• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 	dev->tracer = mlx5_fw_tracer_create(dev);
1096 	dev->hv_vhca = mlx5_hv_vhca_create(dev);
1097 	dev->rsc_dump = mlx5_rsc_dump_create(dev);
1098 
1099 	return 0;
1100 
1101 err_fs:
1102 	mlx5_sf_table_cleanup(dev);
1103 err_sf_table_cleanup:
1104 	mlx5_sf_hw_table_cleanup(dev);
1105 err_sf_hw_table_cleanup:
1106 	mlx5_vhca_event_cleanup(dev);
1107 err_fpga_cleanup:
1108 	mlx5_fpga_cleanup(dev);
1109 err_eswitch_cleanup:
1110 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1111 err_sriov_cleanup:
1112 	mlx5_sriov_cleanup(dev);
1113 err_mpfs_cleanup:
1114 	mlx5_mpfs_cleanup(dev);
1115 err_rl_cleanup:
1116 	mlx5_cleanup_rl_table(dev);
1117 err_tables_cleanup:
1118 	mlx5_geneve_destroy(dev->geneve);
1119 	mlx5_vxlan_destroy(dev->vxlan);
1120 	mlx5_cleanup_clock(dev);
1121 	mlx5_cleanup_reserved_gids(dev);
1122 	mlx5_cq_debugfs_cleanup(dev);
1123 	mlx5_fw_reset_cleanup(dev);
1124 err_events_cleanup:
1125 	mlx5_events_cleanup(dev);
1126 err_eq_cleanup:
1127 	mlx5_eq_table_cleanup(dev);
1128 err_irq_cleanup:
1129 	mlx5_irq_table_cleanup(dev);
1130 err_devcom:
1131 	mlx5_unregister_hca_devcom_comp(dev);
1132 	mlx5_devcom_unregister_device(dev->priv.devc);
1133 
1134 	return err;
1135 }
1136 
mlx5_cleanup_once(struct mlx5_core_dev * dev)1137 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
1138 {
1139 	mlx5_rsc_dump_destroy(dev);
1140 	mlx5_hv_vhca_destroy(dev->hv_vhca);
1141 	mlx5_fw_tracer_destroy(dev->tracer);
1142 	mlx5_dm_cleanup(dev);
1143 	mlx5_fs_core_free(dev);
1144 	mlx5_sf_table_cleanup(dev);
1145 	mlx5_sf_hw_table_cleanup(dev);
1146 	mlx5_vhca_event_cleanup(dev);
1147 	mlx5_fpga_cleanup(dev);
1148 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1149 	mlx5_sriov_cleanup(dev);
1150 	mlx5_mpfs_cleanup(dev);
1151 	mlx5_cleanup_rl_table(dev);
1152 	mlx5_geneve_destroy(dev->geneve);
1153 	mlx5_vxlan_destroy(dev->vxlan);
1154 	mlx5_cleanup_clock(dev);
1155 	mlx5_cleanup_reserved_gids(dev);
1156 	mlx5_cq_debugfs_cleanup(dev);
1157 	mlx5_fw_reset_cleanup(dev);
1158 	mlx5_events_cleanup(dev);
1159 	mlx5_eq_table_cleanup(dev);
1160 	mlx5_irq_table_cleanup(dev);
1161 	mlx5_unregister_hca_devcom_comp(dev);
1162 	mlx5_devcom_unregister_device(dev->priv.devc);
1163 }
1164 
mlx5_function_enable(struct mlx5_core_dev * dev,bool boot,u64 timeout)1165 static int mlx5_function_enable(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1166 {
1167 	int err;
1168 
1169 	mlx5_core_info(dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1170 		       fw_rev_min(dev), fw_rev_sub(dev));
1171 
1172 	/* Only PFs hold the relevant PCIe information for this query */
1173 	if (mlx5_core_is_pf(dev))
1174 		pcie_print_link_status(dev->pdev);
1175 
1176 	/* wait for firmware to accept initialization segments configurations
1177 	 */
1178 	err = wait_fw_init(dev, timeout,
1179 			   mlx5_tout_ms(dev, FW_PRE_INIT_WARN_MESSAGE_INTERVAL),
1180 			   "pre-initializing");
1181 	if (err)
1182 		return err;
1183 
1184 	err = mlx5_cmd_enable(dev);
1185 	if (err) {
1186 		mlx5_core_err(dev, "Failed initializing command interface, aborting\n");
1187 		return err;
1188 	}
1189 
1190 	mlx5_tout_query_iseg(dev);
1191 
1192 	err = wait_fw_init(dev, mlx5_tout_ms(dev, FW_INIT), 0, "initializing");
1193 	if (err)
1194 		goto err_cmd_cleanup;
1195 
1196 	dev->caps.embedded_cpu = mlx5_read_embedded_cpu(dev);
