, fmt, ...) ({ \
2010 	struct mlx5_core_dev *__dev = (dev); \
2011 	mlx5_core_info(__dev, "%s Device state = %d health sensors: %d pci_status: %d. " fmt, \
2012 		       __func__, __dev->state, mlx5_health_check_fatal_sensors(__dev), \
2013 		       __dev->pci_status, ##__VA_ARGS__); \
2014 })
2015 
result2str(enum pci_ers_result result)2016 static const char *result2str(enum pci_ers_result result)
2017 {
2018 	return  result == PCI_ERS_RESULT_NEED_RESET ? "need reset" :
2019 		result == PCI_ERS_RESULT_DISCONNECT ? "disconnect" :
2020 		result == PCI_ERS_RESULT_RECOVERED  ? "recovered" :
2021 		"unknown";
2022 }
2023 
mlx5_pci_err_detected(struct pci_dev * pdev,pci_channel_state_t state)2024 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
2025 					      pci_channel_state_t state)
2026 {
2027 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2028 	enum pci_ers_result res;
2029 
2030 	mlx5_pci_trace(dev, "Enter, pci channel state = %d\n", state);
2031 
2032 	mlx5_enter_error_state(dev, false);
2033 	mlx5_error_sw_reset(dev);
2034 	mlx5_unload_one(dev, false);
2035 	mlx5_drain_health_wq(dev);
2036 	mlx5_pci_disable_device(dev);
2037 
2038 	res = state == pci_channel_io_perm_failure ?
2039 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2040 
2041 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, result = %d, %s\n",
2042 		       __func__, dev->state, dev->pci_status, res, result2str(res));
2043 	return res;
2044 }
2045 
2046 /* wait for the device to show vital signs by waiting
2047  * for the health counter to start counting.
2048  */
wait_vital(struct pci_dev * pdev)2049 static int wait_vital(struct pci_dev *pdev)
2050 {
2051 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2052 	struct mlx5_core_health *health = &dev->priv.health;
2053 	const int niter = 100;
2054 	u32 last_count = 0;
2055 	u32 count;
2056 	int i;
2057 
2058 	for (i = 0; i < niter; i++) {
2059 		count = ioread32be(health->health_counter);
2060 		if (count && count != 0xffffffff) {
2061 			if (last_count && last_count != count) {
2062 				mlx5_core_info(dev,
2063 					       "wait vital counter value 0x%x after %d iterations\n",
2064 					       count, i);
2065 				return 0;
2066 			}
2067 			last_count = count;
2068 		}
2069 		msleep(50);
2070 	}
2071 
2072 	return -ETIMEDOUT;
2073 }
2074 
mlx5_pci_slot_reset(struct pci_dev * pdev)2075 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
2076 {
2077 	enum pci_ers_result res = PCI_ERS_RESULT_DISCONNECT;
2078 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2079 	int err;
2080 
2081 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Enter\n",
2082 		       __func__, dev->state, dev->pci_status);
2083 
2084 	err = mlx5_pci_enable_device(dev);
2085 	if (err) {
2086 		mlx5_core_err(dev, "%s: mlx5_pci_enable_device failed with error code: %d\n",
2087 			      __func__, err);
2088 		goto out;
2089 	}
2090 
2091 	pci_set_master(pdev);
2092 	pci_restore_state(pdev);
2093 	pci_save_state(pdev);
2094 
2095 	err = wait_vital(pdev);
2096 	if (err) {
2097 		mlx5_core_err(dev, "%s: wait vital failed with error code: %d\n",
2098 			      __func__, err);
2099 		goto out;
2100 	}
2101 
2102 	res = PCI_ERS_RESULT_RECOVERED;
2103 out:
2104 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, err = %d, result = %d, %s\n",
