xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/main.c (revision 42b16d3ac371a2fac9b6f08fd75f23f34ba3955a)
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 	bool do_set = false, mem_page_fault = false;
458 	void *set_hca_cap;
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 	/* For best performance, enable memory scheme ODP only when
474 	 * it has page prefetch enabled.
475 	 */
476 	if (MLX5_CAP_ODP_MAX(dev, mem_page_fault) &&
477 	    MLX5_CAP_ODP_MAX(dev, memory_page_fault_scheme_cap.page_prefetch)) {
478 		mem_page_fault = true;
479 		do_set = true;
480 		MLX5_SET(odp_cap, set_hca_cap, mem_page_fault, mem_page_fault);
481 		goto set;
482 	}
483 
484 #define ODP_CAP_SET_MAX(dev, field)                                            \
485 	do {                                                                   \
486 		u32 _res = MLX5_CAP_ODP_MAX(dev, field);                       \
487 		if (_res) {                                                    \
488 			do_set = true;                                         \
489 			MLX5_SET(odp_cap, set_hca_cap, field, _res);           \
490 		}                                                              \
491 	} while (0)
492 
493 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.ud_odp_caps.srq_receive);
494 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.rc_odp_caps.srq_receive);
495 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.srq_receive);
496 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.send);
497 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.receive);
498 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.write);
499 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.read);
500 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.xrc_odp_caps.atomic);
501 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.srq_receive);
502 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.send);
503 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.receive);
504 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.write);
505 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.read);
506 	ODP_CAP_SET_MAX(dev, transport_page_fault_scheme_cap.dc_odp_caps.atomic);
507 
508 set:
509 	if (do_set)
510 		err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ODP);
511 
512 	mlx5_core_dbg(dev, "Using ODP %s scheme\n",
513 		      mem_page_fault ? "memory" : "transport");
514 	return err;
515 }
516 
max_uc_list_get_devlink_param(struct mlx5_core_dev * dev)517 static int max_uc_list_get_devlink_param(struct mlx5_core_dev *dev)
518 {
519 	struct devlink *devlink = priv_to_devlink(dev);
520 	union devlink_param_value val;
521 	int err;
522 
523 	err = devl_param_driverinit_value_get(devlink,
524 					      DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
525 					      &val);
526 	if (!err)
527 		return val.vu32;
528 	mlx5_core_dbg(dev, "Failed to get param. err = %d\n", err);
529 	return err;
530 }
531 
mlx5_is_roce_on(struct mlx5_core_dev * dev)532 bool mlx5_is_roce_on(struct mlx5_core_dev *dev)
533 {
534 	struct devlink *devlink = priv_to_devlink(dev);
535 	union devlink_param_value val;
536 	int err;
537 
538 	err = devl_param_driverinit_value_get(devlink,
539 					      DEVLINK_PARAM_GENERIC_ID_ENABLE_ROCE,
540 					      &val);
541 
542 	if (!err)
543 		return val.vbool;
544 
545 	mlx5_core_dbg(dev, "Failed to get param. err = %d\n", err);
546 	return MLX5_CAP_GEN(dev, roce);
547 }
548 EXPORT_SYMBOL(mlx5_is_roce_on);
549 
handle_hca_cap_2(struct mlx5_core_dev * dev,void * set_ctx)550 static int handle_hca_cap_2(struct mlx5_core_dev *dev, void *set_ctx)
551 {
552 	void *set_hca_cap;
553 	int err;
554 
555 	if (!MLX5_CAP_GEN_MAX(dev, hca_cap_2))
556 		return 0;
557 
558 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL_2);
559 	if (err)
560 		return err;
561 
562 	if (!MLX5_CAP_GEN_2_MAX(dev, sw_vhca_id_valid) ||
563 	    !(dev->priv.sw_vhca_id > 0))
564 		return 0;
565 
566 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
567 				   capability);
568 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_GENERAL_2]->cur,
569 	       MLX5_ST_SZ_BYTES(cmd_hca_cap_2));
570 	MLX5_SET(cmd_hca_cap_2, set_hca_cap, sw_vhca_id_valid, 1);
571 
572 	return set_caps(dev, set_ctx, MLX5_CAP_GENERAL_2);
573 }
574 
handle_hca_cap(struct mlx5_core_dev * dev,void * set_ctx)575 static int handle_hca_cap(struct mlx5_core_dev *dev, void *set_ctx)
576 {
577 	struct mlx5_profile *prof = &dev->profile;
578 	void *set_hca_cap;
579 	int max_uc_list;
580 	int err;
581 
582 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
583 	if (err)
584 		return err;
585 
586 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
587 				   capability);
588 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_GENERAL]->cur,
589 	       MLX5_ST_SZ_BYTES(cmd_hca_cap));
590 
591 	mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
592 		      mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
593 		      128);
594 	/* we limit the size of the pkey table to 128 entries for now */
595 	MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
596 		 to_fw_pkey_sz(dev, 128));
597 
598 	/* Check log_max_qp from HCA caps to set in current profile */
599 	if (prof->log_max_qp == LOG_MAX_SUPPORTED_QPS) {
600 		prof->log_max_qp = min_t(u8, 18, MLX5_CAP_GEN_MAX(dev, log_max_qp));
601 	} else if (MLX5_CAP_GEN_MAX(dev, log_max_qp) < prof->log_max_qp) {
602 		mlx5_core_warn(dev, "log_max_qp value in current profile is %d, changing it to HCA capability limit (%d)\n",
603 			       prof->log_max_qp,
604 			       MLX5_CAP_GEN_MAX(dev, log_max_qp));
605 		prof->log_max_qp = MLX5_CAP_GEN_MAX(dev, log_max_qp);
606 	}
607 	if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
608 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
609 			 prof->log_max_qp);
610 
611 	/* disable cmdif checksum */
612 	MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
613 
614 	/* Enable 4K UAR only when HCA supports it and page size is bigger
615 	 * than 4K.
616 	 */
617 	if (MLX5_CAP_GEN_MAX(dev, uar_4k) && PAGE_SIZE > 4096)
618 		MLX5_SET(cmd_hca_cap, set_hca_cap, uar_4k, 1);
619 
620 	MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
621 
622 	if (MLX5_CAP_GEN_MAX(dev, cache_line_128byte))
623 		MLX5_SET(cmd_hca_cap,
624 			 set_hca_cap,
625 			 cache_line_128byte,
626 			 cache_line_size() >= 128 ? 1 : 0);
627 
628 	if (MLX5_CAP_GEN_MAX(dev, dct))
629 		MLX5_SET(cmd_hca_cap, set_hca_cap, dct, 1);
630 
631 	if (MLX5_CAP_GEN_MAX(dev, pci_sync_for_fw_update_event))
632 		MLX5_SET(cmd_hca_cap, set_hca_cap, pci_sync_for_fw_update_event, 1);
633 	if (MLX5_CAP_GEN_MAX(dev, pci_sync_for_fw_update_with_driver_unload))
634 		MLX5_SET(cmd_hca_cap, set_hca_cap,
635 			 pci_sync_for_fw_update_with_driver_unload, 1);
636 	if (MLX5_CAP_GEN_MAX(dev, pcie_reset_using_hotreset_method))
637 		MLX5_SET(cmd_hca_cap, set_hca_cap,
638 			 pcie_reset_using_hotreset_method, 1);
639 
640 	if (MLX5_CAP_GEN_MAX(dev, num_vhca_ports))
641 		MLX5_SET(cmd_hca_cap,
642 			 set_hca_cap,
643 			 num_vhca_ports,
644 			 MLX5_CAP_GEN_MAX(dev, num_vhca_ports));
645 
646 	if (MLX5_CAP_GEN_MAX(dev, release_all_pages))
647 		MLX5_SET(cmd_hca_cap, set_hca_cap, release_all_pages, 1);
648 
649 	if (MLX5_CAP_GEN_MAX(dev, mkey_by_name))
650 		MLX5_SET(cmd_hca_cap, set_hca_cap, mkey_by_name, 1);
651 
652 	mlx5_vhca_state_cap_handle(dev, set_hca_cap);
653 
654 	if (MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix))
655 		MLX5_SET(cmd_hca_cap, set_hca_cap, num_total_dynamic_vf_msix,
656 			 MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix));
657 
658 	if (MLX5_CAP_GEN(dev, roce_rw_supported) && MLX5_CAP_GEN_MAX(dev, roce))
659 		MLX5_SET(cmd_hca_cap, set_hca_cap, roce,
660 			 mlx5_is_roce_on(dev));
661 
662 	max_uc_list = max_uc_list_get_devlink_param(dev);
663 	if (max_uc_list > 0)
664 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_current_uc_list,
665 			 ilog2(max_uc_list));
666 
667 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
668 }
669 
670 /* Cached MLX5_CAP_GEN(dev, roce) can be out of sync this early in the
671  * boot process.
672  * In case RoCE cap is writable in FW and user/devlink requested to change the
673  * cap, we are yet to query the final state of the above cap.
674  * Hence, the need for this function.
675  *
676  * Returns
677  * True:
678  * 1) RoCE cap is read only in FW and already disabled
679  * OR:
680  * 2) RoCE cap is writable in FW and user/devlink requested it off.
681  *
682  * In any other case, return False.