1197 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_UP);
1198 
1199 	mlx5_start_health_poll(dev);
1200 
1201 	err = mlx5_core_enable_hca(dev, 0);
1202 	if (err) {
1203 		mlx5_core_err(dev, "enable hca failed\n");
1204 		goto stop_health_poll;
1205 	}
1206 
1207 	err = mlx5_core_set_issi(dev);
1208 	if (err) {
1209 		mlx5_core_err(dev, "failed to set issi\n");
1210 		goto err_disable_hca;
1211 	}
1212 
1213 	err = mlx5_satisfy_startup_pages(dev, 1);
1214 	if (err) {
1215 		mlx5_core_err(dev, "failed to allocate boot pages\n");
1216 		goto err_disable_hca;
1217 	}
1218 
1219 	err = mlx5_tout_query_dtor(dev);
1220 	if (err) {
1221 		mlx5_core_err(dev, "failed to read dtor\n");
1222 		goto reclaim_boot_pages;
1223 	}
1224 
1225 	return 0;
1226 
1227 reclaim_boot_pages:
1228 	mlx5_reclaim_startup_pages(dev);
1229 err_disable_hca:
1230 	mlx5_core_disable_hca(dev, 0);
1231 stop_health_poll:
1232 	mlx5_stop_health_poll(dev, boot);
1233 err_cmd_cleanup:
1234 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1235 	mlx5_cmd_disable(dev);
1236 
1237 	return err;
1238 }
1239 
mlx5_function_disable(struct mlx5_core_dev * dev,bool boot)1240 static void mlx5_function_disable(struct mlx5_core_dev *dev, bool boot)
1241 {
1242 	mlx5_reclaim_startup_pages(dev);
1243 	mlx5_core_disable_hca(dev, 0);
1244 	mlx5_stop_health_poll(dev, boot);
1245 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1246 	mlx5_cmd_disable(dev);
1247 }
1248 
mlx5_function_open(struct mlx5_core_dev * dev)1249 static int mlx5_function_open(struct mlx5_core_dev *dev)
1250 {
1251 	int err;
1252 
1253 	err = set_hca_ctrl(dev);
1254 	if (err) {
1255 		mlx5_core_err(dev, "set_hca_ctrl failed\n");
1256 		return err;
1257 	}
1258 
1259 	err = set_hca_cap(dev);
1260 	if (err) {
1261 		mlx5_core_err(dev, "set_hca_cap failed\n");
1262 		return err;
1263 	}
1264 
1265 	err = mlx5_satisfy_startup_pages(dev, 0);
1266 	if (err) {
1267 		mlx5_core_err(dev, "failed to allocate init pages\n");
1268 		return err;
1269 	}
1270 
1271 	err = mlx5_cmd_init_hca(dev, sw_owner_id);
1272 	if (err) {
1273 		mlx5_core_err(dev, "init hca failed\n");
1274 		return err;
1275 	}
1276 
1277 	mlx5_set_driver_version(dev);
1278 
1279 	err = mlx5_query_hca_caps(dev);
1280 	if (err) {
1281 		mlx5_core_err(dev, "query hca failed\n");
1282 		return err;
1283 	}
1284 	mlx5_start_health_fw_log_up(dev);
1285 	return 0;
1286 }
1287 
mlx5_function_close(struct mlx5_core_dev * dev)1288 static int mlx5_function_close(struct mlx5_core_dev *dev)
1289 {
1290 	int err;
1291 
1292 	err = mlx5_cmd_teardown_hca(dev);
1293 	if (err) {
1294 		mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n");
1295 		return err;
1296 	}
1297 
1298 	return 0;
1299 }
1300 
mlx5_function_setup(struct mlx5_core_dev * dev,bool boot,u64 timeout)1301 static int mlx5_function_setup(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1302 {
1303 	int err;
1304 
1305 	err = mlx5_function_enable(dev, boot, timeout);
1306 	if (err)
1307 		return err;
1308 
1309 	err = mlx5_function_open(dev);
1310 	if (err)
1311 		mlx5_function_disable(dev, boot);
1312 	return err;
1313 }
1314 
mlx5_function_teardown(struct mlx5_core_dev * dev,bool boot)1315 static int mlx5_function_teardown(struct mlx5_core_dev *dev, bool boot)
1316 {
1317 	int err = mlx5_function_close(dev);
1318 
1319 	if (!err)
1320 		mlx5_function_disable(dev, boot);
1321 	else
