xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/main.c (revision ab431bc39741e9d9bd3102688439e1864c857a74)
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 	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 
mlx5_core_get_caps(struct mlx5_core_dev * dev,enum mlx5_cap_type cap_type)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 
set_caps(struct mlx5_core_dev * dev,void * in,int opmod)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 
handle_hca_cap_atomic(struct mlx5_core_dev * dev,void * set_ctx)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 
handle_hca_cap_odp(struct mlx5_core_dev * dev,void * set_ctx)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 
max_uc_list_get_devlink_param(struct mlx5_core_dev * dev)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 
mlx5_is_roce_on(struct mlx5_core_dev * dev)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 
handle_hca_cap_2(struct mlx5_core_dev * dev,void * set_ctx)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 
handle_hca_cap(struct mlx5_core_dev * dev,void * set_ctx)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  */
is_roce_fw_disabled(struct mlx5_core_dev * dev)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 
handle_hca_cap_roce(struct mlx5_core_dev * dev,void * set_ctx)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 
handle_hca_cap_port_selection(struct mlx5_core_dev * dev,void * set_ctx)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 
set_hca_cap(struct mlx5_core_dev * dev)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 
set_hca_ctrl(struct mlx5_core_dev * dev)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 
mlx5_core_set_hca_defaults(struct mlx5_core_dev * dev)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 
mlx5_core_enable_hca(struct mlx5_core_dev * dev,u16 func_id)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 
mlx5_core_disable_hca(struct mlx5_core_dev * dev,u16 func_id)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 
mlx5_core_set_issi(struct mlx5_core_dev * dev)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 
mlx5_pci_init(struct mlx5_core_dev * dev,struct pci_dev * pdev,const struct pci_device_id * id)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 
mlx5_pci_close(struct mlx5_core_dev * dev)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 
mlx5_init_once(struct mlx5_core_dev * dev)976 static int mlx5_init_once(struct mlx5_core_dev *dev)
977 {
978 	int err;
979 
980 	dev->priv.devc = mlx5_devcom_register_device(dev);
981 	if (!dev->priv.devc)
982 		mlx5_core_warn(dev, "failed to register devcom device\n");
983 
984 	err = mlx5_query_board_id(dev);
985 	if (err) {
986 		mlx5_core_err(dev, "query board id failed\n");
987 		goto err_devcom;
988 	}
989 
990 	err = mlx5_irq_table_init(dev);
991 	if (err) {
992 		mlx5_core_err(dev, "failed to initialize irq table\n");
993 		goto err_devcom;
994 	}
995 
996 	err = mlx5_eq_table_init(dev);
997 	if (err) {
998 		mlx5_core_err(dev, "failed to initialize eq\n");
999 		goto err_irq_cleanup;
1000 	}
1001 
1002 	err = mlx5_events_init(dev);
1003 	if (err) {
1004 		mlx5_core_err(dev, "failed to initialize events\n");
1005 		goto err_eq_cleanup;
1006 	}
1007 
1008 	err = mlx5_fw_reset_init(dev);
1009 	if (err) {
1010 		mlx5_core_err(dev, "failed to initialize fw reset events\n");
1011 		goto err_events_cleanup;
1012 	}
1013 
1014 	mlx5_cq_debugfs_init(dev);
1015 
1016 	mlx5_init_reserved_gids(dev);
1017 
1018 	err = mlx5_init_clock(dev);
1019 	if (err) {
1020 		mlx5_core_err(dev, "failed to initialize hardware clock\n");
1021 		goto err_tables_cleanup;
1022 	}
1023 
1024 	dev->vxlan = mlx5_vxlan_create(dev);
1025 	dev->geneve = mlx5_geneve_create(dev);
1026 
1027 	err = mlx5_init_rl_table(dev);
1028 	if (err) {
1029 		mlx5_core_err(dev, "Failed to init rate limiting\n");
1030 		goto err_clock_cleanup;
1031 	}
1032 
1033 	err = mlx5_mpfs_init(dev);
1034 	if (err) {
1035 		mlx5_core_err(dev, "Failed to init l2 table %d\n", err);
1036 		goto err_rl_cleanup;
1037 	}
1038 
1039 	err = mlx5_sriov_init(dev);
1040 	if (err) {
1041 		mlx5_core_err(dev, "Failed to init sriov %d\n", err);
1042 		goto err_mpfs_cleanup;
1043 	}
1044 
1045 	err = mlx5_eswitch_init(dev);
1046 	if (err) {
1047 		mlx5_core_err(dev, "Failed to init eswitch %d\n", err);
1048 		goto err_sriov_cleanup;
1049 	}
1050 
1051 	err = mlx5_fpga_init(dev);
1052 	if (err) {
1053 		mlx5_core_err(dev, "Failed to init fpga device %d\n", err);
1054 		goto err_eswitch_cleanup;
1055 	}
1056 
1057 	err = mlx5_vhca_event_init(dev);
1058 	if (err) {
1059 		mlx5_core_err(dev, "Failed to init vhca event notifier %d\n", err);
1060 		goto err_fpga_cleanup;
1061 	}
1062 
1063 	err = mlx5_sf_hw_table_init(dev);
1064 	if (err) {
1065 		mlx5_core_err(dev, "Failed to init SF HW table %d\n", err);
1066 		goto err_sf_hw_table_cleanup;
1067 	}
1068 
1069 	err = mlx5_sf_table_init(dev);
1070 	if (err) {
1071 		mlx5_core_err(dev, "Failed to init SF table %d\n", err);
1072 		goto err_sf_table_cleanup;
1073 	}