2105 		       __func__, dev->state, dev->pci_status, err, res, result2str(res));
2106 	return res;
2107 }
2108 
mlx5_pci_resume(struct pci_dev * pdev)2109 static void mlx5_pci_resume(struct pci_dev *pdev)
2110 {
2111 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2112 	int err;
2113 
2114 	mlx5_pci_trace(dev, "Enter, loading driver..\n");
2115 
2116 	err = mlx5_load_one(dev, false);
2117 
2118 	if (!err)
2119 		devlink_health_reporter_state_update(dev->priv.health.fw_fatal_reporter,
2120 						     DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
2121 
2122 	mlx5_pci_trace(dev, "Done, err = %d, device %s\n", err,
2123 		       !err ? "recovered" : "Failed");
2124 }
2125 
2126 static const struct pci_error_handlers mlx5_err_handler = {
2127 	.error_detected = mlx5_pci_err_detected,
2128 	.slot_reset	= mlx5_pci_slot_reset,
2129 	.resume		= mlx5_pci_resume
2130 };
2131 
mlx5_try_fast_unload(struct mlx5_core_dev * dev)2132 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
2133 {
2134 	bool fast_teardown = false, force_teardown = false;
2135 	int ret = 1;
2136 
2137 	fast_teardown = MLX5_CAP_GEN(dev, fast_teardown);
2138 	force_teardown = MLX5_CAP_GEN(dev, force_teardown);
2139 
2140 	mlx5_core_dbg(dev, "force teardown firmware support=%d\n", force_teardown);
2141 	mlx5_core_dbg(dev, "fast teardown firmware support=%d\n", fast_teardown);
2142 
2143 	if (!fast_teardown && !force_teardown)
2144 		return -EOPNOTSUPP;
2145 
2146 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
2147 		mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
2148 		return -EAGAIN;
2149 	}
2150 
2151 	/* Panic tear down fw command will stop the PCI bus communication
2152 	 * with the HCA, so the health poll is no longer needed.
2153 	 */
2154 	mlx5_stop_health_poll(dev, false);
2155 
2156 	ret = mlx5_cmd_fast_teardown_hca(dev);
2157 	if (!ret)
2158 		goto succeed;
2159 
2160 	ret = mlx5_cmd_force_teardown_hca(dev);
2161 	if (!ret)
2162 		goto succeed;
2163 
2164 	mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", ret);
2165 	mlx5_start_health_poll(dev);
2166 	return ret;
2167 
2168 succeed:
2169 	mlx5_enter_error_state(dev, true);
2170 
2171 	/* Some platforms requiring freeing the IRQ's in the shutdown
2172 	 * flow. If they aren't freed they can't be allocated after
2173 	 * kexec. There is no need to cleanup the mlx5_core software
2174 	 * contexts.
2175 	 */
2176 	mlx5_core_eq_free_irqs(dev);
2177 
2178 	return 0;
2179 }
2180 
shutdown(struct pci_dev * pdev)2181 static void shutdown(struct pci_dev *pdev)
2182 {
2183 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2184 	int err;
2185 
2186 	mlx5_core_info(dev, "Shutdown was called\n");
2187 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2188 	mlx5_drain_health_wq(dev);
2189 	err = mlx5_try_fast_unload(dev);
2190 	if (err)
2191 		mlx5_unload_one(dev, false);
2192 	mlx5_pci_disable_device(dev);
2193 }
2194 
mlx5_suspend(struct pci_dev * pdev,pm_message_t state)2195 static int mlx5_suspend(struct pci_dev *pdev, pm_message_t state)
2196 {
2197 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2198 
2199 	mlx5_unload_one(dev, true);
2200 
2201 	return 0;
2202 }
2203 
mlx5_resume(struct pci_dev * pdev)2204 static int mlx5_resume(struct pci_dev *pdev)
2205 {