1322 		mlx5_stop_health_poll(dev, boot);
1323 
1324 	return err;
1325 }
1326 
mlx5_load(struct mlx5_core_dev * dev)1327 static int mlx5_load(struct mlx5_core_dev *dev)
1328 {
1329 	int err;
1330 
1331 	dev->priv.uar = mlx5_get_uars_page(dev);
1332 	if (IS_ERR(dev->priv.uar)) {
1333 		mlx5_core_err(dev, "Failed allocating uar, aborting\n");
1334 		err = PTR_ERR(dev->priv.uar);
1335 		return err;
1336 	}
1337 
1338 	mlx5_events_start(dev);
1339 	mlx5_pagealloc_start(dev);
1340 
1341 	err = mlx5_irq_table_create(dev);
1342 	if (err) {
1343 		mlx5_core_err(dev, "Failed to alloc IRQs\n");
1344 		goto err_irq_table;
1345 	}
1346 
1347 	err = mlx5_eq_table_create(dev);
1348 	if (err) {
1349 		mlx5_core_err(dev, "Failed to create EQs\n");
1350 		goto err_eq_table;
1351 	}
1352 
1353 	err = mlx5_fw_tracer_init(dev->tracer);
1354 	if (err) {
1355 		mlx5_core_err(dev, "Failed to init FW tracer %d\n", err);
1356 		mlx5_fw_tracer_destroy(dev->tracer);
1357 		dev->tracer = NULL;
1358 	}
1359 
1360 	mlx5_fw_reset_events_start(dev);
1361 	mlx5_hv_vhca_init(dev->hv_vhca);
1362 
1363 	err = mlx5_rsc_dump_init(dev);
1364 	if (err) {
1365 		mlx5_core_err(dev, "Failed to init Resource dump %d\n", err);
1366 		mlx5_rsc_dump_destroy(dev);
1367 		dev->rsc_dump = NULL;
1368 	}
1369 
1370 	err = mlx5_fpga_device_start(dev);
1371 	if (err) {
1372 		mlx5_core_err(dev, "fpga device start failed %d\n", err);
1373 		goto err_fpga_start;
1374 	}
1375 
1376 	err = mlx5_fs_core_init(dev);
1377 	if (err) {
1378 		mlx5_core_err(dev, "Failed to init flow steering\n");
1379 		goto err_fs;
1380 	}
1381 
1382 	err = mlx5_core_set_hca_defaults(dev);
1383 	if (err) {
1384 		mlx5_core_err(dev, "Failed to set hca defaults\n");
1385 		goto err_set_hca;
1386 	}
1387 
1388 	mlx5_vhca_event_start(dev);
1389 
1390 	err = mlx5_sf_hw_table_create(dev);
1391 	if (err) {
1392 		mlx5_core_err(dev, "sf table create failed %d\n", err);
1393 		goto err_vhca;
1394 	}
1395 
1396 	err = mlx5_ec_init(dev);
1397 	if (err) {
1398 		mlx5_core_err(dev, "Failed to init embedded CPU\n");
1399 		goto err_ec;
1400 	}
1401 
1402 	mlx5_lag_add_mdev(dev);
1403 	err = mlx5_sriov_attach(dev);
1404 	if (err) {
1405 		mlx5_core_err(dev, "sriov init failed %d\n", err);
1406 		goto err_sriov;
1407 	}
1408 
1409 	mlx5_sf_dev_table_create(dev);
1410 
1411 	err = mlx5_devlink_traps_register(priv_to_devlink(dev));
1412 	if (err)
1413 		goto err_traps_reg;
1414 
1415 	return 0;
1416 
1417 err_traps_reg:
1418 	mlx5_sf_dev_table_destroy(dev);
1419 	mlx5_sriov_detach(dev);
1420 err_sriov:
1421 	mlx5_lag_remove_mdev(dev);
1422 	mlx5_ec_cleanup(dev);
1423 err_ec:
1424 	mlx5_sf_hw_table_destroy(dev);
1425 err_vhca:
1426 	mlx5_vhca_event_stop(dev);
1427 err_set_hca:
1428 	mlx5_fs_core_cleanup(dev);
1429 err_fs:
1430 	mlx5_fpga_device_stop(dev);
1431 err_fpga_start:
1432 	mlx5_rsc_dump_cleanup(dev);
1433 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1434 	mlx5_fw_reset_events_stop(dev);
1435 	mlx5_fw_tracer_cleanup(dev->tracer);
1436 	mlx5_eq_table_destroy(dev);
1437 err_eq_table:
1438 	mlx5_irq_table_destroy(dev);
1439 err_irq_table:
1440 	mlx5_pagealloc_stop(dev);
1441 	mlx5_events_stop(dev);
1442 	mlx5_put_uars_page(dev, dev->priv.uar);
1443 	return err;
1444 }
1445 
mlx5_unload(struct mlx5_core_dev * dev)1446 static void mlx5_unload(struct mlx5_core_dev *dev)
1447 {
1448 	mlx5_eswitch_disable(dev->priv.eswitch);