1074 
1075 	err = mlx5_fs_core_alloc(dev);
1076 	if (err) {
1077 		mlx5_core_err(dev, "Failed to alloc flow steering\n");
1078 		goto err_fs;
1079 	}
1080 
1081 	dev->dm = mlx5_dm_create(dev);
1082 	dev->st = mlx5_st_create(dev);
1083 	dev->tracer = mlx5_fw_tracer_create(dev);
1084 	dev->hv_vhca = mlx5_hv_vhca_create(dev);
1085 	dev->rsc_dump = mlx5_rsc_dump_create(dev);
1086 
1087 	return 0;
1088 
1089 err_fs:
1090 	mlx5_sf_table_cleanup(dev);
1091 err_sf_table_cleanup:
1092 	mlx5_sf_hw_table_cleanup(dev);
1093 err_sf_hw_table_cleanup:
1094 	mlx5_vhca_event_cleanup(dev);
1095 err_fpga_cleanup:
1096 	mlx5_fpga_cleanup(dev);
1097 err_eswitch_cleanup:
1098 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1099 err_sriov_cleanup:
1100 	mlx5_sriov_cleanup(dev);
1101 err_mpfs_cleanup:
1102 	mlx5_mpfs_cleanup(dev);
1103 err_rl_cleanup:
1104 	mlx5_cleanup_rl_table(dev);
1105 err_clock_cleanup:
1106 	mlx5_geneve_destroy(dev->geneve);
1107 	mlx5_vxlan_destroy(dev->vxlan);
1108 	mlx5_cleanup_clock(dev);
1109 err_tables_cleanup:
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_devcom_unregister_device(dev->priv.devc);
1121 
1122 	return err;
1123 }
1124 
mlx5_cleanup_once(struct mlx5_core_dev * dev)1125 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
1126 {
1127 	mlx5_rsc_dump_destroy(dev);
1128 	mlx5_hv_vhca_destroy(dev->hv_vhca);
1129 	mlx5_fw_tracer_destroy(dev->tracer);
1130 	mlx5_st_destroy(dev);
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_devcom_unregister_device(dev->priv.devc);
1151 }
1152 
mlx5_function_enable(struct mlx5_core_dev * dev,bool boot,u64 timeout)1153 static int mlx5_function_enable(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1154 {
1155 	int err;
1156 
1157 	mlx5_core_info(dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1158 		       fw_rev_min(dev), fw_rev_sub(dev));
1159 
1160 	/* Only PFs hold the relevant PCIe information for this query */
1161 	if (mlx5_core_is_pf(dev))
1162 		pcie_print_link_status(dev->pdev);
1163 
1164 	/* wait for firmware to accept initialization segments configurations
1165 	 */
1166 	err = wait_fw_init(dev, timeout,
1167 			   mlx5_tout_ms(dev, FW_PRE_INIT_WARN_MESSAGE_INTERVAL),
1168 			   "pre-initializing");
1169 	if (err)
1170 		return err;
1171 
1172 	err = mlx5_cmd_enable(dev);
1173 	if (err) {
1174 		mlx5_core_err(dev, "Failed initializing command interface, aborting\n");
1175 		return err;
1176 	}
1177 
1178 	mlx5_tout_query_iseg(dev);
1179 
1180 	err = wait_fw_init(dev, mlx5_tout_ms(dev, FW_INIT), 0, "initializing");
1181 	if (err)
1182 		goto err_cmd_cleanup;
1183 
1184 	dev->caps.embedded_cpu = mlx5_read_embedded_cpu(dev);
1185 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_UP);
1186 
1187 	err = mlx5_core_enable_hca(dev, 0);
1188 	if (err) {
1189 		mlx5_core_err(dev, "enable hca failed\n");
1190 		goto err_cmd_cleanup;
1191 	}
1192 
1193 	mlx5_start_health_poll(dev);
1194 
1195 	err = mlx5_core_set_issi(dev);
1196 	if (err) {
1197 		mlx5_core_err(dev, "failed to set issi\n");
1198 		goto stop_health_poll;
1199 	}
1200 
1201 	err = mlx5_satisfy_startup_pages(dev, 1);
1202 	if (err) {
1203 		mlx5_core_err(dev, "failed to allocate boot pages\n");
1204 		goto stop_health_poll;
1205 	}
1206 
1207 	err = mlx5_tout_query_dtor(dev);
1208 	if (err) {
1209 		mlx5_core_err(dev, "failed to read dtor\n");
1210 		goto reclaim_boot_pages;
1211 	}
1212 
1213 	return 0;
1214 
1215 reclaim_boot_pages:
1216 	mlx5_reclaim_startup_pages(dev);
1217 stop_health_poll:
1218 	mlx5_stop_health_poll(dev, boot);
1219 	mlx5_core_disable_hca(dev, 0);
1220 err_cmd_cleanup:
1221 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1222 	mlx5_cmd_disable(dev);
1223 
1224 	return err;
1225 }
1226 
mlx5_function_disable(struct mlx5_core_dev * dev,bool boot)1227 static void mlx5_function_disable(struct mlx5_core_dev *dev, bool boot)
1228 {
1229 	mlx5_reclaim_startup_pages(dev);
1230 	mlx5_stop_health_poll(dev, boot);
1231 	mlx5_core_disable_hca(dev, 0);
1232 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1233 	mlx5_cmd_disable(dev);
1234 }
1235 
mlx5_function_open(struct mlx5_core_dev * dev)1236 static int mlx5_function_open(struct mlx5_core_dev *dev)
1237 {
1238 	int err;
1239 
1240 	err = set_hca_ctrl(dev);
1241 	if (err) {
1242 		mlx5_core_err(dev, "set_hca_ctrl failed\n");
1243 		return err;
1244 	}
1245 
1246 	err = set_hca_cap(dev);
1247 	if (err) {
1248 		mlx5_core_err(dev, "set_hca_cap failed\n");
1249 		return err;
1250 	}
1251 
1252 	err = mlx5_satisfy_startup_pages(dev, 0);
1253 	if (err) {
1254 		mlx5_core_err(dev, "failed to allocate init pages\n");
1255 		return err;
1256 	}
1257 
1258 	err = mlx5_cmd_init_hca(dev, sw_owner_id);
1259 	if (err) {
1260 		mlx5_core_err(dev, "init hca failed\n");
1261 		return err;
1262 	}
1263 
1264 	mlx5_set_driver_version(dev);
1265 
1266 	err = mlx5_query_hca_caps(dev);
1267 	if (err) {
1268 		mlx5_core_err(dev, "query hca failed\n");
1269 		return err;
1270 	}
1271 	mlx5_start_health_fw_log_up(dev);
1272 	return 0;
1273 }
1274 