683  */
is_roce_fw_disabled(struct mlx5_core_dev * dev)684 static bool is_roce_fw_disabled(struct mlx5_core_dev *dev)
685 {
686 	return (MLX5_CAP_GEN(dev, roce_rw_supported) && !mlx5_is_roce_on(dev)) ||
687 		(!MLX5_CAP_GEN(dev, roce_rw_supported) && !MLX5_CAP_GEN(dev, roce));
688 }
689 
handle_hca_cap_roce(struct mlx5_core_dev * dev,void * set_ctx)690 static int handle_hca_cap_roce(struct mlx5_core_dev *dev, void *set_ctx)
691 {
692 	void *set_hca_cap;
693 	int err;
694 
695 	if (is_roce_fw_disabled(dev))
696 		return 0;
697 
698 	err = mlx5_core_get_caps(dev, MLX5_CAP_ROCE);
699 	if (err)
700 		return err;
701 
702 	if (MLX5_CAP_ROCE(dev, sw_r_roce_src_udp_port) ||
703 	    !MLX5_CAP_ROCE_MAX(dev, sw_r_roce_src_udp_port))
704 		return 0;
705 
706 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
707 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_ROCE]->cur,
708 	       MLX5_ST_SZ_BYTES(roce_cap));
709 	MLX5_SET(roce_cap, set_hca_cap, sw_r_roce_src_udp_port, 1);
710 
711 	if (MLX5_CAP_ROCE_MAX(dev, qp_ooo_transmit_default))
712 		MLX5_SET(roce_cap, set_hca_cap, qp_ooo_transmit_default, 1);
713 
714 	err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ROCE);
715 	return err;
716 }
717 
handle_hca_cap_port_selection(struct mlx5_core_dev * dev,void * set_ctx)718 static int handle_hca_cap_port_selection(struct mlx5_core_dev *dev,
719 					 void *set_ctx)
720 {
721 	void *set_hca_cap;
722 	int err;
723 
724 	if (!MLX5_CAP_GEN(dev, port_selection_cap))
725 		return 0;
726 
727 	err = mlx5_core_get_caps(dev, MLX5_CAP_PORT_SELECTION);
728 	if (err)
729 		return err;
730 
731 	if (MLX5_CAP_PORT_SELECTION(dev, port_select_flow_table_bypass) ||
732 	    !MLX5_CAP_PORT_SELECTION_MAX(dev, port_select_flow_table_bypass))
733 		return 0;
734 
735 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
736 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_PORT_SELECTION]->cur,
737 	       MLX5_ST_SZ_BYTES(port_selection_cap));
738 	MLX5_SET(port_selection_cap, set_hca_cap, port_select_flow_table_bypass, 1);
739 
740 	err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_PORT_SELECTION);
741 
742 	return err;
743 }
744 
set_hca_cap(struct mlx5_core_dev * dev)745 static int set_hca_cap(struct mlx5_core_dev *dev)
746 {
747 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
748 	void *set_ctx;
749 	int err;
750 
751 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
752 	if (!set_ctx)
753 		return -ENOMEM;
754 
755 	err = handle_hca_cap(dev, set_ctx);
756 	if (err) {
757 		mlx5_core_err(dev, "handle_hca_cap failed\n");
758 		goto out;
759 	}
760 
761 	memset(set_ctx, 0, set_sz);
762 	err = handle_hca_cap_atomic(dev, set_ctx);
763 	if (err) {
764 		mlx5_core_err(dev, "handle_hca_cap_atomic failed\n");
765 		goto out;
766 	}
767 
768 	memset(set_ctx, 0, set_sz);
769 	err = handle_hca_cap_odp(dev, set_ctx);
770 	if (err) {
771 		mlx5_core_err(dev, "handle_hca_cap_odp failed\n");
772 		goto out;
773 	}
774 
775 	memset(set_ctx, 0, set_sz);
776 	err = handle_hca_cap_roce(dev, set_ctx);
777 	if (err) {
778 		mlx5_core_err(dev, "handle_hca_cap_roce failed\n");
779 		goto out;
780 	}
781 
782 	memset(set_ctx, 0, set_sz);
783 	err = handle_hca_cap_2(dev, set_ctx);
784 	if (err) {
785 		mlx5_core_err(dev, "handle_hca_cap_2 failed\n");
786 		goto out;
787 	}
788 
789 	memset(set_ctx, 0, set_sz);
790 	err = handle_hca_cap_port_selection(dev, set_ctx);
791 	if (err) {
792 		mlx5_core_err(dev, "handle_hca_cap_port_selection failed\n");
793 		goto out;
794 	}
795 
796 out:
797 	kfree(set_ctx);
798 	return err;
799 }
800 
set_hca_ctrl(struct mlx5_core_dev * dev)801 static int set_hca_ctrl(struct mlx5_core_dev *dev)
802 {
803 	struct mlx5_reg_host_endianness he_in;
804 	struct mlx5_reg_host_endianness he_out;
805 	int err;
806 
807 	if (!mlx5_core_is_pf(dev))
808 		return 0;
809 
810 	memset(&he_in, 0, sizeof(he_in));
811 	he_in.he = MLX5_SET_HOST_ENDIANNESS;
812 	err = mlx5_core_access_reg(dev, &he_in,  sizeof(he_in),
813 					&he_out, sizeof(he_out),
814 					MLX5_REG_HOST_ENDIANNESS, 0, 1);
815 	return err;
816 }
817 
mlx5_core_set_hca_defaults(struct mlx5_core_dev * dev)818 static int mlx5_core_set_hca_defaults(struct mlx5_core_dev *dev)
819 {
820 	int ret = 0;
821 
822 	/* Disable local_lb by default */
823 	if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH)
824 		ret = mlx5_nic_vport_update_local_lb(dev, false);
825 
826 	return ret;
827 }
828 
mlx5_core_enable_hca(struct mlx5_core_dev * dev,u16 func_id)829 int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
830 {
831 	u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {};
832 
833 	MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
834 	MLX5_SET(enable_hca_in, in, function_id, func_id);
835 	MLX5_SET(enable_hca_in, in, embedded_cpu_function,
836 		 dev->caps.embedded_cpu);
837 	return mlx5_cmd_exec_in(dev, enable_hca, in);
838 }
839 
mlx5_core_disable_hca(struct mlx5_core_dev * dev,u16 func_id)840 int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
841 {
842 	u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {};
843 
844 	MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
845 	MLX5_SET(disable_hca_in, in, function_id, func_id);
846 	MLX5_SET(enable_hca_in, in, embedded_cpu_function,
847 		 dev->caps.embedded_cpu);
848 	return mlx5_cmd_exec_in(dev, disable_hca, in);
849 }
850 
mlx5_core_set_issi(struct mlx5_core_dev * dev)851 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
852 {
853 	u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {};
854 	u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {};
855 	u32 sup_issi;
856 	int err;
857 
858 	MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
859 	err = mlx5_cmd_exec_inout(dev, query_issi, query_in, query_out);
860 	if (err) {
861 		u32 syndrome = MLX5_GET(query_issi_out, query_out, syndrome);
862 		u8 status = MLX5_GET(query_issi_out, query_out, status);
863 
864 		if (!status || syndrome == MLX5_DRIVER_SYND) {
865 			mlx5_core_err(dev, "Failed to query ISSI err(%d) status(%d) synd(%d)\n",
866 				      err, status, syndrome);
867 			return err;
868 		}
869 
870 		mlx5_core_warn(dev, "Query ISSI is not supported by FW, ISSI is 0\n");
871 		dev->issi = 0;
872 		return 0;
873 	}
874 
875 	sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
876 
877 	if (sup_issi & (1 << 1)) {
878 		u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {};
879 
880 		MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
881 		MLX5_SET(set_issi_in, set_in, current_issi, 1);
882 		err = mlx5_cmd_exec_in(dev, set_issi, set_in);
883 		if (err) {
884 			mlx5_core_err(dev, "Failed to set ISSI to 1 err(%d)\n",
885 				      err);
886 			return err;
887 		}
888 
889 		dev->issi = 1;
890 
891 		return 0;
892 	} else if (sup_issi & (1 << 0) || !sup_issi) {
893 		return 0;
894 	}
895 
896 	return -EOPNOTSUPP;
897 }
898 
mlx5_pci_init(struct mlx5_core_dev * dev,struct pci_dev * pdev,const struct pci_device_id * id)899 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct pci_dev *pdev,
900 			 const struct pci_device_id *id)
901 {
902 	int err = 0;
903 
904 	mutex_init(&dev->pci_status_mutex);
905 	pci_set_drvdata(dev->pdev, dev);
906 
907 	dev->bar_addr = pci_resource_start(pdev, 0);
908 
909 	err = mlx5_pci_enable_device(dev);
910 	if (err) {
911 		mlx5_core_err(dev, "Cannot enable PCI device, aborting\n");
912 		return err;
913 	}
914 
915 	err = request_bar(pdev);
916 	if (err) {
917 		mlx5_core_err(dev, "error requesting BARs, aborting\n");
918 		goto err_disable;
919 	}
920 
921 	pci_set_master(pdev);
922 
923 	err = set_dma_caps(pdev);
924 	if (err) {
925 		mlx5_core_err(dev, "Failed setting DMA capabilities mask, aborting\n");
926 		goto err_clr_master;
927 	}
928 
929 	if (pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP32) &&
930 	    pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP64) &&
931 	    pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP128))
932 		mlx5_core_dbg(dev, "Enabling pci atomics failed\n");
933 
934 	dev->iseg_base = dev->bar_addr;
935 	dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
936 	if (!dev->iseg) {
937 		err = -ENOMEM;
938 		mlx5_core_err(dev, "Failed mapping initialization segment, aborting\n");
939 		goto err_clr_master;
940 	}
941 
942 	mlx5_pci_vsc_init(dev);
943 
944 	err = pci_enable_ptm(pdev, NULL);
945 	if (err)
946 		mlx5_core_info(dev, "PTM is not supported by PCIe\n");
947 
948 	return 0;
949 
950 err_clr_master:
951 	release_bar(dev->pdev);
952 err_disable:
953 	mlx5_pci_disable_device(dev);
954 	return err;
955 }
956 
mlx5_pci_close(struct mlx5_core_dev * dev)957 static void mlx5_pci_close(struct mlx5_core_dev *dev)
958 {
959 	/* health work might still be active, and it needs pci bar in
960 	 * order to know the NIC state. Therefore, drain the health WQ
961 	 * before removing the pci bars
962 	 */
963 	mlx5_drain_health_wq(dev);
964 	pci_disable_ptm(dev->pdev);
965 	iounmap(dev->iseg);
966 	release_bar(dev->pdev);
967 	mlx5_pci_disable_device(dev);
968 }
969 
mlx5_register_hca_devcom_comp(struct mlx5_core_dev * dev)970 static void mlx5_register_hca_devcom_comp(struct mlx5_core_dev *dev)
971 {
972 	/* This component is use to sync adding core_dev to lag_dev and to sync
973 	 * changes of mlx5_adev_devices between LAG layer and other layers.