1449 	mlx5_devlink_traps_unregister(priv_to_devlink(dev));
1450 	mlx5_sf_dev_table_destroy(dev);
1451 	mlx5_sriov_detach(dev);
1452 	mlx5_lag_remove_mdev(dev);
1453 	mlx5_ec_cleanup(dev);
1454 	mlx5_sf_hw_table_destroy(dev);
1455 	mlx5_vhca_event_stop(dev);
1456 	mlx5_fs_core_cleanup(dev);
1457 	mlx5_fpga_device_stop(dev);
1458 	mlx5_rsc_dump_cleanup(dev);
1459 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1460 	mlx5_fw_reset_events_stop(dev);
1461 	mlx5_fw_tracer_cleanup(dev->tracer);
1462 	mlx5_eq_table_destroy(dev);
1463 	mlx5_irq_table_destroy(dev);
1464 	mlx5_pagealloc_stop(dev);
1465 	mlx5_events_stop(dev);
1466 	mlx5_put_uars_page(dev, dev->priv.uar);
1467 }
1468 
mlx5_init_one_devl_locked(struct mlx5_core_dev * dev)1469 int mlx5_init_one_devl_locked(struct mlx5_core_dev *dev)
1470 {
1471 	bool light_probe = mlx5_dev_is_lightweight(dev);
1472 	int err = 0;
1473 
1474 	mutex_lock(&dev->intf_state_mutex);
1475 	dev->state = MLX5_DEVICE_STATE_UP;
1476 
1477 	err = mlx5_function_setup(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1478 	if (err)
1479 		goto err_function;
1480 
1481 	err = mlx5_init_once(dev);
1482 	if (err) {
1483 		mlx5_core_err(dev, "sw objs init failed\n");
1484 		goto function_teardown;
1485 	}
1486 
1487 	/* In case of light_probe, mlx5_devlink is already registered.
1488 	 * Hence, don't register devlink again.
1489 	 */
1490 	if (!light_probe) {
1491 		err = mlx5_devlink_params_register(priv_to_devlink(dev));
1492 		if (err)
1493 			goto err_devlink_params_reg;
1494 	}
1495 
1496 	err = mlx5_load(dev);
1497 	if (err)
1498 		goto err_load;
1499 
1500 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1501 
1502 	err = mlx5_register_device(dev);
1503 	if (err)
1504 		goto err_register;
1505 
1506 	err = mlx5_crdump_enable(dev);
1507 	if (err)
1508 		mlx5_core_err(dev, "mlx5_crdump_enable failed with error code %d\n", err);
1509 
1510 	err = mlx5_hwmon_dev_register(dev);
1511 	if (err)
1512 		mlx5_core_err(dev, "mlx5_hwmon_dev_register failed with error code %d\n", err);
1513 
1514 	mutex_unlock(&dev->intf_state_mutex);
1515 	return 0;
1516 
1517 err_register:
1518 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1519 	mlx5_unload(dev);
1520 err_load:
1521 	if (!light_probe)
1522 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1523 err_devlink_params_reg:
1524 	mlx5_cleanup_once(dev);
1525 function_teardown:
1526 	mlx5_function_teardown(dev, true);
1527 err_function:
1528 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1529 	mutex_unlock(&dev->intf_state_mutex);
1530 	return err;
1531 }
1532 
mlx5_init_one(struct mlx5_core_dev * dev)1533 int mlx5_init_one(struct mlx5_core_dev *dev)
1534 {
1535 	struct devlink *devlink = priv_to_devlink(dev);
1536 	int err;
1537 
1538 	devl_lock(devlink);
1539 	devl_register(devlink);
1540 	err = mlx5_init_one_devl_locked(dev);
1541 	if (err)
1542 		devl_unregister(devlink);
1543 	devl_unlock(devlink);
1544 	return err;
1545 }
1546 
mlx5_uninit_one(struct mlx5_core_dev * dev)1547 void mlx5_uninit_one(struct mlx5_core_dev *dev)
1548 {
1549 	struct devlink *devlink = priv_to_devlink(dev);
1550 
1551 	devl_lock(devlink);
1552 	mutex_lock(&dev->intf_state_mutex);
1553 
1554 	mlx5_hwmon_dev_unregister(dev);
1555 	mlx5_crdump_disable(dev);
1556 	mlx5_unregister_device(dev);
1557 
1558 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1559 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1560 			       __func__);