mlx5_function_close(struct mlx5_core_dev * dev)1275 static int mlx5_function_close(struct mlx5_core_dev *dev)
1276 {
1277 	int err;
1278 
1279 	err = mlx5_cmd_teardown_hca(dev);
1280 	if (err) {
1281 		mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n");
1282 		return err;
1283 	}
1284 
1285 	return 0;
1286 }
1287 
mlx5_function_setup(struct mlx5_core_dev * dev,bool boot,u64 timeout)1288 static int mlx5_function_setup(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1289 {
1290 	int err;
1291 
1292 	err = mlx5_function_enable(dev, boot, timeout);
1293 	if (err)
1294 		return err;
1295 
1296 	err = mlx5_function_open(dev);
1297 	if (err)
1298 		mlx5_function_disable(dev, boot);
1299 	return err;
1300 }
1301 
mlx5_function_teardown(struct mlx5_core_dev * dev,bool boot)1302 static int mlx5_function_teardown(struct mlx5_core_dev *dev, bool boot)
1303 {
1304 	int err = mlx5_function_close(dev);
1305 
1306 	if (!err)
1307 		mlx5_function_disable(dev, boot);
1308 	else
1309 		mlx5_stop_health_poll(dev, boot);
1310 
1311 	return err;
1312 }
1313 
mlx5_load(struct mlx5_core_dev * dev)1314 static int mlx5_load(struct mlx5_core_dev *dev)
1315 {
1316 	int err;
1317 
1318 	err = mlx5_alloc_bfreg(dev, &dev->priv.bfreg, false, false);
1319 	if (err) {
1320 		mlx5_core_err(dev, "Failed allocating bfreg, %d\n", err);
1321 		return err;
1322 	}
1323 
1324 	mlx5_events_start(dev);
1325 	mlx5_pagealloc_start(dev);
1326 
1327 	err = mlx5_irq_table_create(dev);
1328 	if (err) {
1329 		mlx5_core_err(dev, "Failed to alloc IRQs\n");
1330 		goto err_irq_table;
1331 	}
1332 
1333 	err = mlx5_eq_table_create(dev);
1334 	if (err) {
1335 		mlx5_core_err(dev, "Failed to create EQs\n");
1336 		goto err_eq_table;
1337 	}
1338 
1339 	mlx5_clock_load(dev);
1340 
1341 	err = mlx5_fw_tracer_init(dev->tracer);
1342 	if (err) {
1343 		mlx5_core_err(dev, "Failed to init FW tracer %d\n", err);
1344 		mlx5_fw_tracer_destroy(dev->tracer);
1345 		dev->tracer = NULL;
1346 	}
1347 
1348 	mlx5_fw_reset_events_start(dev);
1349 	mlx5_hv_vhca_init(dev->hv_vhca);
1350 
1351 	err = mlx5_rsc_dump_init(dev);
1352 	if (err) {
1353 		mlx5_core_err(dev, "Failed to init Resource dump %d\n", err);
1354 		mlx5_rsc_dump_destroy(dev);
1355 		dev->rsc_dump = NULL;
1356 	}
1357 
1358 	err = mlx5_fpga_device_start(dev);
1359 	if (err) {
1360 		mlx5_core_err(dev, "fpga device start failed %d\n", err);
1361 		goto err_fpga_start;
1362 	}
1363 
1364 	err = mlx5_fs_core_init(dev);
1365 	if (err) {
1366 		mlx5_core_err(dev, "Failed to init flow steering\n");
1367 		goto err_fs;
1368 	}
1369 
1370 	err = mlx5_core_set_hca_defaults(dev);
1371 	if (err) {
1372 		mlx5_core_err(dev, "Failed to set hca defaults\n");
1373 		goto err_set_hca;
1374 	}
1375 
1376 	mlx5_vhca_event_start(dev);
1377 
1378 	err = mlx5_sf_hw_table_create(dev);
1379 	if (err) {
1380 		mlx5_core_err(dev, "sf table create failed %d\n", err);
1381 		goto err_vhca;
1382 	}
1383 
1384 	err = mlx5_ec_init(dev);
1385 	if (err) {
1386 		mlx5_core_err(dev, "Failed to init embedded CPU\n");
1387 		goto err_ec;
1388 	}
1389 
1390 	mlx5_lag_add_mdev(dev);
1391 	err = mlx5_sriov_attach(dev);
1392 	if (err) {
1393 		mlx5_core_err(dev, "sriov init failed %d\n", err);
1394 		goto err_sriov;
1395 	}
1396 
1397 	mlx5_sf_dev_table_create(dev);
1398 
1399 	err = mlx5_devlink_traps_register(priv_to_devlink(dev));
1400 	if (err)
1401 		goto err_traps_reg;
1402 
1403 	return 0;
1404 
1405 err_traps_reg:
1406 	mlx5_sf_dev_table_destroy(dev);
1407 	mlx5_sriov_detach(dev);
1408 err_sriov:
1409 	mlx5_lag_remove_mdev(dev);
1410 	mlx5_ec_cleanup(dev);
1411 err_ec:
1412 	mlx5_sf_hw_table_destroy(dev);
1413 err_vhca:
1414 	mlx5_vhca_event_stop(dev);
1415 err_set_hca:
1416 	mlx5_fs_core_cleanup(dev);
1417 err_fs:
1418 	mlx5_fpga_device_stop(dev);
1419 err_fpga_start:
1420 	mlx5_rsc_dump_cleanup(dev);
1421 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1422 	mlx5_fw_reset_events_stop(dev);
1423 	mlx5_fw_tracer_cleanup(dev->tracer);
1424 	mlx5_clock_unload(dev);
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_free_bfreg(dev, &dev->priv.bfreg);
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_clock_unload(dev);
1452 	mlx5_eq_table_destroy(dev);
1453 	mlx5_irq_table_destroy(dev);
1454 	mlx5_pagealloc_stop(dev);
1455 	mlx5_events_stop(dev);
1456 	mlx5_free_bfreg(dev, &dev->priv.bfreg);
1457 }
1458 
mlx5_init_one_devl_locked(struct mlx5_core_dev * dev)1459 int mlx5_init_one_devl_locked(struct mlx5_core_dev *dev)
1460 {
1461 	bool light_probe = mlx5_dev_is_lightweight(dev);
1462 	int err = 0;
1463 
1464 	mutex_lock(&dev->intf_state_mutex);
1465 	dev->state = MLX5_DEVICE_STATE_UP;
1466 
1467 	err = mlx5_function_setup(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1468 	if (err)
1469 		goto err_function;
1470 
1471 	err = mlx5_init_once(dev);
1472 	if (err) {
1473 		mlx5_core_err(dev, "sw objs init failed\n");
1474 		goto function_teardown;
1475 	}
1476 
1477 	/* In case of light_probe, mlx5_devlink is already registered.
1478 	 * Hence, don't register devlink again.