974 	 */
975 	if (!mlx5_lag_is_supported(dev))
976 		return;
977 
978 	dev->priv.hca_devcom_comp =
979 		mlx5_devcom_register_component(dev->priv.devc, MLX5_DEVCOM_HCA_PORTS,
980 					       mlx5_query_nic_system_image_guid(dev),
981 					       NULL, dev);
982 	if (IS_ERR(dev->priv.hca_devcom_comp))
983 		mlx5_core_err(dev, "Failed to register devcom HCA component\n");
984 }
985 
mlx5_unregister_hca_devcom_comp(struct mlx5_core_dev * dev)986 static void mlx5_unregister_hca_devcom_comp(struct mlx5_core_dev *dev)
987 {
988 	mlx5_devcom_unregister_component(dev->priv.hca_devcom_comp);
989 }
990 
mlx5_init_once(struct mlx5_core_dev * dev)991 static int mlx5_init_once(struct mlx5_core_dev *dev)
992 {
993 	int err;
994 
995 	dev->priv.devc = mlx5_devcom_register_device(dev);
996 	if (IS_ERR(dev->priv.devc))
997 		mlx5_core_warn(dev, "failed to register devcom device %ld\n",
998 			       PTR_ERR(dev->priv.devc));
999 	mlx5_register_hca_devcom_comp(dev);
1000 
1001 	err = mlx5_query_board_id(dev);
1002 	if (err) {
1003 		mlx5_core_err(dev, "query board id failed\n");
1004 		goto err_devcom;
1005 	}
1006 
1007 	err = mlx5_irq_table_init(dev);
1008 	if (err) {
1009 		mlx5_core_err(dev, "failed to initialize irq table\n");
1010 		goto err_devcom;
1011 	}
1012 
1013 	err = mlx5_eq_table_init(dev);
1014 	if (err) {
1015 		mlx5_core_err(dev, "failed to initialize eq\n");
1016 		goto err_irq_cleanup;
1017 	}
1018 
1019 	err = mlx5_events_init(dev);
1020 	if (err) {
1021 		mlx5_core_err(dev, "failed to initialize events\n");
1022 		goto err_eq_cleanup;
1023 	}
1024 
1025 	err = mlx5_fw_reset_init(dev);
1026 	if (err) {
1027 		mlx5_core_err(dev, "failed to initialize fw reset events\n");
1028 		goto err_events_cleanup;
1029 	}
1030 
1031 	mlx5_cq_debugfs_init(dev);
1032 
1033 	mlx5_init_reserved_gids(dev);
1034 
1035 	mlx5_init_clock(dev);
1036 
1037 	dev->vxlan = mlx5_vxlan_create(dev);
1038 	dev->geneve = mlx5_geneve_create(dev);
1039 
1040 	err = mlx5_init_rl_table(dev);
1041 	if (err) {
1042 		mlx5_core_err(dev, "Failed to init rate limiting\n");
1043 		goto err_tables_cleanup;
1044 	}
1045 
1046 	err = mlx5_mpfs_init(dev);
1047 	if (err) {
1048 		mlx5_core_err(dev, "Failed to init l2 table %d\n", err);
1049 		goto err_rl_cleanup;
1050 	}
1051 
1052 	err = mlx5_sriov_init(dev);
1053 	if (err) {
1054 		mlx5_core_err(dev, "Failed to init sriov %d\n", err);
1055 		goto err_mpfs_cleanup;
1056 	}
1057 
1058 	err = mlx5_eswitch_init(dev);
1059 	if (err) {
1060 		mlx5_core_err(dev, "Failed to init eswitch %d\n", err);
1061 		goto err_sriov_cleanup;
1062 	}
1063 
1064 	err = mlx5_fpga_init(dev);
1065 	if (err) {
1066 		mlx5_core_err(dev, "Failed to init fpga device %d\n", err);
1067 		goto err_eswitch_cleanup;
1068 	}
1069 
1070 	err = mlx5_vhca_event_init(dev);
1071 	if (err) {
1072 		mlx5_core_err(dev, "Failed to init vhca event notifier %d\n", err);
1073 		goto err_fpga_cleanup;
1074 	}
1075 
1076 	err = mlx5_sf_hw_table_init(dev);
1077 	if (err) {
1078 		mlx5_core_err(dev, "Failed to init SF HW table %d\n", err);
1079 		goto err_sf_hw_table_cleanup;
1080 	}
1081 
1082 	err = mlx5_sf_table_init(dev);
1083 	if (err) {
1084 		mlx5_core_err(dev, "Failed to init SF table %d\n", err);
1085 		goto err_sf_table_cleanup;
1086 	}
1087 
1088 	err = mlx5_fs_core_alloc(dev);
1089 	if (err) {
1090 		mlx5_core_err(dev, "Failed to alloc flow steering\n");
1091 		goto err_fs;
1092 	}
1093 
1094 	dev->dm = mlx5_dm_create(dev);
1095 	if (IS_ERR(dev->dm))
1096 		mlx5_core_warn(dev, "Failed to init device memory %ld\n", PTR_ERR(dev->dm));
1097 
1098 	dev->tracer = mlx5_fw_tracer_create(dev);
1099 	dev->hv_vhca = mlx5_hv_vhca_create(dev);
1100 	dev->rsc_dump = mlx5_rsc_dump_create(dev);
1101 
1102 	return 0;
1103 
1104 err_fs:
1105 	mlx5_sf_table_cleanup(dev);
1106 err_sf_table_cleanup:
1107 	mlx5_sf_hw_table_cleanup(dev);
1108 err_sf_hw_table_cleanup:
1109 	mlx5_vhca_event_cleanup(dev);
1110 err_fpga_cleanup:
1111 	mlx5_fpga_cleanup(dev);
1112 err_eswitch_cleanup:
1113 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1114 err_sriov_cleanup:
1115 	mlx5_sriov_cleanup(dev);
1116 err_mpfs_cleanup:
1117 	mlx5_mpfs_cleanup(dev);
1118 err_rl_cleanup:
1119 	mlx5_cleanup_rl_table(dev);
1120 err_tables_cleanup:
1121 	mlx5_geneve_destroy(dev->geneve);
1122 	mlx5_vxlan_destroy(dev->vxlan);
1123 	mlx5_cleanup_clock(dev);
1124 	mlx5_cleanup_reserved_gids(dev);
1125 	mlx5_cq_debugfs_cleanup(dev);
1126 	mlx5_fw_reset_cleanup(dev);
1127 err_events_cleanup:
1128 	mlx5_events_cleanup(dev);
1129 err_eq_cleanup:
1130 	mlx5_eq_table_cleanup(dev);
1131 err_irq_cleanup:
1132 	mlx5_irq_table_cleanup(dev);
1133 err_devcom:
1134 	mlx5_unregister_hca_devcom_comp(dev);
1135 	mlx5_devcom_unregister_device(dev->priv.devc);
1136 
1137 	return err;
1138 }
1139 
mlx5_cleanup_once(struct mlx5_core_dev * dev)1140 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
1141 {
1142 	mlx5_rsc_dump_destroy(dev);
1143 	mlx5_hv_vhca_destroy(dev->hv_vhca);
1144 	mlx5_fw_tracer_destroy(dev->tracer);
1145 	mlx5_dm_cleanup(dev);
1146 	mlx5_fs_core_free(dev);
1147 	mlx5_sf_table_cleanup(dev);
1148 	mlx5_sf_hw_table_cleanup(dev);
1149 	mlx5_vhca_event_cleanup(dev);
1150 	mlx5_fpga_cleanup(dev);
1151 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1152 	mlx5_sriov_cleanup(dev);
1153 	mlx5_mpfs_cleanup(dev);
1154 	mlx5_cleanup_rl_table(dev);
1155 	mlx5_geneve_destroy(dev->geneve);
1156 	mlx5_vxlan_destroy(dev->vxlan);
1157 	mlx5_cleanup_clock(dev);
1158 	mlx5_cleanup_reserved_gids(dev);
1159 	mlx5_cq_debugfs_cleanup(dev);
1160 	mlx5_fw_reset_cleanup(dev);
1161 	mlx5_events_cleanup(dev);
1162 	mlx5_eq_table_cleanup(dev);
1163 	mlx5_irq_table_cleanup(dev);
1164 	mlx5_unregister_hca_devcom_comp(dev);
1165 	mlx5_devcom_unregister_device(dev->priv.devc);
1166 }
1167 
mlx5_function_enable(struct mlx5_core_dev * dev,bool boot,u64 timeout)1168 static int mlx5_function_enable(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1169 {
1170 	int err;
1171 
1172 	mlx5_core_info(dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1173 		       fw_rev_min(dev), fw_rev_sub(dev));
1174 
1175 	/* Only PFs hold the relevant PCIe information for this query */
1176 	if (mlx5_core_is_pf(dev))
1177 		pcie_print_link_status(dev->pdev);
1178 
1179 	/* wait for firmware to accept initialization segments configurations
1180 	 */
1181 	err = wait_fw_init(dev, timeout,
1182 			   mlx5_tout_ms(dev, FW_PRE_INIT_WARN_MESSAGE_INTERVAL),
1183 			   "pre-initializing");
1184 	if (err)
1185 		return err;
1186 
1187 	err = mlx5_cmd_enable(dev);
1188 	if (err) {
1189 		mlx5_core_err(dev, "Failed initializing command interface, aborting\n");
1190 		return err;
1191 	}
1192 
1193 	mlx5_tout_query_iseg(dev);
1194 
1195 	err = wait_fw_init(dev, mlx5_tout_ms(dev, FW_INIT), 0, "initializing");
1196 	if (err)
1197 		goto err_cmd_cleanup;
1198 
1199 	dev->caps.embedded_cpu = mlx5_read_embedded_cpu(dev);
1200 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_UP);
1201 
1202 	mlx5_start_health_poll(dev);
1203 
1204 	err = mlx5_core_enable_hca(dev, 0);
1205 	if (err) {
1206 		mlx5_core_err(dev, "enable hca failed\n");
1207 		goto stop_health_poll;
1208 	}
1209 
1210 	err = mlx5_core_set_issi(dev);
1211 	if (err) {
1212 		mlx5_core_err(dev, "failed to set issi\n");
1213 		goto err_disable_hca;
1214 	}
1215 
1216 	err = mlx5_satisfy_startup_pages(dev, 1);
1217 	if (err) {
1218 		mlx5_core_err(dev, "failed to allocate boot pages\n");
1219 		goto err_disable_hca;
1220 	}
1221 