2206 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2207 
2208 	return mlx5_load_one(dev, false);
2209 }
2210 
2211 static const struct pci_device_id mlx5_core_pci_table[] = {
2212 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTIB) },
2213 	{ PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF},	/* Connect-IB VF */
2214 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4) },
2215 	{ PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4 VF */
2216 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4_LX) },
2217 	{ PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4LX VF */
2218 	{ PCI_VDEVICE(MELLANOX, 0x1017) },			/* ConnectX-5, PCIe 3.0 */
2219 	{ PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 VF */
2220 	{ PCI_VDEVICE(MELLANOX, 0x1019) },			/* ConnectX-5 Ex */
2221 	{ PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 Ex VF */
2222 	{ PCI_VDEVICE(MELLANOX, 0x101b) },			/* ConnectX-6 */
2223 	{ PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF},	/* ConnectX-6 VF */
2224 	{ PCI_VDEVICE(MELLANOX, 0x101d) },			/* ConnectX-6 Dx */
2225 	{ PCI_VDEVICE(MELLANOX, 0x101e), MLX5_PCI_DEV_IS_VF},	/* ConnectX Family mlx5Gen Virtual Function */
2226 	{ PCI_VDEVICE(MELLANOX, 0x101f) },			/* ConnectX-6 LX */
2227 	{ PCI_VDEVICE(MELLANOX, 0x1021) },			/* ConnectX-7 */
2228 	{ PCI_VDEVICE(MELLANOX, 0x1023) },			/* ConnectX-8 */
2229 	{ PCI_VDEVICE(MELLANOX, 0x1025) },			/* ConnectX-9 */
2230 	{ PCI_VDEVICE(MELLANOX, 0xa2d2) },			/* BlueField integrated ConnectX-5 network controller */
2231 	{ PCI_VDEVICE(MELLANOX, 0xa2d3), MLX5_PCI_DEV_IS_VF},	/* BlueField integrated ConnectX-5 network controller VF */
2232 	{ PCI_VDEVICE(MELLANOX, 0xa2d6) },			/* BlueField-2 integrated ConnectX-6 Dx network controller */
2233 	{ PCI_VDEVICE(MELLANOX, 0xa2dc) },			/* BlueField-3 integrated ConnectX-7 network controller */
2234 	{ PCI_VDEVICE(MELLANOX, 0xa2df) },			/* BlueField-4 integrated ConnectX-8 network controller */
2235 	{ 0, }
2236 };
2237 
2238 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
2239 
mlx5_disable_device(struct mlx5_core_dev * dev)2240 void mlx5_disable_device(struct mlx5_core_dev *dev)
2241 {
2242 	mlx5_error_sw_reset(dev);
2243 	mlx5_unload_one_devl_locked(dev, false);
2244 }
2245 
mlx5_recover_device(struct mlx5_core_dev * dev)2246 int mlx5_recover_device(struct mlx5_core_dev *dev)
2247 {
2248 	if (!mlx5_core_is_sf(dev)) {
2249 		mlx5_pci_disable_device(dev);
2250 		if (mlx5_pci_slot_reset(dev->pdev) != PCI_ERS_RESULT_RECOVERED)
2251 			return -EIO;
2252 	}
2253 
2254 	return mlx5_load_one_devl_locked(dev, true);
2255 }
2256 
2257 static struct pci_driver mlx5_core_driver = {
2258 	.name           = KBUILD_MODNAME,
2259 	.id_table       = mlx5_core_pci_table,
2260 	.probe          = probe_one,
2261 	.remove         = remove_one,
2262 	.suspend        = mlx5_suspend,
2263 	.resume         = mlx5_resume,
2264 	.shutdown	= shutdown,
2265 	.err_handler	= &mlx5_err_handler,
2266 	.sriov_configure   = mlx5_core_sriov_configure,
2267 	.sriov_get_vf_total_msix = mlx5_sriov_get_vf_total_msix,
2268 	.sriov_set_msix_vec_count = mlx5_core_sriov_set_msix_vec_count,
2269 };
2270 
2271 /**
2272  * mlx5_vf_get_core_dev - Get the mlx5 core device from a given VF PCI device if
2273  *                     mlx5_core is its driver.