1479 	 */
1480 	if (!light_probe) {
1481 		err = mlx5_devlink_params_register(priv_to_devlink(dev));
1482 		if (err)
1483 			goto err_devlink_params_reg;
1484 	}
1485 
1486 	err = mlx5_load(dev);
1487 	if (err)
1488 		goto err_load;
1489 
1490 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1491 
1492 	err = mlx5_register_device(dev);
1493 	if (err)
1494 		goto err_register;
1495 
1496 	err = mlx5_crdump_enable(dev);
1497 	if (err)
1498 		mlx5_core_err(dev, "mlx5_crdump_enable failed with error code %d\n", err);
1499 
1500 	err = mlx5_hwmon_dev_register(dev);
1501 	if (err)
1502 		mlx5_core_err(dev, "mlx5_hwmon_dev_register failed with error code %d\n", err);
1503 
1504 	mutex_unlock(&dev->intf_state_mutex);
1505 	return 0;
1506 
1507 err_register:
1508 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1509 	mlx5_unload(dev);
1510 err_load:
1511 	if (!light_probe)
1512 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1513 err_devlink_params_reg:
1514 	mlx5_cleanup_once(dev);
1515 function_teardown:
1516 	mlx5_function_teardown(dev, true);
1517 err_function:
1518 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1519 	mutex_unlock(&dev->intf_state_mutex);
1520 	return err;
1521 }
1522 
mlx5_init_one(struct mlx5_core_dev * dev)1523 int mlx5_init_one(struct mlx5_core_dev *dev)
1524 {
1525 	struct devlink *devlink = priv_to_devlink(dev);
1526 	int err;
1527 
1528 	devl_lock(devlink);
1529 	devl_register(devlink);
1530 	err = mlx5_init_one_devl_locked(dev);
1531 	if (err)
1532 		devl_unregister(devlink);
1533 	devl_unlock(devlink);
1534 	return err;
1535 }
1536 
mlx5_uninit_one(struct mlx5_core_dev * dev)1537 void mlx5_uninit_one(struct mlx5_core_dev *dev)
1538 {
1539 	struct devlink *devlink = priv_to_devlink(dev);
1540 
1541 	devl_lock(devlink);
1542 	mutex_lock(&dev->intf_state_mutex);
1543 
1544 	mlx5_hwmon_dev_unregister(dev);
1545 	mlx5_crdump_disable(dev);
1546 	mlx5_unregister_device(dev);
1547 
1548 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1549 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1550 			       __func__);
1551 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1552 		mlx5_cleanup_once(dev);
1553 		goto out;
1554 	}
1555 
1556 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1557 	mlx5_unload(dev);
1558 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1559 	mlx5_cleanup_once(dev);
1560 	mlx5_function_teardown(dev, true);
1561 out:
1562 	mutex_unlock(&dev->intf_state_mutex);
1563 	devl_unregister(devlink);
1564 	devl_unlock(devlink);
1565 }
1566 
mlx5_load_one_devl_locked(struct mlx5_core_dev * dev,bool recovery)1567 int mlx5_load_one_devl_locked(struct mlx5_core_dev *dev, bool recovery)
1568 {
1569 	int err = 0;
1570 	u64 timeout;
1571 
1572 	devl_assert_locked(priv_to_devlink(dev));
1573 	mutex_lock(&dev->intf_state_mutex);
1574 	if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1575 		mlx5_core_warn(dev, "interface is up, NOP\n");
1576 		goto out;
1577 	}
1578 	/* remove any previous indication of internal error */
1579 	dev->state = MLX5_DEVICE_STATE_UP;
1580 
1581 	if (recovery)
1582 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_ON_RECOVERY_TIMEOUT);
1583 	else
1584 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT);
1585 	err = mlx5_function_setup(dev, false, timeout);
1586 	if (err)
1587 		goto err_function;
1588 
1589 	err = mlx5_load(dev);
1590 	if (err)
1591 		goto err_load;
1592 
1593 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1594 
1595 	err = mlx5_attach_device(dev);
1596 	if (err)
1597 		goto err_attach;
1598 
1599 	mutex_unlock(&dev->intf_state_mutex);
1600 	return 0;
1601 
1602 err_attach:
1603 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1604 	mlx5_unload(dev);
1605 err_load:
1606 	mlx5_function_teardown(dev, false);
1607 err_function:
1608 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1609 out:
1610 	mutex_unlock(&dev->intf_state_mutex);
1611 	return err;
1612 }
1613 
mlx5_load_one(struct mlx5_core_dev * dev,bool recovery)1614 int mlx5_load_one(struct mlx5_core_dev *dev, bool recovery)
1615 {
1616 	struct devlink *devlink = priv_to_devlink(dev);
1617 	int ret;
1618 
1619 	devl_lock(devlink);
1620 	ret = mlx5_load_one_devl_locked(dev, recovery);
1621 	devl_unlock(devlink);
1622 	return ret;
1623 }
1624 
mlx5_unload_one_devl_locked(struct mlx5_core_dev * dev,bool suspend)1625 void mlx5_unload_one_devl_locked(struct mlx5_core_dev *dev, bool suspend)
1626 {
1627 	devl_assert_locked(priv_to_devlink(dev));
1628 	mutex_lock(&dev->intf_state_mutex);
1629 
1630 	mlx5_detach_device(dev, suspend);
1631 
1632 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1633 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1634 			       __func__);
1635 		goto out;
1636 	}
1637 
1638 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1639 	mlx5_unload(dev);
1640 	mlx5_function_teardown(dev, false);
1641 out:
1642 	mutex_unlock(&dev->intf_state_mutex);
1643 }
1644 
mlx5_unload_one(struct mlx5_core_dev * dev,bool suspend)1645 void mlx5_unload_one(struct mlx5_core_dev *dev, bool suspend)
1646 {
1647 	struct devlink *devlink = priv_to_devlink(dev);
1648 
1649 	devl_lock(devlink);
1650 	mlx5_unload_one_devl_locked(dev, suspend);
1651 	devl_unlock(devlink);
1652 }
1653 
1654 /* In case of light probe, we don't need a full query of hca_caps, but only the bellow caps.
1655  * A full query of hca_caps will be done when the device will reload.
1656  */
mlx5_query_hca_caps_light(struct mlx5_core_dev * dev)1657 static int mlx5_query_hca_caps_light(struct mlx5_core_dev *dev)
1658 {
1659 	int err;
1660 
1661 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
1662 	if (err)
1663 		return err;
1664 
1665 	if (MLX5_CAP_GEN(dev, eth_net_offloads)) {
1666 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_ETHERNET_OFFLOADS,
1667 					      HCA_CAP_OPMOD_GET_CUR);
1668 		if (err)
1669 			return err;
1670 	}
1671 
1672 	if (MLX5_CAP_GEN(dev, nic_flow_table) ||
1673 	    MLX5_CAP_GEN(dev, ipoib_enhanced_offloads)) {
1674 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_FLOW_TABLE,
1675 					      HCA_CAP_OPMOD_GET_CUR);
1676 		if (err)
1677 			return err;
1678 	}
1679 
1680 	if (MLX5_CAP_GEN_64(dev, general_obj_types) &
1681 		MLX5_GENERAL_OBJ_TYPES_CAP_VIRTIO_NET_Q) {
1682 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_VDPA_EMULATION,
1683 					      HCA_CAP_OPMOD_GET_CUR);
1684 		if (err)
1685 			return err;
1686 	}
1687 
1688 	return 0;
1689 }
1690 
mlx5_init_one_light(struct mlx5_core_dev * dev)1691 int mlx5_init_one_light(struct mlx5_core_dev *dev)
1692 {
1693 	struct devlink *devlink = priv_to_devlink(dev);
1694 	int err;
1695 
1696 	devl_lock(devlink);
1697 	devl_register(devlink);
1698 	dev->state = MLX5_DEVICE_STATE_UP;
1699 	err = mlx5_function_enable(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1700 	if (err) {
1701 		mlx5_core_warn(dev, "mlx5_function_enable err=%d\n", err);
1702 		goto out;
1703 	}
1704 
1705 	err = mlx5_query_hca_caps_light(dev);
1706 	if (err) {
1707 		mlx5_core_warn(dev, "mlx5_query_hca_caps_light err=%d\n", err);
1708 		goto query_hca_caps_err;
1709 	}
1710 
1711 	err = mlx5_devlink_params_register(priv_to_devlink(dev));
1712 	if (err) {
1713 		mlx5_core_warn(dev, "mlx5_devlink_param_reg err = %d\n", err);
1714 		goto query_hca_caps_err;
1715 	}
1716 
1717 	devl_unlock(devlink);
1718 	return 0;
1719 
1720 query_hca_caps_err:
1721 	mlx5_function_disable(dev, true);
1722 out:
1723 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1724 	devl_unregister(devlink);
1725 	devl_unlock(devlink);
1726 	return err;
1727 }
1728 
mlx5_uninit_one_light(struct mlx5_core_dev * dev)1729 void mlx5_uninit_one_light(struct mlx5_core_dev *dev)
1730 {
1731 	struct devlink *devlink = priv_to_devlink(dev);
1732 
1733 	devl_lock(devlink);
1734 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1735 	devl_unregister(devlink);
1736 	devl_unlock(devlink);
1737 	if (dev->state != MLX5_DEVICE_STATE_UP)
1738 		return;
1739 	mlx5_function_disable(dev, true);
1740 }
1741 
1742 /* xxx_light() function are used in order to configure the device without full
1743  * init (light init). e.g.: There isn't a point in reload a device to light state.