1222 	err = mlx5_tout_query_dtor(dev);
1223 	if (err) {
1224 		mlx5_core_err(dev, "failed to read dtor\n");
1225 		goto reclaim_boot_pages;
1226 	}
1227 
1228 	return 0;
1229 
1230 reclaim_boot_pages:
1231 	mlx5_reclaim_startup_pages(dev);
1232 err_disable_hca:
1233 	mlx5_core_disable_hca(dev, 0);
1234 stop_health_poll:
1235 	mlx5_stop_health_poll(dev, boot);
1236 err_cmd_cleanup:
1237 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1238 	mlx5_cmd_disable(dev);
1239 
1240 	return err;
1241 }
1242 
mlx5_function_disable(struct mlx5_core_dev * dev,bool boot)1243 static void mlx5_function_disable(struct mlx5_core_dev *dev, bool boot)
1244 {
1245 	mlx5_reclaim_startup_pages(dev);
1246 	mlx5_core_disable_hca(dev, 0);
1247 	mlx5_stop_health_poll(dev, boot);
1248 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1249 	mlx5_cmd_disable(dev);
1250 }
1251 
mlx5_function_open(struct mlx5_core_dev * dev)1252 static int mlx5_function_open(struct mlx5_core_dev *dev)
1253 {
1254 	int err;
1255 
1256 	err = set_hca_ctrl(dev);
1257 	if (err) {
1258 		mlx5_core_err(dev, "set_hca_ctrl failed\n");
1259 		return err;
1260 	}
1261 
1262 	err = set_hca_cap(dev);
1263 	if (err) {
1264 		mlx5_core_err(dev, "set_hca_cap failed\n");
1265 		return err;
1266 	}
1267 
1268 	err = mlx5_satisfy_startup_pages(dev, 0);
1269 	if (err) {
1270 		mlx5_core_err(dev, "failed to allocate init pages\n");
1271 		return err;
1272 	}
1273 
1274 	err = mlx5_cmd_init_hca(dev, sw_owner_id);
1275 	if (err) {
1276 		mlx5_core_err(dev, "init hca failed\n");
1277 		return err;
1278 	}
1279 
1280 	mlx5_set_driver_version(dev);
1281 
1282 	err = mlx5_query_hca_caps(dev);
1283 	if (err) {
1284 		mlx5_core_err(dev, "query hca failed\n");
1285 		return err;
1286 	}
1287 	mlx5_start_health_fw_log_up(dev);
1288 	return 0;
1289 }
1290 
mlx5_function_close(struct mlx5_core_dev * dev)1291 static int mlx5_function_close(struct mlx5_core_dev *dev)
1292 {
1293 	int err;
1294 
1295 	err = mlx5_cmd_teardown_hca(dev);
1296 	if (err) {
1297 		mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n");
1298 		return err;
1299 	}
1300 
1301 	return 0;
1302 }
1303 
mlx5_function_setup(struct mlx5_core_dev * dev,bool boot,u64 timeout)1304 static int mlx5_function_setup(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1305 {
1306 	int err;
1307 
1308 	err = mlx5_function_enable(dev, boot, timeout);
1309 	if (err)
1310 		return err;
1311 
1312 	err = mlx5_function_open(dev);
1313 	if (err)
1314 		mlx5_function_disable(dev, boot);
1315 	return err;
1316 }
1317 
mlx5_function_teardown(struct mlx5_core_dev * dev,bool boot)1318 static int mlx5_function_teardown(struct mlx5_core_dev *dev, bool boot)
1319 {
1320 	int err = mlx5_function_close(dev);
1321 
1322 	if (!err)
1323 		mlx5_function_disable(dev, boot);
1324 	else
1325 		mlx5_stop_health_poll(dev, boot);
1326 
1327 	return err;
1328 }
1329 
mlx5_load(struct mlx5_core_dev * dev)1330 static int mlx5_load(struct mlx5_core_dev *dev)
1331 {
1332 	int err;
1333 
1334 	dev->priv.uar = mlx5_get_uars_page(dev);
1335 	if (IS_ERR(dev->priv.uar)) {
1336 		mlx5_core_err(dev, "Failed allocating uar, aborting\n");
1337 		err = PTR_ERR(dev->priv.uar);
1338 		return err;
1339 	}
1340 
1341 	mlx5_events_start(dev);
1342 	mlx5_pagealloc_start(dev);
1343 
1344 	err = mlx5_irq_table_create(dev);
1345 	if (err) {
1346 		mlx5_core_err(dev, "Failed to alloc IRQs\n");
1347 		goto err_irq_table;
1348 	}
1349 
1350 	err = mlx5_eq_table_create(dev);
1351 	if (err) {
1352 		mlx5_core_err(dev, "Failed to create EQs\n");
1353 		goto err_eq_table;
1354 	}
1355 
1356 	err = mlx5_fw_tracer_init(dev->tracer);
1357 	if (err) {
1358 		mlx5_core_err(dev, "Failed to init FW tracer %d\n", err);
1359 		mlx5_fw_tracer_destroy(dev->tracer);
1360 		dev->tracer = NULL;
1361 	}
1362 
1363 	mlx5_fw_reset_events_start(dev);
1364 	mlx5_hv_vhca_init(dev->hv_vhca);
1365 
1366 	err = mlx5_rsc_dump_init(dev);
1367 	if (err) {
1368 		mlx5_core_err(dev, "Failed to init Resource dump %d\n", err);
1369 		mlx5_rsc_dump_destroy(dev);
1370 		dev->rsc_dump = NULL;
1371 	}
1372 
1373 	err = mlx5_fpga_device_start(dev);
1374 	if (err) {
1375 		mlx5_core_err(dev, "fpga device start failed %d\n", err);
1376 		goto err_fpga_start;
1377 	}
1378 
1379 	err = mlx5_fs_core_init(dev);
1380 	if (err) {
1381 		mlx5_core_err(dev, "Failed to init flow steering\n");
1382 		goto err_fs;
1383 	}
1384 
1385 	err = mlx5_core_set_hca_defaults(dev);
1386 	if (err) {
1387 		mlx5_core_err(dev, "Failed to set hca defaults\n");
1388 		goto err_set_hca;
1389 	}
1390 
1391 	mlx5_vhca_event_start(dev);
1392 
1393 	err = mlx5_sf_hw_table_create(dev);
1394 	if (err) {
1395 		mlx5_core_err(dev, "sf table create failed %d\n", err);
1396 		goto err_vhca;
1397 	}
1398 
1399 	err = mlx5_ec_init(dev);
1400 	if (err) {
1401 		mlx5_core_err(dev, "Failed to init embedded CPU\n");
1402 		goto err_ec;
1403 	}
1404 
1405 	mlx5_lag_add_mdev(dev);
1406 	err = mlx5_sriov_attach(dev);
1407 	if (err) {
1408 		mlx5_core_err(dev, "sriov init failed %d\n", err);
1409 		goto err_sriov;
1410 	}
1411 
1412 	mlx5_sf_dev_table_create(dev);
1413 
1414 	err = mlx5_devlink_traps_register(priv_to_devlink(dev));
1415 	if (err)
1416 		goto err_traps_reg;
1417 
1418 	return 0;
1419 
1420 err_traps_reg:
1421 	mlx5_sf_dev_table_destroy(dev);
1422 	mlx5_sriov_detach(dev);
1423 err_sriov:
1424 	mlx5_lag_remove_mdev(dev);
1425 	mlx5_ec_cleanup(dev);
1426 err_ec:
1427 	mlx5_sf_hw_table_destroy(dev);
1428 err_vhca:
1429 	mlx5_vhca_event_stop(dev);
1430 err_set_hca:
1431 	mlx5_fs_core_cleanup(dev);
1432 err_fs:
1433 	mlx5_fpga_device_stop(dev);
1434 err_fpga_start:
1435 	mlx5_rsc_dump_cleanup(dev);
1436 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1437 	mlx5_fw_reset_events_stop(dev);
1438 	mlx5_fw_tracer_cleanup(dev->tracer);
1439 	mlx5_eq_table_destroy(dev);
1440 err_eq_table:
1441 	mlx5_irq_table_destroy(dev);
1442 err_irq_table:
1443 	mlx5_pagealloc_stop(dev);
1444 	mlx5_events_stop(dev);
1445 	mlx5_put_uars_page(dev, dev->priv.uar);
1446 	return err;
1447 }
1448 
mlx5_unload(struct mlx5_core_dev * dev)1449 static void mlx5_unload(struct mlx5_core_dev *dev)
1450 {
1451 	mlx5_eswitch_disable(dev->priv.eswitch);
1452 	mlx5_devlink_traps_unregister(priv_to_devlink(dev));
1453 	mlx5_sf_dev_table_destroy(dev);
1454 	mlx5_sriov_detach(dev);
1455 	mlx5_lag_remove_mdev(dev);
1456 	mlx5_ec_cleanup(dev);
1457 	mlx5_sf_hw_table_destroy(dev);
1458 	mlx5_vhca_event_stop(dev);
1459 	mlx5_fs_core_cleanup(dev);
1460 	mlx5_fpga_device_stop(dev);
1461 	mlx5_rsc_dump_cleanup(dev);
1462 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1463 	mlx5_fw_reset_events_stop(dev);
1464 	mlx5_fw_tracer_cleanup(dev->tracer);
1465 	mlx5_eq_table_destroy(dev);
1466 	mlx5_irq_table_destroy(dev);
1467 	mlx5_pagealloc_stop(dev);
1468 	mlx5_events_stop(dev);
1469 	mlx5_put_uars_page(dev, dev->priv.uar);
1470 }
1471 
mlx5_init_one_devl_locked(struct mlx5_core_dev * dev)1472 int mlx5_init_one_devl_locked(struct mlx5_core_dev *dev)
1473 {
1474 	bool light_probe = mlx5_dev_is_lightweight(dev);
1475 	int err = 0;
1476 
1477 	mutex_lock(&dev->intf_state_mutex);
1478 	dev->state = MLX5_DEVICE_STATE_UP;
1479 
1480 	err = mlx5_function_setup(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1481 	if (err)
1482 		goto err_function;
1483 
1484 	err = mlx5_init_once(dev);
1485 	if (err) {
1486 		mlx5_core_err(dev, "sw objs init failed\n");
1487 		goto function_teardown;
1488 	}
1489 
1490 	/* In case of light_probe, mlx5_devlink is already registered.