1561 		mlx5_devlink_params_unregister(priv_to_devlink(dev));
1562 		mlx5_cleanup_once(dev);
1563 		goto out;
1564 	}
1565 
1566 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1567 	mlx5_unload(dev);
1568 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1569 	mlx5_cleanup_once(dev);
1570 	mlx5_function_teardown(dev, true);
1571 out:
1572 	mutex_unlock(&dev->intf_state_mutex);
1573 	devl_unregister(devlink);
1574 	devl_unlock(devlink);
1575 }
1576 
mlx5_load_one_devl_locked(struct mlx5_core_dev * dev,bool recovery)1577 int mlx5_load_one_devl_locked(struct mlx5_core_dev *dev, bool recovery)
1578 {
1579 	int err = 0;
1580 	u64 timeout;
1581 
1582 	devl_assert_locked(priv_to_devlink(dev));
1583 	mutex_lock(&dev->intf_state_mutex);
1584 	if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1585 		mlx5_core_warn(dev, "interface is up, NOP\n");
1586 		goto out;
1587 	}
1588 	/* remove any previous indication of internal error */
1589 	dev->state = MLX5_DEVICE_STATE_UP;
1590 
1591 	if (recovery)
1592 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_ON_RECOVERY_TIMEOUT);
1593 	else
1594 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT);
1595 	err = mlx5_function_setup(dev, false, timeout);
1596 	if (err)
1597 		goto err_function;
1598 
1599 	err = mlx5_load(dev);
1600 	if (err)
1601 		goto err_load;
1602 
1603 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1604 
1605 	err = mlx5_attach_device(dev);
1606 	if (err)
1607 		goto err_attach;
1608 
1609 	mutex_unlock(&dev->intf_state_mutex);
1610 	return 0;
1611 
1612 err_attach:
1613 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1614 	mlx5_unload(dev);
1615 err_load:
1616 	mlx5_function_teardown(dev, false);
1617 err_function:
1618 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1619 out:
1620 	mutex_unlock(&dev->intf_state_mutex);
1621 	return err;
1622 }
1623 
mlx5_load_one(struct mlx5_core_dev * dev,bool recovery)1624 int mlx5_load_one(struct mlx5_core_dev *dev, bool recovery)
1625 {
1626 	struct devlink *devlink = priv_to_devlink(dev);
1627 	int ret;
1628 
1629 	devl_lock(devlink);
1630 	ret = mlx5_load_one_devl_locked(dev, recovery);
1631 	devl_unlock(devlink);
1632 	return ret;
1633 }
1634 
mlx5_unload_one_devl_locked(struct mlx5_core_dev * dev,bool suspend)1635 void mlx5_unload_one_devl_locked(struct mlx5_core_dev *dev, bool suspend)
1636 {
1637 	devl_assert_locked(priv_to_devlink(dev));
1638 	mutex_lock(&dev->intf_state_mutex);
1639 
1640 	mlx5_detach_device(dev, suspend);
1641 
1642 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1643 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1644 			       __func__);
1645 		goto out;
1646 	}
1647 
1648 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1649 	mlx5_unload(dev);
1650 	mlx5_function_teardown(dev, false);
1651 out:
1652 	mutex_unlock(&dev->intf_state_mutex);
1653 }
1654 
mlx5_unload_one(struct mlx5_core_dev * dev,bool suspend)1655 void mlx5_unload_one(struct mlx5_core_dev *dev, bool suspend)
1656 {
1657 	struct devlink *devlink = priv_to_devlink(dev);
1658 
1659 	devl_lock(devlink);
1660 	mlx5_unload_one_devl_locked(dev, suspend);
1661 	devl_unlock(devlink);
1662 }
1663 
1664 /* In case of light probe, we don't need a full query of hca_caps, but only the bellow caps.
1665  * A full query of hca_caps will be done when the device will reload.