1744  * Hence, mlx5_load_one_light() isn't needed.
1745  */
1746 
mlx5_unload_one_light(struct mlx5_core_dev * dev)1747 void mlx5_unload_one_light(struct mlx5_core_dev *dev)
1748 {
1749 	if (dev->state != MLX5_DEVICE_STATE_UP)
1750 		return;
1751 	mlx5_function_disable(dev, false);
1752 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1753 }
1754 
1755 static const int types[] = {
1756 	MLX5_CAP_GENERAL,
1757 	MLX5_CAP_GENERAL_2,
1758 	MLX5_CAP_ETHERNET_OFFLOADS,
1759 	MLX5_CAP_IPOIB_ENHANCED_OFFLOADS,
1760 	MLX5_CAP_ODP,
1761 	MLX5_CAP_ATOMIC,
1762 	MLX5_CAP_ROCE,
1763 	MLX5_CAP_IPOIB_OFFLOADS,
1764 	MLX5_CAP_FLOW_TABLE,
1765 	MLX5_CAP_ESWITCH_FLOW_TABLE,
1766 	MLX5_CAP_ESWITCH,
1767 	MLX5_CAP_QOS,
1768 	MLX5_CAP_DEBUG,
1769 	MLX5_CAP_DEV_MEM,
1770 	MLX5_CAP_DEV_EVENT,
1771 	MLX5_CAP_TLS,
1772 	MLX5_CAP_VDPA_EMULATION,
1773 	MLX5_CAP_IPSEC,
1774 	MLX5_CAP_PORT_SELECTION,
1775 	MLX5_CAP_PSP,
1776 	MLX5_CAP_MACSEC,
1777 	MLX5_CAP_ADV_VIRTUALIZATION,
1778 	MLX5_CAP_CRYPTO,
1779 	MLX5_CAP_SHAMPO,
1780 	MLX5_CAP_ADV_RDMA,
1781 };
1782 
mlx5_hca_caps_free(struct mlx5_core_dev * dev)1783 static void mlx5_hca_caps_free(struct mlx5_core_dev *dev)
1784 {
1785 	int type;
1786 	int i;
1787 
1788 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1789 		type = types[i];
1790 		kfree(dev->caps.hca[type]);
1791 	}
1792 }
1793 
mlx5_hca_caps_alloc(struct mlx5_core_dev * dev)1794 static int mlx5_hca_caps_alloc(struct mlx5_core_dev *dev)
1795 {
1796 	struct mlx5_hca_cap *cap;
1797 	int type;
1798 	int i;
1799 
1800 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1801 		cap = kzalloc(sizeof(*cap), GFP_KERNEL);
1802 		if (!cap)
1803 			goto err;
1804 		type = types[i];
1805 		dev->caps.hca[type] = cap;
1806 	}
1807 
1808 	return 0;
1809 
1810 err:
1811 	mlx5_hca_caps_free(dev);
1812 	return -ENOMEM;
1813 }
1814 
vhca_id_show(struct seq_file * file,void * priv)1815 static int vhca_id_show(struct seq_file *file, void *priv)
1816 {
1817 	struct mlx5_core_dev *dev = file->private;
1818 
1819 	seq_printf(file, "0x%x\n", MLX5_CAP_GEN(dev, vhca_id));
1820 	return 0;
1821 }
1822 
1823 DEFINE_SHOW_ATTRIBUTE(vhca_id);
1824 
mlx5_mdev_init(struct mlx5_core_dev * dev,int profile_idx)1825 int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
1826 {
1827 	struct mlx5_priv *priv = &dev->priv;
1828 	int err;
1829 
1830 	memcpy(&dev->profile, &profile[profile_idx], sizeof(dev->profile));
1831 	lockdep_register_key(&dev->lock_key);
1832 	mutex_init(&dev->intf_state_mutex);
1833 	lockdep_set_class(&dev->intf_state_mutex, &dev->lock_key);
1834 	mutex_init(&dev->mlx5e_res.uplink_netdev_lock);
1835 	mutex_init(&dev->wc_state_lock);
1836 
1837 	mutex_init(&priv->bfregs.reg_head.lock);
1838 	mutex_init(&priv->bfregs.wc_head.lock);
1839 	INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1840 	INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1841 
1842 	mutex_init(&priv->alloc_mutex);
1843 	mutex_init(&priv->pgdir_mutex);
1844 	INIT_LIST_HEAD(&priv->pgdir_list);
1845 
1846 	priv->numa_node = dev_to_node(mlx5_core_dma_dev(dev));
1847 	priv->dbg.dbg_root = debugfs_create_dir(dev_name(dev->device),
1848 						mlx5_debugfs_root);
1849 	debugfs_create_file("vhca_id", 0400, priv->dbg.dbg_root, dev, &vhca_id_fops);
1850 	INIT_LIST_HEAD(&priv->traps);
1851 
1852 	err = mlx5_cmd_init(dev);
1853 	if (err) {
1854 		mlx5_core_err(dev, "Failed initializing cmdif SW structs, aborting\n");
1855 		goto err_cmd_init;
1856 	}
1857 
1858 	err = mlx5_tout_init(dev);
1859 	if (err) {
1860 		mlx5_core_err(dev, "Failed initializing timeouts, aborting\n");
1861 		goto err_timeout_init;
1862 	}
1863 
1864 	err = mlx5_health_init(dev);
1865 	if (err)
1866 		goto err_health_init;
1867 
1868 	err = mlx5_pagealloc_init(dev);
1869 	if (err)
1870 		goto err_pagealloc_init;
1871 
1872 	err = mlx5_adev_init(dev);
1873 	if (err)
1874 		goto err_adev_init;
1875 
1876 	err = mlx5_hca_caps_alloc(dev);
1877 	if (err)
1878 		goto err_hca_caps;
1879 
1880 	/* The conjunction of sw_vhca_id with sw_owner_id will be a global
1881 	 * unique id per function which uses mlx5_core.
1882 	 * Those values are supplied to FW as part of the init HCA command to
1883 	 * be used by both driver and FW when it's applicable.