1491 	 * Hence, don't register devlink again.
1492 	 */
1493 	if (!light_probe) {
1494 		err = mlx5_devlink_params_register(priv_to_devlink(dev));
1495 		if (err)
1496 			goto err_devlink_params_reg;
1497 	}
1498 
1499 	err = mlx5_load(dev);
1500 	if (err)
1501 		goto err_load;
1502 
1503 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1504 
1505 	err = mlx5_register_device(dev);
1506 	if (err)
1507 		goto err_register;
1508 
1509 	err = mlx5_crdump_enable(dev);
1510 	if (err)
1511 		mlx5_core_err(dev, "mlx5_crdump_enable failed with error code %d\n", err);
1512 
1513 	err = mlx5_hwmon_dev_register(dev);
1514 	if (err)
1515 		mlx5_core_err(dev, "mlx5_hwmon_dev_register failed with error code %d\n", err);
1516 
1517 	mutex_unlock(&dev->intf_state_mutex);
1518 	return 0;
1519 
1520 err_register:
1521 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1522 	mlx5_unload(dev);
1523 err_load:
1524 	if (!light_probe)
1525 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1526 err_devlink_params_reg:
1527 	mlx5_cleanup_once(dev);
1528 function_teardown:
1529 	mlx5_function_teardown(dev, true);
1530 err_function:
1531 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1532 	mutex_unlock(&dev->intf_state_mutex);
1533 	return err;
1534 }
1535 
mlx5_init_one(struct mlx5_core_dev * dev)1536 int mlx5_init_one(struct mlx5_core_dev *dev)
1537 {
1538 	struct devlink *devlink = priv_to_devlink(dev);
1539 	int err;
1540 
1541 	devl_lock(devlink);
1542 	devl_register(devlink);
1543 	err = mlx5_init_one_devl_locked(dev);
1544 	if (err)
1545 		devl_unregister(devlink);
1546 	devl_unlock(devlink);
1547 	return err;
1548 }
1549 
mlx5_uninit_one(struct mlx5_core_dev * dev)1550 void mlx5_uninit_one(struct mlx5_core_dev *dev)
1551 {
1552 	struct devlink *devlink = priv_to_devlink(dev);
1553 
1554 	devl_lock(devlink);
1555 	mutex_lock(&dev->intf_state_mutex);
1556 
1557 	mlx5_hwmon_dev_unregister(dev);
1558 	mlx5_crdump_disable(dev);
1559 	mlx5_unregister_device(dev);
1560 
1561 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1562 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1563 			       __func__);
1564 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1565 		mlx5_cleanup_once(dev);
1566 		goto out;
1567 	}
1568 
1569 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1570 	mlx5_unload(dev);
1571 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1572 	mlx5_cleanup_once(dev);
1573 	mlx5_function_teardown(dev, true);
1574 out:
1575 	mutex_unlock(&dev->intf_state_mutex);
1576 	devl_unregister(devlink);
1577 	devl_unlock(devlink);
1578 }
1579 
mlx5_load_one_devl_locked(struct mlx5_core_dev * dev,bool recovery)1580 int mlx5_load_one_devl_locked(struct mlx5_core_dev *dev, bool recovery)
1581 {
1582 	int err = 0;
1583 	u64 timeout;
1584 
1585 	devl_assert_locked(priv_to_devlink(dev));
1586 	mutex_lock(&dev->intf_state_mutex);
1587 	if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1588 		mlx5_core_warn(dev, "interface is up, NOP\n");
1589 		goto out;
1590 	}
1591 	/* remove any previous indication of internal error */
1592 	dev->state = MLX5_DEVICE_STATE_UP;
1593 
1594 	if (recovery)
1595 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_ON_RECOVERY_TIMEOUT);
1596 	else
1597 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT);
1598 	err = mlx5_function_setup(dev, false, timeout);
1599 	if (err)
1600 		goto err_function;
1601 
1602 	err = mlx5_load(dev);
1603 	if (err)
1604 		goto err_load;
1605 
1606 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1607 
1608 	err = mlx5_attach_device(dev);
1609 	if (err)
1610 		goto err_attach;
1611 
1612 	mutex_unlock(&dev->intf_state_mutex);
1613 	return 0;
1614 
1615 err_attach:
1616 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1617 	mlx5_unload(dev);
1618 err_load:
1619 	mlx5_function_teardown(dev, false);
1620 err_function:
1621 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1622 out:
1623 	mutex_unlock(&dev->intf_state_mutex);
1624 	return err;
1625 }
1626 
mlx5_load_one(struct mlx5_core_dev * dev,bool recovery)1627 int mlx5_load_one(struct mlx5_core_dev *dev, bool recovery)
1628 {
1629 	struct devlink *devlink = priv_to_devlink(dev);
1630 	int ret;
1631 
1632 	devl_lock(devlink);
1633 	ret = mlx5_load_one_devl_locked(dev, recovery);
1634 	devl_unlock(devlink);
1635 	return ret;
1636 }
1637 
mlx5_unload_one_devl_locked(struct mlx5_core_dev * dev,bool suspend)1638 void mlx5_unload_one_devl_locked(struct mlx5_core_dev *dev, bool suspend)
1639 {
1640 	devl_assert_locked(priv_to_devlink(dev));
1641 	mutex_lock(&dev->intf_state_mutex);
1642 
1643 	mlx5_detach_device(dev, suspend);
1644 
1645 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1646 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1647 			       __func__);
1648 		goto out;
1649 	}
1650 
1651 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1652 	mlx5_unload(dev);
1653 	mlx5_function_teardown(dev, false);
1654 out:
1655 	mutex_unlock(&dev->intf_state_mutex);
1656 }
1657 
mlx5_unload_one(struct mlx5_core_dev * dev,bool suspend)1658 void mlx5_unload_one(struct mlx5_core_dev *dev, bool suspend)
1659 {
1660 	struct devlink *devlink = priv_to_devlink(dev);
1661 
1662 	devl_lock(devlink);
1663 	mlx5_unload_one_devl_locked(dev, suspend);
1664 	devl_unlock(devlink);
1665 }
1666 
1667 /* In case of light probe, we don't need a full query of hca_caps, but only the bellow caps.
1668  * A full query of hca_caps will be done when the device will reload.