1666  */
mlx5_query_hca_caps_light(struct mlx5_core_dev * dev)1667 static int mlx5_query_hca_caps_light(struct mlx5_core_dev *dev)
1668 {
1669 	int err;
1670 
1671 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
1672 	if (err)
1673 		return err;
1674 
1675 	if (MLX5_CAP_GEN(dev, eth_net_offloads)) {
1676 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_ETHERNET_OFFLOADS,
1677 					      HCA_CAP_OPMOD_GET_CUR);
1678 		if (err)
1679 			return err;
1680 	}
1681 
1682 	if (MLX5_CAP_GEN(dev, nic_flow_table) ||
1683 	    MLX5_CAP_GEN(dev, ipoib_enhanced_offloads)) {
1684 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_FLOW_TABLE,
1685 					      HCA_CAP_OPMOD_GET_CUR);
1686 		if (err)
1687 			return err;
1688 	}
1689 
1690 	if (MLX5_CAP_GEN_64(dev, general_obj_types) &
1691 		MLX5_GENERAL_OBJ_TYPES_CAP_VIRTIO_NET_Q) {
1692 		err = mlx5_core_get_caps_mode(dev, MLX5_CAP_VDPA_EMULATION,
1693 					      HCA_CAP_OPMOD_GET_CUR);
1694 		if (err)
1695 			return err;
1696 	}
1697 
1698 	return 0;
1699 }
1700 
mlx5_init_one_light(struct mlx5_core_dev * dev)1701 int mlx5_init_one_light(struct mlx5_core_dev *dev)
1702 {
1703 	struct devlink *devlink = priv_to_devlink(dev);
1704 	int err;
1705 
1706 	devl_lock(devlink);
1707 	devl_register(devlink);
1708 	dev->state = MLX5_DEVICE_STATE_UP;
1709 	err = mlx5_function_enable(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1710 	if (err) {
1711 		mlx5_core_warn(dev, "mlx5_function_enable err=%d\n", err);
1712 		goto out;
1713 	}
1714 
1715 	err = mlx5_query_hca_caps_light(dev);
1716 	if (err) {
1717 		mlx5_core_warn(dev, "mlx5_query_hca_caps_light err=%d\n", err);
1718 		goto query_hca_caps_err;
1719 	}
1720 
1721 	err = mlx5_devlink_params_register(priv_to_devlink(dev));
1722 	if (err) {
1723 		mlx5_core_warn(dev, "mlx5_devlink_param_reg err = %d\n", err);
1724 		goto query_hca_caps_err;
1725 	}
1726 
1727 	devl_unlock(devlink);
1728 	return 0;
1729 
1730 query_hca_caps_err:
1731 	mlx5_function_disable(dev, true);
1732 out:
1733 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1734 	devl_unregister(devlink);
1735 	devl_unlock(devlink);
1736 	return err;
1737 }
1738 
mlx5_uninit_one_light(struct mlx5_core_dev * dev)1739 void mlx5_uninit_one_light(struct mlx5_core_dev *dev)
1740 {
1741 	struct devlink *devlink = priv_to_devlink(dev);
1742 
1743 	devl_lock(devlink);
1744 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1745 	devl_unregister(devlink);
1746 	devl_unlock(devlink);
1747 	if (dev->state != MLX5_DEVICE_STATE_UP)
1748 		return;
1749 	mlx5_function_disable(dev, true);
1750 }
1751 
1752 /* xxx_light() function are used in order to configure the device without full
1753  * init (light init). e.g.: There isn't a point in reload a device to light state.
1754  * Hence, mlx5_load_one_light() isn't needed.