1884 	 */
1885 	dev->priv.sw_vhca_id = ida_alloc_range(&sw_vhca_ida, 1,
1886 					       MAX_SW_VHCA_ID,
1887 					       GFP_KERNEL);
1888 	if (dev->priv.sw_vhca_id < 0)
1889 		mlx5_core_err(dev, "failed to allocate sw_vhca_id, err=%d\n",
1890 			      dev->priv.sw_vhca_id);
1891 
1892 	return 0;
1893 
1894 err_hca_caps:
1895 	mlx5_adev_cleanup(dev);
1896 err_adev_init:
1897 	mlx5_pagealloc_cleanup(dev);
1898 err_pagealloc_init:
1899 	mlx5_health_cleanup(dev);
1900 err_health_init:
1901 	mlx5_tout_cleanup(dev);
1902 err_timeout_init:
1903 	mlx5_cmd_cleanup(dev);
1904 err_cmd_init:
1905 	debugfs_remove(dev->priv.dbg.dbg_root);
1906 	mutex_destroy(&priv->pgdir_mutex);
1907 	mutex_destroy(&priv->alloc_mutex);
1908 	mutex_destroy(&priv->bfregs.wc_head.lock);
1909 	mutex_destroy(&priv->bfregs.reg_head.lock);
1910 	mutex_destroy(&dev->intf_state_mutex);
1911 	lockdep_unregister_key(&dev->lock_key);
1912 	return err;
1913 }
1914 
mlx5_mdev_uninit(struct mlx5_core_dev * dev)1915 void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
1916 {
1917 	struct mlx5_priv *priv = &dev->priv;
1918 
1919 	if (priv->sw_vhca_id > 0)
1920 		ida_free(&sw_vhca_ida, dev->priv.sw_vhca_id);
1921 
1922 	mlx5_hca_caps_free(dev);
1923 	mlx5_adev_cleanup(dev);
1924 	mlx5_pagealloc_cleanup(dev);
1925 	mlx5_health_cleanup(dev);
1926 	mlx5_tout_cleanup(dev);
1927 	mlx5_cmd_cleanup(dev);
1928 	debugfs_remove_recursive(dev->priv.dbg.dbg_root);
1929 	mutex_destroy(&priv->pgdir_mutex);
1930 	mutex_destroy(&priv->alloc_mutex);
1931 	mutex_destroy(&priv->bfregs.wc_head.lock);
1932 	mutex_destroy(&priv->bfregs.reg_head.lock);
1933 	mutex_destroy(&dev->wc_state_lock);
1934 	mutex_destroy(&dev->mlx5e_res.uplink_netdev_lock);
1935 	mutex_destroy(&dev->intf_state_mutex);
1936 	lockdep_unregister_key(&dev->lock_key);
1937 }
1938 
probe_one(struct pci_dev * pdev,const struct pci_device_id * id)1939 static int probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
1940 {
1941 	struct mlx5_core_dev *dev;
1942 	struct devlink *devlink;
1943 	int err;
1944 
1945 	devlink = mlx5_devlink_alloc(&pdev->dev);
1946 	if (!devlink) {
1947 		dev_err(&pdev->dev, "devlink alloc failed\n");
1948 		return -ENOMEM;
1949 	}
1950 
1951 	dev = devlink_priv(devlink);
1952 	dev->device = &pdev->dev;
1953 	dev->pdev = pdev;
1954 
1955 	dev->coredev_type = id->driver_data & MLX5_PCI_DEV_IS_VF ?
1956 			 MLX5_COREDEV_VF : MLX5_COREDEV_PF;
1957 
1958 	dev->priv.adev_idx = mlx5_adev_idx_alloc();
1959 	if (dev->priv.adev_idx < 0) {
1960 		err = dev->priv.adev_idx;
1961 		goto adev_init_err;
1962 	}
1963 
1964 	err = mlx5_mdev_init(dev, prof_sel);
1965 	if (err)
1966 		goto mdev_init_err;
1967 
1968 	err = mlx5_pci_init(dev, pdev, id);
1969 	if (err) {
1970 		mlx5_core_err(dev, "mlx5_pci_init failed with error code %d\n",
1971 			      err);
1972 		goto pci_init_err;
1973 	}
1974 
1975 	err = mlx5_init_one(dev);
1976 	if (err) {
1977 		mlx5_core_err(dev, "mlx5_init_one failed with error code %d\n",
1978 			      err);
1979 		goto err_init_one;
1980 	}
1981 
1982 	pci_save_state(pdev);
1983 	return 0;
1984 
1985 err_init_one:
1986 	mlx5_pci_close(dev);
1987 pci_init_err:
1988 	mlx5_mdev_uninit(dev);
1989 mdev_init_err:
1990 	mlx5_adev_idx_free(dev->priv.adev_idx);
1991 adev_init_err:
1992 	mlx5_devlink_free(devlink);
1993 
1994 	return err;
1995 }
1996 
remove_one(struct pci_dev * pdev)1997 static void remove_one(struct pci_dev *pdev)
1998 {
1999 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2000 	struct devlink *devlink = priv_to_devlink(dev);
2001 
2002 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2003 	mlx5_drain_fw_reset(dev);
2004 	mlx5_drain_health_wq(dev);
2005 	mlx5_sriov_disable(pdev, false);
2006 	mlx5_uninit_one(dev);
2007 	mlx5_pci_close(dev);
2008 	mlx5_mdev_uninit(dev);
2009 	mlx5_adev_idx_free(dev->priv.adev_idx);
2010 	mlx5_devlink_free(devlink);
2011 }
2012 
2013 #define mlx5_pci_trace(dev, fmt, ...) ({ \
2014 	struct mlx5_core_dev *__dev = (dev); \
2015 	mlx5_core_info(__dev, "%s Device state = %d health sensors: %d pci_status: %d. " fmt, \
2016 		       __func__, __dev->state, mlx5_health_check_fatal_sensors(__dev), \
2017 		       __dev->pci_status, ##__VA_ARGS__); \
2018 })
2019 
result2str(enum pci_ers_result result)2020 static const char *result2str(enum pci_ers_result result)
2021 {
2022 	return  result == PCI_ERS_RESULT_NEED_RESET ? "need reset" :
2023 		result == PCI_ERS_RESULT_DISCONNECT ? "disconnect" :
2024 		result == PCI_ERS_RESULT_RECOVERED  ? "recovered" :
2025 		"unknown";
2026 }
2027 
mlx5_pci_err_detected(struct pci_dev * pdev,pci_channel_state_t state)2028 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
2029 					      pci_channel_state_t state)
2030 {
2031 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2032 	enum pci_ers_result res;
2033 
2034 	mlx5_pci_trace(dev, "Enter, pci channel state = %d\n", state);
2035 
2036 	mlx5_enter_error_state(dev, false);
2037 	mlx5_error_sw_reset(dev);
2038 	mlx5_unload_one(dev, false);
2039 	mlx5_drain_health_wq(dev);
2040 	mlx5_pci_disable_device(dev);
2041 
2042 	res = state == pci_channel_io_perm_failure ?