1669  */
mlx5_query_hca_caps_light(struct mlx5_core_dev * dev)1670 static int mlx5_query_hca_caps_light(struct mlx5_core_dev *dev)
1671 {
1672 	int err;
1673 
1674 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
1675 	if (err)
1676 		return err;
1677 
1678 	if (MLX5_CAP_GEN(dev, eth_net_offloads)) {
1679 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_ETHERNET_OFFLOADS,
1680 					      HCA_CAP_OPMOD_GET_CUR);
1681 		if (err)
1682 			return err;
1683 	}
1684 
1685 	if (MLX5_CAP_GEN(dev, nic_flow_table) ||
1686 	    MLX5_CAP_GEN(dev, ipoib_enhanced_offloads)) {
1687 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_FLOW_TABLE,
1688 					      HCA_CAP_OPMOD_GET_CUR);
1689 		if (err)
1690 			return err;
1691 	}
1692 
1693 	if (MLX5_CAP_GEN_64(dev, general_obj_types) &
1694 		MLX5_GENERAL_OBJ_TYPES_CAP_VIRTIO_NET_Q) {
1695 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_VDPA_EMULATION,
1696 					      HCA_CAP_OPMOD_GET_CUR);
1697 		if (err)
1698 			return err;
1699 	}
1700 
1701 	return 0;
1702 }
1703 
mlx5_init_one_light(struct mlx5_core_dev * dev)1704 int mlx5_init_one_light(struct mlx5_core_dev *dev)
1705 {
1706 	struct devlink *devlink = priv_to_devlink(dev);
1707 	int err;
1708 
1709 	devl_lock(devlink);
1710 	devl_register(devlink);
1711 	dev->state = MLX5_DEVICE_STATE_UP;
1712 	err = mlx5_function_enable(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1713 	if (err) {
1714 		mlx5_core_warn(dev, "mlx5_function_enable err=%d\n", err);
1715 		goto out;
1716 	}
1717 
1718 	err = mlx5_query_hca_caps_light(dev);
1719 	if (err) {
1720 		mlx5_core_warn(dev, "mlx5_query_hca_caps_light err=%d\n", err);
1721 		goto query_hca_caps_err;
1722 	}
1723 
1724 	err = mlx5_devlink_params_register(priv_to_devlink(dev));
1725 	if (err) {
1726 		mlx5_core_warn(dev, "mlx5_devlink_param_reg err = %d\n", err);
1727 		goto query_hca_caps_err;
1728 	}
1729 
1730 	devl_unlock(devlink);
1731 	return 0;
1732 
1733 query_hca_caps_err:
1734 	mlx5_function_disable(dev, true);
1735 out:
1736 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1737 	devl_unregister(devlink);
1738 	devl_unlock(devlink);
1739 	return err;
1740 }
1741 
mlx5_uninit_one_light(struct mlx5_core_dev * dev)1742 void mlx5_uninit_one_light(struct mlx5_core_dev *dev)
1743 {
1744 	struct devlink *devlink = priv_to_devlink(dev);
1745 
1746 	devl_lock(devlink);
1747 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1748 	devl_unregister(devlink);
1749 	devl_unlock(devlink);
1750 	if (dev->state != MLX5_DEVICE_STATE_UP)
1751 		return;
1752 	mlx5_function_disable(dev, true);
1753 }
1754 
1755 /* xxx_light() function are used in order to configure the device without full
1756  * init (light init). e.g.: There isn't a point in reload a device to light state.
1757  * Hence, mlx5_load_one_light() isn't needed.
1758  */
1759 
mlx5_unload_one_light(struct mlx5_core_dev * dev)1760 void mlx5_unload_one_light(struct mlx5_core_dev *dev)
1761 {
1762 	if (dev->state != MLX5_DEVICE_STATE_UP)
1763 		return;
1764 	mlx5_function_disable(dev, false);
1765 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1766 }
1767 
1768 static const int types[] = {
1769 	MLX5_CAP_GENERAL,
1770 	MLX5_CAP_GENERAL_2,
1771 	MLX5_CAP_ETHERNET_OFFLOADS,
1772 	MLX5_CAP_IPOIB_ENHANCED_OFFLOADS,
1773 	MLX5_CAP_ODP,
1774 	MLX5_CAP_ATOMIC,
1775 	MLX5_CAP_ROCE,
1776 	MLX5_CAP_IPOIB_OFFLOADS,
1777 	MLX5_CAP_FLOW_TABLE,
1778 	MLX5_CAP_ESWITCH_FLOW_TABLE,
1779 	MLX5_CAP_ESWITCH,
1780 	MLX5_CAP_QOS,
1781 	MLX5_CAP_DEBUG,
1782 	MLX5_CAP_DEV_MEM,
1783 	MLX5_CAP_DEV_EVENT,
1784 	MLX5_CAP_TLS,
1785 	MLX5_CAP_VDPA_EMULATION,
1786 	MLX5_CAP_IPSEC,
1787 	MLX5_CAP_PORT_SELECTION,
1788 	MLX5_CAP_MACSEC,
1789 	MLX5_CAP_ADV_VIRTUALIZATION,
1790 	MLX5_CAP_CRYPTO,
1791 };
1792 
mlx5_hca_caps_free(struct mlx5_core_dev * dev)1793 static void mlx5_hca_caps_free(struct mlx5_core_dev *dev)
1794 {
1795 	int type;
1796 	int i;
1797 
1798 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1799 		type = types[i];
1800 		kfree(dev->caps.hca[type]);
1801 	}
1802 }
1803 
mlx5_hca_caps_alloc(struct mlx5_core_dev * dev)1804 static int mlx5_hca_caps_alloc(struct mlx5_core_dev *dev)
1805 {
1806 	struct mlx5_hca_cap *cap;
1807 	int type;
1808 	int i;
1809 
1810 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1811 		cap = kzalloc(sizeof(*cap), GFP_KERNEL);
1812 		if (!cap)
1813 			goto err;
1814 		type = types[i];
1815 		dev->caps.hca[type] = cap;
1816 	}
1817 
1818 	return 0;
1819 
1820 err:
1821 	mlx5_hca_caps_free(dev);
1822 	return -ENOMEM;
1823 }
1824 
vhca_id_show(struct seq_file * file,void * priv)1825 static int vhca_id_show(struct seq_file *file, void *priv)
1826 {
1827 	struct mlx5_core_dev *dev = file->private;
1828 
1829 	seq_printf(file, "0x%x\n", MLX5_CAP_GEN(dev, vhca_id));
1830 	return 0;
1831 }
1832 
1833 DEFINE_SHOW_ATTRIBUTE(vhca_id);
1834 
mlx5_mdev_init(struct mlx5_core_dev * dev,int profile_idx)1835 int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
1836 {
1837 	struct mlx5_priv *priv = &dev->priv;
1838 	int err;
1839 
1840 	memcpy(&dev->profile, &profile[profile_idx], sizeof(dev->profile));
1841 	lockdep_register_key(&dev->lock_key);
1842 	mutex_init(&dev->intf_state_mutex);
1843 	lockdep_set_class(&dev->intf_state_mutex, &dev->lock_key);
1844 	mutex_init(&dev->mlx5e_res.uplink_netdev_lock);
1845 	mutex_init(&dev->wc_state_lock);
1846 
1847 	mutex_init(&priv->bfregs.reg_head.lock);
1848 	mutex_init(&priv->bfregs.wc_head.lock);
1849 	INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1850 	INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1851 
1852 	mutex_init(&priv->alloc_mutex);
1853 	mutex_init(&priv->pgdir_mutex);
1854 	INIT_LIST_HEAD(&priv->pgdir_list);
1855 
1856 	priv->numa_node = dev_to_node(mlx5_core_dma_dev(dev));
1857 	priv->dbg.dbg_root = debugfs_create_dir(dev_name(dev->device),
1858 						mlx5_debugfs_root);
1859 	debugfs_create_file("vhca_id", 0400, priv->dbg.dbg_root, dev, &vhca_id_fops);
1860 	INIT_LIST_HEAD(&priv->traps);
1861 
1862 	err = mlx5_cmd_init(dev);
1863 	if (err) {
1864 		mlx5_core_err(dev, "Failed initializing cmdif SW structs, aborting\n");
1865 		goto err_cmd_init;
1866 	}
1867 
1868 	err = mlx5_tout_init(dev);
1869 	if (err) {
1870 		mlx5_core_err(dev, "Failed initializing timeouts, aborting\n");
1871 		goto err_timeout_init;
1872 	}
1873 
1874 	err = mlx5_health_init(dev);
1875 	if (err)
1876 		goto err_health_init;
1877 
1878 	err = mlx5_pagealloc_init(dev);
1879 	if (err)
1880 		goto err_pagealloc_init;
1881 
1882 	err = mlx5_adev_init(dev);
1883 	if (err)
1884 		goto err_adev_init;
1885 
1886 	err = mlx5_hca_caps_alloc(dev);
1887 	if (err)
1888 		goto err_hca_caps;
1889 
1890 	/* The conjunction of sw_vhca_id with sw_owner_id will be a global
1891 	 * unique id per function which uses mlx5_core.
1892 	 * Those values are supplied to FW as part of the init HCA command to
1893 	 * be used by both driver and FW when it's applicable.