1755  */
1756 
mlx5_unload_one_light(struct mlx5_core_dev * dev)1757 void mlx5_unload_one_light(struct mlx5_core_dev *dev)
1758 {
1759 	if (dev->state != MLX5_DEVICE_STATE_UP)
1760 		return;
1761 	mlx5_function_disable(dev, false);
1762 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1763 }
1764 
1765 static const int types[] = {
1766 	MLX5_CAP_GENERAL,
1767 	MLX5_CAP_GENERAL_2,
1768 	MLX5_CAP_ETHERNET_OFFLOADS,
1769 	MLX5_CAP_IPOIB_ENHANCED_OFFLOADS,
1770 	MLX5_CAP_ODP,
1771 	MLX5_CAP_ATOMIC,
1772 	MLX5_CAP_ROCE,
1773 	MLX5_CAP_IPOIB_OFFLOADS,
1774 	MLX5_CAP_FLOW_TABLE,
1775 	MLX5_CAP_ESWITCH_FLOW_TABLE,
1776 	MLX5_CAP_ESWITCH,
1777 	MLX5_CAP_QOS,
1778 	MLX5_CAP_DEBUG,
1779 	MLX5_CAP_DEV_MEM,
1780 	MLX5_CAP_DEV_EVENT,
1781 	MLX5_CAP_TLS,
1782 	MLX5_CAP_VDPA_EMULATION,
1783 	MLX5_CAP_IPSEC,
1784 	MLX5_CAP_PORT_SELECTION,
1785 	MLX5_CAP_MACSEC,
1786 	MLX5_CAP_ADV_VIRTUALIZATION,
1787 	MLX5_CAP_CRYPTO,
1788 };
1789 
mlx5_hca_caps_free(struct mlx5_core_dev * dev)1790 static void mlx5_hca_caps_free(struct mlx5_core_dev *dev)
1791 {
1792 	int type;
1793 	int i;
1794 
1795 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1796 		type = types[i];
1797 		kfree(dev->caps.hca[type]);
1798 	}
1799 }
1800 
mlx5_hca_caps_alloc(struct mlx5_core_dev * dev)1801 static int mlx5_hca_caps_alloc(struct mlx5_core_dev *dev)
1802 {
1803 	struct mlx5_hca_cap *cap;
1804 	int type;
1805 	int i;
1806 
1807 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1808 		cap = kzalloc(sizeof(*cap), GFP_KERNEL);
1809 		if (!cap)
1810 			goto err;
1811 		type = types[i];
1812 		dev->caps.hca[type] = cap;
1813 	}
1814 
1815 	return 0;
1816 
1817 err:
1818 	mlx5_hca_caps_free(dev);
1819 	return -ENOMEM;
1820 }
1821 
vhca_id_show(struct seq_file * file,void * priv)1822 static int vhca_id_show(struct seq_file *file, void *priv)
1823 {
1824 	struct mlx5_core_dev *dev = file->private;
1825 
1826 	seq_printf(file, "0x%x\n", MLX5_CAP_GEN(dev, vhca_id));
1827 	return 0;
1828 }
1829 
1830 DEFINE_SHOW_ATTRIBUTE(vhca_id);
1831 
mlx5_mdev_init(struct mlx5_core_dev * dev,int profile_idx)1832 int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
1833 {
1834 	struct mlx5_priv *priv = &dev->priv;
1835 	int err;
1836 
1837 	memcpy(&dev->profile, &profile[profile_idx], sizeof(dev->profile));
1838 	lockdep_register_key(&dev->lock_key);
1839 	mutex_init(&dev->intf_state_mutex);
1840 	lockdep_set_class(&dev->intf_state_mutex, &dev->lock_key);
1841 	mutex_init(&dev->mlx5e_res.uplink_netdev_lock);
1842 	mutex_init(&dev->wc_state_lock);
1843 
1844 	mutex_init(&priv->bfregs.reg_head.lock);
1845 	mutex_init(&priv->bfregs.wc_head.lock);
1846 	INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1847 	INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1848 
1849 	mutex_init(&priv->alloc_mutex);
1850 	mutex_init(&priv->pgdir_mutex);
1851 	INIT_LIST_HEAD(&priv->pgdir_list);
1852 
1853 	priv->numa_node = dev_to_node(mlx5_core_dma_dev(dev));
1854 	priv->dbg.dbg_root = debugfs_create_dir(dev_name(dev->device),
1855 						mlx5_debugfs_root);
1856 	debugfs_create_file("vhca_id", 0400, priv->dbg.dbg_root, dev, &vhca_id_fops);
1857 	INIT_LIST_HEAD(&priv->traps);
1858 
1859 	err = mlx5_cmd_init(dev);
1860 	if (err) {
1861 		mlx5_core_err(dev, "Failed initializing cmdif SW structs, aborting\n");
1862 		goto err_cmd_init;
1863 	}
1864 
1865 	err = mlx5_tout_init(dev);
1866 	if (err) {
1867 		mlx5_core_err(dev, "Failed initializing timeouts, aborting\n");
1868 		goto err_timeout_init;
1869 	}
1870 
1871 	err = mlx5_health_init(dev);
1872 	if (err)
1873 		goto err_health_init;
1874 
1875 	err = mlx5_pagealloc_init(dev);
1876 	if (err)
1877 		goto err_pagealloc_init;
1878 
1879 	err = mlx5_adev_init(dev);
1880 	if (err)
1881 		goto err_adev_init;
1882 
1883 	err = mlx5_hca_caps_alloc(dev);
1884 	if (err)
1885 		goto err_hca_caps;
1886 
1887 	/* The conjunction of sw_vhca_id with sw_owner_id will be a global
1888 	 * unique id per function which uses mlx5_core.