2043 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2044 
2045 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, result = %d, %s\n",
2046 		       __func__, dev->state, dev->pci_status, res, result2str(res));
2047 	return res;
2048 }
2049 
2050 /* wait for the device to show vital signs by waiting
2051  * for the health counter to start counting.
2052  */
wait_vital(struct pci_dev * pdev)2053 static int wait_vital(struct pci_dev *pdev)
2054 {
2055 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2056 	struct mlx5_core_health *health = &dev->priv.health;
2057 	const int niter = 100;
2058 	u32 last_count = 0;
2059 	u32 count;
2060 	int i;
2061 
2062 	for (i = 0; i < niter; i++) {
2063 		count = ioread32be(health->health_counter);
2064 		if (count && count != 0xffffffff) {
2065 			if (last_count && last_count != count) {
2066 				mlx5_core_info(dev,
2067 					       "wait vital counter value 0x%x after %d iterations\n",
2068 					       count, i);
2069 				return 0;
2070 			}
2071 			last_count = count;
2072 		}
2073 		msleep(50);
2074 	}
2075 
2076 	return -ETIMEDOUT;
2077 }
2078 
mlx5_pci_slot_reset(struct pci_dev * pdev)2079 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
2080 {
2081 	enum pci_ers_result res = PCI_ERS_RESULT_DISCONNECT;
2082 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2083 	int err;
2084 
2085 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Enter\n",
2086 		       __func__, dev->state, dev->pci_status);
2087 
2088 	err = mlx5_pci_enable_device(dev);
2089 	if (err) {
2090 		mlx5_core_err(dev, "%s: mlx5_pci_enable_device failed with error code: %d\n",
2091 			      __func__, err);
2092 		goto out;
2093 	}
2094 
2095 	pci_set_master(pdev);
2096 	pci_restore_state(pdev);
2097 	pci_save_state(pdev);
2098 
2099 	err = wait_vital(pdev);
2100 	if (err) {
2101 		mlx5_core_err(dev, "%s: wait vital failed with error code: %d\n",
2102 			      __func__, err);
2103 		goto out;
2104 	}
2105 
2106 	res = PCI_ERS_RESULT_RECOVERED;
2107 out:
2108 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, err = %d, result = %d, %s\n",
2109 		       __func__, dev->state, dev->pci_status, err, res, result2str(res));
2110 	return res;
2111 }
2112 
mlx5_pci_resume(struct pci_dev * pdev)2113 static void mlx5_pci_resume(struct pci_dev *pdev)
2114 {
2115 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2116 	int err;
2117 
2118 	mlx5_pci_trace(dev, "Enter, loading driver..\n");
2119 
2120 	err = mlx5_load_one(dev, false);
2121 
2122 	if (!err)
2123 		devlink_health_reporter_state_update(dev->priv.health.fw_fatal_reporter,
2124 						     DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
2125 
2126 	mlx5_pci_trace(dev, "Done, err = %d, device %s\n", err,
2127 		       !err ? "recovered" : "Failed");
2128 }
2129 
2130 static const struct pci_error_handlers mlx5_err_handler = {
2131 	.error_detected = mlx5_pci_err_detected,
2132 	.slot_reset	= mlx5_pci_slot_reset,
2133 	.resume		= mlx5_pci_resume
2134 };
2135 
mlx5_try_fast_unload(struct mlx5_core_dev * dev)2136 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
2137 {
2138 	bool fast_teardown = false, force_teardown = false;
2139 	int ret = 1;
2140 
2141 	fast_teardown = MLX5_CAP_GEN(dev, fast_teardown);
2142 	force_teardown = MLX5_CAP_GEN(dev, force_teardown);
2143 
2144 	mlx5_core_dbg(dev, "force teardown firmware support=%d\n", force_teardown);
2145 	mlx5_core_dbg(dev, "fast teardown firmware support=%d\n", fast_teardown);
2146 
2147 	if (!fast_teardown && !force_teardown)
2148 		return -EOPNOTSUPP;
2149 
2150 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
2151 		mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
2152 		return -EAGAIN;
2153 	}
2154 
2155 	/* Panic tear down fw command will stop the PCI bus communication
2156 	 * with the HCA, so the health poll is no longer needed.
2157 	 */
2158 	mlx5_stop_health_poll(dev, false);
2159 
2160 	ret = mlx5_cmd_fast_teardown_hca(dev);
2161 	if (!ret)
2162 		goto succeed;
2163 
2164 	ret = mlx5_cmd_force_teardown_hca(dev);
2165 	if (!ret)
2166 		goto succeed;
2167 
2168 	mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", ret);
2169 	mlx5_start_health_poll(dev);
2170 	return ret;
2171 
2172 succeed:
2173 	mlx5_enter_error_state(dev, true);
2174 
2175 	/* Some platforms requiring freeing the IRQ's in the shutdown
2176 	 * flow. If they aren't freed they can't be allocated after
2177 	 * kexec. There is no need to cleanup the mlx5_core software
2178 	 * contexts.