1894 	 */
1895 	dev->priv.sw_vhca_id = ida_alloc_range(&sw_vhca_ida, 1,
1896 					       MAX_SW_VHCA_ID,
1897 					       GFP_KERNEL);
1898 	if (dev->priv.sw_vhca_id < 0)
1899 		mlx5_core_err(dev, "failed to allocate sw_vhca_id, err=%d\n",
1900 			      dev->priv.sw_vhca_id);
1901 
1902 	return 0;
1903 
1904 err_hca_caps:
1905 	mlx5_adev_cleanup(dev);
1906 err_adev_init:
1907 	mlx5_pagealloc_cleanup(dev);
1908 err_pagealloc_init:
1909 	mlx5_health_cleanup(dev);
1910 err_health_init:
1911 	mlx5_tout_cleanup(dev);
1912 err_timeout_init:
1913 	mlx5_cmd_cleanup(dev);
1914 err_cmd_init:
1915 	debugfs_remove(dev->priv.dbg.dbg_root);
1916 	mutex_destroy(&priv->pgdir_mutex);
1917 	mutex_destroy(&priv->alloc_mutex);
1918 	mutex_destroy(&priv->bfregs.wc_head.lock);
1919 	mutex_destroy(&priv->bfregs.reg_head.lock);
1920 	mutex_destroy(&dev->intf_state_mutex);
1921 	lockdep_unregister_key(&dev->lock_key);
1922 	return err;
1923 }
1924 
mlx5_mdev_uninit(struct mlx5_core_dev * dev)1925 void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
1926 {
1927 	struct mlx5_priv *priv = &dev->priv;
1928 
1929 	if (priv->sw_vhca_id > 0)
1930 		ida_free(&sw_vhca_ida, dev->priv.sw_vhca_id);
1931 
1932 	mlx5_hca_caps_free(dev);
1933 	mlx5_adev_cleanup(dev);
1934 	mlx5_pagealloc_cleanup(dev);
1935 	mlx5_health_cleanup(dev);
1936 	mlx5_tout_cleanup(dev);
1937 	mlx5_cmd_cleanup(dev);
1938 	debugfs_remove_recursive(dev->priv.dbg.dbg_root);
1939 	mutex_destroy(&priv->pgdir_mutex);
1940 	mutex_destroy(&priv->alloc_mutex);
1941 	mutex_destroy(&priv->bfregs.wc_head.lock);
1942 	mutex_destroy(&priv->bfregs.reg_head.lock);
1943 	mutex_destroy(&dev->wc_state_lock);
1944 	mutex_destroy(&dev->mlx5e_res.uplink_netdev_lock);
1945 	mutex_destroy(&dev->intf_state_mutex);
1946 	lockdep_unregister_key(&dev->lock_key);
1947 }
1948 
probe_one(struct pci_dev * pdev,const struct pci_device_id * id)1949 static int probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
1950 {
1951 	struct mlx5_core_dev *dev;
1952 	struct devlink *devlink;
1953 	int err;
1954 
1955 	devlink = mlx5_devlink_alloc(&pdev->dev);
1956 	if (!devlink) {
1957 		dev_err(&pdev->dev, "devlink alloc failed\n");
1958 		return -ENOMEM;
1959 	}
1960 
1961 	dev = devlink_priv(devlink);
1962 	dev->device = &pdev->dev;
1963 	dev->pdev = pdev;
1964 
1965 	dev->coredev_type = id->driver_data & MLX5_PCI_DEV_IS_VF ?
1966 			 MLX5_COREDEV_VF : MLX5_COREDEV_PF;
1967 
1968 	dev->priv.adev_idx = mlx5_adev_idx_alloc();
1969 	if (dev->priv.adev_idx < 0) {
1970 		err = dev->priv.adev_idx;
1971 		goto adev_init_err;
1972 	}
1973 
1974 	err = mlx5_mdev_init(dev, prof_sel);
1975 	if (err)
1976 		goto mdev_init_err;
1977 
1978 	err = mlx5_pci_init(dev, pdev, id);
1979 	if (err) {
1980 		mlx5_core_err(dev, "mlx5_pci_init failed with error code %d\n",
1981 			      err);
1982 		goto pci_init_err;
1983 	}
1984 
1985 	err = mlx5_init_one(dev);
1986 	if (err) {
1987 		mlx5_core_err(dev, "mlx5_init_one failed with error code %d\n",
1988 			      err);
1989 		goto err_init_one;
1990 	}
1991 
1992 	pci_save_state(pdev);
1993 	return 0;
1994 
1995 err_init_one:
1996 	mlx5_pci_close(dev);
1997 pci_init_err:
1998 	mlx5_mdev_uninit(dev);
1999 mdev_init_err:
2000 	mlx5_adev_idx_free(dev->priv.adev_idx);
2001 adev_init_err:
2002 	mlx5_devlink_free(devlink);
2003 
2004 	return err;
2005 }
2006 
remove_one(struct pci_dev * pdev)2007 static void remove_one(struct pci_dev *pdev)
2008 {
2009 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2010 	struct devlink *devlink = priv_to_devlink(dev);
2011 
2012 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2013 	mlx5_drain_fw_reset(dev);
2014 	mlx5_drain_health_wq(dev);
2015 	mlx5_sriov_disable(pdev, false);
2016 	mlx5_uninit_one(dev);
2017 	mlx5_pci_close(dev);
2018 	mlx5_mdev_uninit(dev);
2019 	mlx5_adev_idx_free(dev->priv.adev_idx);
2020 	mlx5_devlink_free(devlink);
2021 }
2022 
2023 #define mlx5_pci_trace(dev, fmt, ...) ({ \
2024 	struct mlx5_core_dev *__dev = (dev); \
2025 	mlx5_core_info(__dev, "%s Device state = %d health sensors: %d pci_status: %d. " fmt, \
2026 		       __func__, __dev->state, mlx5_health_check_fatal_sensors(__dev), \
2027 		       __dev->pci_status, ##__VA_ARGS__); \
2028 })
2029 
result2str(enum pci_ers_result result)2030 static const char *result2str(enum pci_ers_result result)
2031 {
2032 	return  result == PCI_ERS_RESULT_NEED_RESET ? "need reset" :
2033 		result == PCI_ERS_RESULT_DISCONNECT ? "disconnect" :
2034 		result == PCI_ERS_RESULT_RECOVERED  ? "recovered" :
2035 		"unknown";
2036 }
2037 
mlx5_pci_err_detected(struct pci_dev * pdev,pci_channel_state_t state)2038 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
2039 					      pci_channel_state_t state)
2040 {
2041 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2042 	enum pci_ers_result res;
2043 
2044 	mlx5_pci_trace(dev, "Enter, pci channel state = %d\n", state);
2045 
2046 	mlx5_enter_error_state(dev, false);
2047 	mlx5_error_sw_reset(dev);
2048 	mlx5_unload_one(dev, false);
2049 	mlx5_drain_health_wq(dev);
2050 	mlx5_pci_disable_device(dev);
2051 
2052 	res = state == pci_channel_io_perm_failure ?
2053 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2054 
2055 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, result = %d, %s\n",
2056 		       __func__, dev->state, dev->pci_status, res, result2str(res));
2057 	return res;
2058 }
2059 
2060 /* wait for the device to show vital signs by waiting
2061  * for the health counter to start counting.
2062  */
wait_vital(struct pci_dev * pdev)2063 static int wait_vital(struct pci_dev *pdev)
2064 {
2065 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2066 	struct mlx5_core_health *health = &dev->priv.health;
2067 	const int niter = 100;
2068 	u32 last_count = 0;
2069 	u32 count;
2070 	int i;
2071 
2072 	for (i = 0; i < niter; i++) {
2073 		count = ioread32be(health->health_counter);
2074 		if (count && count != 0xffffffff) {
2075 			if (last_count && last_count != count) {
2076 				mlx5_core_info(dev,
2077 					       "wait vital counter value 0x%x after %d iterations\n",
2078 					       count, i);
2079 				return 0;
2080 			}
2081 			last_count = count;
2082 		}
2083 		msleep(50);
2084 	}
2085 
2086 	return -ETIMEDOUT;
2087 }
2088 
mlx5_pci_slot_reset(struct pci_dev * pdev)2089 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
2090 {
2091 	enum pci_ers_result res = PCI_ERS_RESULT_DISCONNECT;
2092 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2093 	int err;
2094 
2095 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Enter\n",
2096 		       __func__, dev->state, dev->pci_status);
2097 
2098 	err = mlx5_pci_enable_device(dev);
2099 	if (err) {
2100 		mlx5_core_err(dev, "%s: mlx5_pci_enable_device failed with error code: %d\n",
2101 			      __func__, err);
2102 		goto out;
2103 	}
2104 
2105 	pci_set_master(pdev);
2106 	pci_restore_state(pdev);
2107 	pci_save_state(pdev);
2108 
2109 	err = wait_vital(pdev);
2110 	if (err) {
2111 		mlx5_core_err(dev, "%s: wait vital failed with error code: %d\n",
2112 			      __func__, err);
2113 		goto out;
2114 	}
2115 
2116 	res = PCI_ERS_RESULT_RECOVERED;
2117 out:
2118 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, err = %d, result = %d, %s\n",
2119 		       __func__, dev->state, dev->pci_status, err, res, result2str(res));
2120 	return res;
2121 }
2122 
mlx5_pci_resume(struct pci_dev * pdev)2123 static void mlx5_pci_resume(struct pci_dev *pdev)
2124 {
2125 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2126 	int err;
2127 
2128 	mlx5_pci_trace(dev, "Enter, loading driver..\n");
2129 
2130 	err = mlx5_load_one(dev, false);
2131 
2132 	if (!err)
2133 		devlink_health_reporter_state_update(dev->priv.health.fw_fatal_reporter,
2134 						     DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
2135 
2136 	mlx5_pci_trace(dev, "Done, err = %d, device %s\n", err,
2137 		       !err ? "recovered" : "Failed");
2138 }
2139 
2140 static const struct pci_error_handlers mlx5_err_handler = {
2141 	.error_detected = mlx5_pci_err_detected,
2142 	.slot_reset	= mlx5_pci_slot_reset,
2143 	.resume		= mlx5_pci_resume
2144 };
2145 
mlx5_try_fast_unload(struct mlx5_core_dev * dev)2146 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
2147 {
2148 	bool fast_teardown = false, force_teardown = false;
2149 	int ret = 1;
2150 
2151 	fast_teardown = MLX5_CAP_GEN(dev, fast_teardown);
2152 	force_teardown = MLX5_CAP_GEN(dev, force_teardown);
2153 
2154 	mlx5_core_dbg(dev, "force teardown firmware support=%d\n", force_teardown);
2155 	mlx5_core_dbg(dev, "fast teardown firmware support=%d\n", fast_teardown);
2156 
2157 	if (!fast_teardown && !force_teardown)
2158 		return -EOPNOTSUPP;
2159 
2160 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
2161 		mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
2162 		return -EAGAIN;
2163 	}
2164 
2165 	/* Panic tear down fw command will stop the PCI bus communication
2166 	 * with the HCA, so the health poll is no longer needed.