2274  * @pdev: The associated PCI device.
2275  *
2276  * Upon return the interface state lock stay held to let caller uses it safely.
2277  * Caller must ensure to use the returned mlx5 device for a narrow window
2278  * and put it back with mlx5_vf_put_core_dev() immediately once usage was over.
2279  *
2280  * Return: Pointer to the associated mlx5_core_dev or NULL.
2281  */
mlx5_vf_get_core_dev(struct pci_dev * pdev)2282 struct mlx5_core_dev *mlx5_vf_get_core_dev(struct pci_dev *pdev)
2283 {
2284 	struct mlx5_core_dev *mdev;
2285 
2286 	mdev = pci_iov_get_pf_drvdata(pdev, &mlx5_core_driver);
2287 	if (IS_ERR(mdev))
2288 		return NULL;
2289 
2290 	mutex_lock(&mdev->intf_state_mutex);
2291 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &mdev->intf_state)) {
2292 		mutex_unlock(&mdev->intf_state_mutex);
2293 		return NULL;
2294 	}
2295 
2296 	return mdev;
2297 }
2298 EXPORT_SYMBOL(mlx5_vf_get_core_dev);
2299 
2300 /**
2301  * mlx5_vf_put_core_dev - Put the mlx5 core device back.
2302  * @mdev: The mlx5 core device.
2303  *
2304  * Upon return the interface state lock is unlocked and caller should not
2305  * access the mdev any more.
2306  */
mlx5_vf_put_core_dev(struct mlx5_core_dev * mdev)2307 void mlx5_vf_put_core_dev(struct mlx5_core_dev *mdev)
2308 {
2309 	mutex_unlock(&mdev->intf_state_mutex);
2310 }
2311 EXPORT_SYMBOL(mlx5_vf_put_core_dev);
2312 
mlx5_core_verify_params(void)2313 static void mlx5_core_verify_params(void)
2314 {
2315 	if (prof_sel >= ARRAY_SIZE(profile)) {
2316 		pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
2317 			prof_sel,
2318 			ARRAY_SIZE(profile) - 1,
2319 			MLX5_DEFAULT_PROF);
2320 		prof_sel = MLX5_DEFAULT_PROF;
2321 	}
2322 }
2323 
mlx5_init(void)2324 static int __init mlx5_init(void)
2325 {
2326 	int err;
2327 
2328 	WARN_ONCE(strcmp(MLX5_ADEV_NAME, KBUILD_MODNAME),
2329 		  "mlx5_core name not in sync with kernel module name");
2330 
2331 	get_random_bytes(&sw_owner_id, sizeof(sw_owner_id));
2332 
2333 	mlx5_core_verify_params();
2334 	mlx5_register_debugfs();
2335 
2336 	err = mlx5e_init();
2337 	if (err)
2338 		goto err_debug;
2339 
2340 	err = mlx5_sf_driver_register();
2341 	if (err)
2342 		goto err_sf;
2343 
2344 	err = pci_register_driver(&mlx5_core_driver);
2345 	if (err)
2346 		goto err_pci;
2347 
2348 	return 0;
2349 
2350 err_pci:
2351 	mlx5_sf_driver_unregister();
2352 err_sf:
2353 	mlx5e_cleanup();
2354 err_debug:
2355 	mlx5_unregister_debugfs();
2356 	return err;
2357 }
2358 
mlx5_cleanup(void)2359 static void __exit mlx5_cleanup(void)
2360 {
2361 	pci_unregister_driver(&mlx5_core_driver);
2362 	mlx5_sf_driver_unregister();
2363 	mlx5e_cleanup();
2364 	mlx5_unregister_debugfs();
2365 }
2366 
2367 module_init(mlx5_init);
2368 module_exit(mlx5_cleanup);
2369