2179 	 */
2180 	mlx5_core_eq_free_irqs(dev);
2181 
2182 	return 0;
2183 }
2184 
shutdown(struct pci_dev * pdev)2185 static void shutdown(struct pci_dev *pdev)
2186 {
2187 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2188 	int err;
2189 
2190 	mlx5_core_info(dev, "Shutdown was called\n");
2191 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2192 	mlx5_drain_health_wq(dev);
2193 	err = mlx5_try_fast_unload(dev);
2194 	if (err)
2195 		mlx5_unload_one(dev, false);
2196 	mlx5_pci_disable_device(dev);
2197 }
2198 
mlx5_suspend(struct pci_dev * pdev,pm_message_t state)2199 static int mlx5_suspend(struct pci_dev *pdev, pm_message_t state)
2200 {
2201 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2202 
2203 	mlx5_unload_one(dev, true);
2204 
2205 	return 0;
2206 }
2207 
mlx5_resume(struct pci_dev * pdev)2208 static int mlx5_resume(struct pci_dev *pdev)
2209 {
2210 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2211 
2212 	return mlx5_load_one(dev, false);
2213 }
2214 
2215 static const struct pci_device_id mlx5_core_pci_table[] = {
2216 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTIB) },
2217 	{ PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF},	/* Connect-IB VF */
2218 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4) },
2219 	{ PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4 VF */
2220 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4_LX) },
2221 	{ PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4LX VF */
2222 	{ PCI_VDEVICE(MELLANOX, 0x1017) },			/* ConnectX-5, PCIe 3.0 */
2223 	{ PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 VF */
2224 	{ PCI_VDEVICE(MELLANOX, 0x1019) },			/* ConnectX-5 Ex */
2225 	{ PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 Ex VF */
2226 	{ PCI_VDEVICE(MELLANOX, 0x101b) },			/* ConnectX-6 */
2227 	{ PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF},	/* ConnectX-6 VF */
2228 	{ PCI_VDEVICE(MELLANOX, 0x101d) },			/* ConnectX-6 Dx */
2229 	{ PCI_VDEVICE(MELLANOX, 0x101e), MLX5_PCI_DEV_IS_VF},	/* ConnectX Family mlx5Gen Virtual Function */
2230 	{ PCI_VDEVICE(MELLANOX, 0x101f) },			/* ConnectX-6 LX */
2231 	{ PCI_VDEVICE(MELLANOX, 0x1021) },			/* ConnectX-7 */
2232 	{ PCI_VDEVICE(MELLANOX, 0x1023) },			/* ConnectX-8 */
2233 	{ PCI_VDEVICE(MELLANOX, 0x1025) },			/* ConnectX-9 */
2234 	{ PCI_VDEVICE(MELLANOX, 0x1027) },			/* ConnectX-10 */
2235 	{ PCI_VDEVICE(MELLANOX, 0xa2d2) },			/* BlueField integrated ConnectX-5 network controller */
2236 	{ PCI_VDEVICE(MELLANOX, 0xa2d3), MLX5_PCI_DEV_IS_VF},	/* BlueField integrated ConnectX-5 network controller VF */
2237 	{ PCI_VDEVICE(MELLANOX, 0xa2d6) },			/* BlueField-2 integrated ConnectX-6 Dx network controller */
2238 	{ PCI_VDEVICE(MELLANOX, 0xa2dc) },			/* BlueField-3 integrated ConnectX-7 network controller */
2239 	{ PCI_VDEVICE(MELLANOX, 0xa2df) },			/* BlueField-4 integrated ConnectX-8 network controller */
2240 	{ 0, }
2241 };
2242 
2243 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
2244 
mlx5_disable_device(struct mlx5_core_dev * dev)2245 void mlx5_disable_device(struct mlx5_core_dev *dev)
2246 {
2247 	mlx5_error_sw_reset(dev);
2248 	mlx5_unload_one_devl_locked(dev, false);
2249 }
2250 
mlx5_recover_device(struct mlx5_core_dev * dev)2251 int mlx5_recover_device(struct mlx5_core_dev *dev)
2252 {
2253 	if (!mlx5_core_is_sf(dev)) {
2254 		mlx5_pci_disable_device(dev);
2255 		if (mlx5_pci_slot_reset(dev->pdev) != PCI_ERS_RESULT_RECOVERED)
2256 			return -EIO;
2257 	}
2258 
2259 	return mlx5_load_one_devl_locked(dev, true);
2260 }
2261 
2262 static struct pci_driver mlx5_core_driver = {
2263 	.name           = KBUILD_MODNAME,
2264 	.id_table       = mlx5_core_pci_table,
2265 	.probe          = probe_one,
2266 	.remove         = remove_one,
2267 	.suspend        = mlx5_suspend,
2268 	.resume         = mlx5_resume,
2269 	.shutdown	= shutdown,
2270 	.err_handler	= &mlx5_err_handler,
2271 	.sriov_configure   = mlx5_core_sriov_configure,
2272 	.sriov_get_vf_total_msix = mlx5_sriov_get_vf_total_msix,
2273 	.sriov_set_msix_vec_count = mlx5_core_sriov_set_msix_vec_count,
2274 };
2275 
2276 /**
2277  * mlx5_vf_get_core_dev - Get the mlx5 core device from a given VF PCI device if
2278  *                     mlx5_core is its driver.
2279  * @pdev: The associated PCI device.
2280  *
2281  * Upon return the interface state lock stay held to let caller uses it safely.
2282  * Caller must ensure to use the returned mlx5 device for a narrow window
2283  * and put it back with mlx5_vf_put_core_dev() immediately once usage was over.
2284  *
2285  * Return: Pointer to the associated mlx5_core_dev or NULL.
2286  */
mlx5_vf_get_core_dev(struct pci_dev * pdev)2287 struct mlx5_core_dev *mlx5_vf_get_core_dev(struct pci_dev *pdev)
2288 {
2289 	struct mlx5_core_dev *mdev;
2290 
2291 	mdev = pci_iov_get_pf_drvdata(pdev, &mlx5_core_driver);
2292 	if (IS_ERR(mdev))
2293 		return NULL;
2294 
2295 	mutex_lock(&mdev->intf_state_mutex);
2296 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &mdev->intf_state)) {
2297 		mutex_unlock(&mdev->intf_state_mutex);
2298 		return NULL;
2299 	}
2300 
2301 	return mdev;
2302 }
2303 EXPORT_SYMBOL(mlx5_vf_get_core_dev);
2304 
2305 /**
2306  * mlx5_vf_put_core_dev - Put the mlx5 core device back.
2307  * @mdev: The mlx5 core device.
2308  *
2309  * Upon return the interface state lock is unlocked and caller should not
2310  * access the mdev any more.
2311  */
mlx5_vf_put_core_dev(struct mlx5_core_dev * mdev)2312 void mlx5_vf_put_core_dev(struct mlx5_core_dev *mdev)
2313 {
2314 	mutex_unlock(&mdev->intf_state_mutex);
2315 }
2316 EXPORT_SYMBOL(mlx5_vf_put_core_dev);
2317 
mlx5_core_verify_params(void)2318 static void mlx5_core_verify_params(void)
2319 {
2320 	if (prof_sel >= ARRAY_SIZE(profile)) {
2321 		pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
2322 			prof_sel,
2323 			ARRAY_SIZE(profile) - 1,
2324 			MLX5_DEFAULT_PROF);
2325 		prof_sel = MLX5_DEFAULT_PROF;
2326 	}
2327 }
2328 
mlx5_init(void)2329 static int __init mlx5_init(void)
2330 {
2331 	int err;
2332 
2333 	WARN_ONCE(strcmp(MLX5_ADEV_NAME, KBUILD_MODNAME),
2334 		  "mlx5_core name not in sync with kernel module name");
2335 
2336 	get_random_bytes(&sw_owner_id, sizeof(sw_owner_id));
2337 
2338 	mlx5_core_verify_params();
2339 	mlx5_register_debugfs();
2340 
2341 	err = mlx5e_init();
2342 	if (err)
2343 		goto err_debug;
2344 
2345 	err = mlx5_sf_driver_register();
2346 	if (err)
2347 		goto err_sf;
2348 
2349 	err = pci_register_driver(&mlx5_core_driver);
2350 	if (err)
2351 		goto err_pci;
2352 
2353 	return 0;
2354 
2355 err_pci:
2356 	mlx5_sf_driver_unregister();
2357 err_sf:
2358 	mlx5e_cleanup();
2359 err_debug:
2360 	mlx5_unregister_debugfs();
2361 	return err;
2362 }
2363 
mlx5_cleanup(void)2364 static void __exit mlx5_cleanup(void)
2365 {
2366 	pci_unregister_driver(&mlx5_core_driver);
2367 	mlx5_sf_driver_unregister();
2368 	mlx5e_cleanup();
2369 	mlx5_unregister_debugfs();
2370 }
2371 
2372 module_init(mlx5_init);
2373 module_exit(mlx5_cleanup);
2374