2167 	 */
2168 	mlx5_stop_health_poll(dev, false);
2169 
2170 	ret = mlx5_cmd_fast_teardown_hca(dev);
2171 	if (!ret)
2172 		goto succeed;
2173 
2174 	ret = mlx5_cmd_force_teardown_hca(dev);
2175 	if (!ret)
2176 		goto succeed;
2177 
2178 	mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", ret);
2179 	mlx5_start_health_poll(dev);
2180 	return ret;
2181 
2182 succeed:
2183 	mlx5_enter_error_state(dev, true);
2184 
2185 	/* Some platforms requiring freeing the IRQ's in the shutdown
2186 	 * flow. If they aren't freed they can't be allocated after
2187 	 * kexec. There is no need to cleanup the mlx5_core software
2188 	 * contexts.
2189 	 */
2190 	mlx5_core_eq_free_irqs(dev);
2191 
2192 	return 0;
2193 }
2194 
shutdown(struct pci_dev * pdev)2195 static void shutdown(struct pci_dev *pdev)
2196 {
2197 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2198 	int err;
2199 
2200 	mlx5_core_info(dev, "Shutdown was called\n");
2201 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2202 	mlx5_drain_health_wq(dev);
2203 	err = mlx5_try_fast_unload(dev);
2204 	if (err)
2205 		mlx5_unload_one(dev, false);
2206 	mlx5_pci_disable_device(dev);
2207 }
2208 
mlx5_suspend(struct pci_dev * pdev,pm_message_t state)2209 static int mlx5_suspend(struct pci_dev *pdev, pm_message_t state)
2210 {
2211 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2212 
2213 	mlx5_unload_one(dev, true);
2214 
2215 	return 0;
2216 }
2217 
mlx5_resume(struct pci_dev * pdev)2218 static int mlx5_resume(struct pci_dev *pdev)
2219 {
2220 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2221 
2222 	return mlx5_load_one(dev, false);
2223 }
2224 
2225 static const struct pci_device_id mlx5_core_pci_table[] = {
2226 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTIB) },
2227 	{ PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF},	/* Connect-IB VF */
2228 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4) },
2229 	{ PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4 VF */
2230 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4_LX) },
2231 	{ PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4LX VF */
2232 	{ PCI_VDEVICE(MELLANOX, 0x1017) },			/* ConnectX-5, PCIe 3.0 */
2233 	{ PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 VF */
2234 	{ PCI_VDEVICE(MELLANOX, 0x1019) },			/* ConnectX-5 Ex */
2235 	{ PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 Ex VF */
2236 	{ PCI_VDEVICE(MELLANOX, 0x101b) },			/* ConnectX-6 */
2237 	{ PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF},	/* ConnectX-6 VF */
2238 	{ PCI_VDEVICE(MELLANOX, 0x101d) },			/* ConnectX-6 Dx */
2239 	{ PCI_VDEVICE(MELLANOX, 0x101e), MLX5_PCI_DEV_IS_VF},	/* ConnectX Family mlx5Gen Virtual Function */
2240 	{ PCI_VDEVICE(MELLANOX, 0x101f) },			/* ConnectX-6 LX */
2241 	{ PCI_VDEVICE(MELLANOX, 0x1021) },			/* ConnectX-7 */
2242 	{ PCI_VDEVICE(MELLANOX, 0x1023) },			/* ConnectX-8 */
2243 	{ PCI_VDEVICE(MELLANOX, 0x1025) },			/* ConnectX-9 */
2244 	{ PCI_VDEVICE(MELLANOX, 0xa2d2) },			/* BlueField integrated ConnectX-5 network controller */
2245 	{ PCI_VDEVICE(MELLANOX, 0xa2d3), MLX5_PCI_DEV_IS_VF},	/* BlueField integrated ConnectX-5 network controller VF */
2246 	{ PCI_VDEVICE(MELLANOX, 0xa2d6) },			/* BlueField-2 integrated ConnectX-6 Dx network controller */
2247 	{ PCI_VDEVICE(MELLANOX, 0xa2dc) },			/* BlueField-3 integrated ConnectX-7 network controller */
2248 	{ PCI_VDEVICE(MELLANOX, 0xa2df) },			/* BlueField-4 integrated ConnectX-8 network controller */
2249 	{ 0, }
2250 };
2251 
2252 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
2253 
mlx5_disable_device(struct mlx5_core_dev * dev)2254 void mlx5_disable_device(struct mlx5_core_dev *dev)
2255 {
2256 	mlx5_error_sw_reset(dev);
2257 	mlx5_unload_one_devl_locked(dev, false);
2258 }
2259 
mlx5_recover_device(struct mlx5_core_dev * dev)2260 int mlx5_recover_device(struct mlx5_core_dev *dev)
2261 {
2262 	if (!mlx5_core_is_sf(dev)) {
2263 		mlx5_pci_disable_device(dev);
2264 		if (mlx5_pci_slot_reset(dev->pdev) != PCI_ERS_RESULT_RECOVERED)
2265 			return -EIO;
2266 	}
2267 
2268 	return mlx5_load_one_devl_locked(dev, true);
2269 }
2270 
2271 static struct pci_driver mlx5_core_driver = {
2272 	.name           = KBUILD_MODNAME,
2273 	.id_table       = mlx5_core_pci_table,
2274 	.probe          = probe_one,
2275 	.remove         = remove_one,
2276 	.suspend        = mlx5_suspend,
2277 	.resume         = mlx5_resume,
2278 	.shutdown	= shutdown,
2279 	.err_handler	= &mlx5_err_handler,
2280 	.sriov_configure   = mlx5_core_sriov_configure,
2281 	.sriov_get_vf_total_msix = mlx5_sriov_get_vf_total_msix,
2282 	.sriov_set_msix_vec_count = mlx5_core_sriov_set_msix_vec_count,
2283 };
2284 
2285 /**
2286  * mlx5_vf_get_core_dev - Get the mlx5 core device from a given VF PCI device if
2287  *                     mlx5_core is its driver.
2288  * @pdev: The associated PCI device.
2289  *
2290  * Upon return the interface state lock stay held to let caller uses it safely.
2291  * Caller must ensure to use the returned mlx5 device for a narrow window
2292  * and put it back with mlx5_vf_put_core_dev() immediately once usage was over.
2293  *
2294  * Return: Pointer to the associated mlx5_core_dev or NULL.
2295  */
mlx5_vf_get_core_dev(struct pci_dev * pdev)2296 struct mlx5_core_dev *mlx5_vf_get_core_dev(struct pci_dev *pdev)
2297 {
2298 	struct mlx5_core_dev *mdev;
2299 
2300 	mdev = pci_iov_get_pf_drvdata(pdev, &mlx5_core_driver);
2301 	if (IS_ERR(mdev))
2302 		return NULL;
2303 
2304 	mutex_lock(&mdev->intf_state_mutex);
2305 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &mdev->intf_state)) {
2306 		mutex_unlock(&mdev->intf_state_mutex);
2307 		return NULL;
2308 	}
2309 
2310 	return mdev;
2311 }
2312 EXPORT_SYMBOL(mlx5_vf_get_core_dev);
2313 
2314 /**
2315  * mlx5_vf_put_core_dev - Put the mlx5 core device back.
2316  * @mdev: The mlx5 core device.
2317  *
2318  * Upon return the interface state lock is unlocked and caller should not
2319  * access the mdev any more.
2320  */
mlx5_vf_put_core_dev(struct mlx5_core_dev * mdev)2321 void mlx5_vf_put_core_dev(struct mlx5_core_dev *mdev)
2322 {
2323 	mutex_unlock(&mdev->intf_state_mutex);
2324 }
2325 EXPORT_SYMBOL(mlx5_vf_put_core_dev);
2326 
mlx5_core_verify_params(void)2327 static void mlx5_core_verify_params(void)
2328 {
2329 	if (prof_sel >= ARRAY_SIZE(profile)) {
2330 		pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
2331 			prof_sel,
2332 			ARRAY_SIZE(profile) - 1,
2333 			MLX5_DEFAULT_PROF);
2334 		prof_sel = MLX5_DEFAULT_PROF;
2335 	}
2336 }
2337 
mlx5_init(void)2338 static int __init mlx5_init(void)
2339 {
2340 	int err;
2341 
2342 	WARN_ONCE(strcmp(MLX5_ADEV_NAME, KBUILD_MODNAME),
2343 		  "mlx5_core name not in sync with kernel module name");
2344 
2345 	get_random_bytes(&sw_owner_id, sizeof(sw_owner_id));
2346 
2347 	mlx5_core_verify_params();
2348 	mlx5_register_debugfs();
2349 
2350 	err = mlx5e_init();
2351 	if (err)
2352 		goto err_debug;
2353 
2354 	err = mlx5_sf_driver_register();
2355 	if (err)
2356 		goto err_sf;
2357 
2358 	err = pci_register_driver(&mlx5_core_driver);
2359 	if (err)
2360 		goto err_pci;
2361 
2362 	return 0;
2363 
2364 err_pci:
2365 	mlx5_sf_driver_unregister();
2366 err_sf:
2367 	mlx5e_cleanup();
2368 err_debug:
2369 	mlx5_unregister_debugfs();
2370 	return err;
2371 }
2372 
mlx5_cleanup(void)2373 static void __exit mlx5_cleanup(void)
2374 {
2375 	pci_unregister_driver(&mlx5_core_driver);
2376 	mlx5_sf_driver_unregister();
2377 	mlx5e_cleanup();
2378 	mlx5_unregister_debugfs();
2379 }
2380 
2381 module_init(mlx5_init);
2382 module_exit(mlx5_cleanup);
2383