1889 	 * Those values are supplied to FW as part of the init HCA command to
1890 	 * be used by both driver and FW when it's applicable.
1891 	 */
1892 	dev->priv.sw_vhca_id = ida_alloc_range(&sw_vhca_ida, 1,
1893 					       MAX_SW_VHCA_ID,
1894 					       GFP_KERNEL);
1895 	if (dev->priv.sw_vhca_id < 0)
1896 		mlx5_core_err(dev, "failed to allocate sw_vhca_id, err=%d\n",
1897 			      dev->priv.sw_vhca_id);
1898 
1899 	return 0;
1900 
1901 err_hca_caps:
1902 	mlx5_adev_cleanup(dev);
1903 err_adev_init:
1904 	mlx5_pagealloc_cleanup(dev);
1905 err_pagealloc_init:
1906 	mlx5_health_cleanup(dev);
1907 err_health_init:
1908 	mlx5_tout_cleanup(dev);
1909 err_timeout_init:
1910 	mlx5_cmd_cleanup(dev);
1911 err_cmd_init:
1912 	debugfs_remove(dev->priv.dbg.dbg_root);
1913 	mutex_destroy(&priv->pgdir_mutex);
1914 	mutex_destroy(&priv->alloc_mutex);
1915 	mutex_destroy(&priv->bfregs.wc_head.lock);
1916 	mutex_destroy(&priv->bfregs.reg_head.lock);
1917 	mutex_destroy(&dev->intf_state_mutex);
1918 	lockdep_unregister_key(&dev->lock_key);
1919 	return err;
1920 }
1921 
mlx5_mdev_uninit(struct mlx5_core_dev * dev)1922 void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
1923 {
1924 	struct mlx5_priv *priv = &dev->priv;
1925 
1926 	if (priv->sw_vhca_id > 0)
1927 		ida_free(&sw_vhca_ida, dev->priv.sw_vhca_id);
1928 
1929 	mlx5_hca_caps_free(dev);
1930 	mlx5_adev_cleanup(dev);
1931 	mlx5_pagealloc_cleanup(dev);
1932 	mlx5_health_cleanup(dev);
1933 	mlx5_tout_cleanup(dev);
1934 	mlx5_cmd_cleanup(dev);
1935 	debugfs_remove_recursive(dev->priv.dbg.dbg_root);
1936 	mutex_destroy(&priv->pgdir_mutex);
1937 	mutex_destroy(&priv->alloc_mutex);
1938 	mutex_destroy(&priv->bfregs.wc_head.lock);
1939 	mutex_destroy(&priv->bfregs.reg_head.lock);
1940 	mutex_destroy(&dev->wc_state_lock);
1941 	mutex_destroy(&dev->mlx5e_res.uplink_netdev_lock);
1942 	mutex_destroy(&dev->intf_state_mutex);
1943 	lockdep_unregister_key(&dev->lock_key);
1944 }
1945 
probe_one(struct pci_dev * pdev,const struct pci_device_id * id)1946 static int probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
1947 {
1948 	struct mlx5_core_dev *dev;
1949 	struct devlink *devlink;
1950 	int err;
1951 
1952 	devlink = mlx5_devlink_alloc(&pdev->dev);
1953 	if (!devlink) {
1954 		dev_err(&pdev->dev, "devlink alloc failed\n");
1955 		return -ENOMEM;
1956 	}
1957 
1958 	dev = devlink_priv(devlink);
1959 	dev->device = &pdev->dev;
1960 	dev->pdev = pdev;
1961 
1962 	dev->coredev_type = id->driver_data & MLX5_PCI_DEV_IS_VF ?
1963 			 MLX5_COREDEV_VF : MLX5_COREDEV_PF;
1964 
1965 	dev->priv.adev_idx = mlx5_adev_idx_alloc();
1966 	if (dev->priv.adev_idx < 0) {
1967 		err = dev->priv.adev_idx;
1968 		goto adev_init_err;
1969 	}
1970 
1971 	err = mlx5_mdev_init(dev, prof_sel);
1972 	if (err)
1973 		goto mdev_init_err;
1974 
1975 	err = mlx5_pci_init(dev, pdev, id);
1976 	if (err) {
1977 		mlx5_core_err(dev, "mlx5_pci_init failed with error code %d\n",
1978 			      err);
1979 		goto pci_init_err;
1980 	}
1981 
1982 	err = mlx5_init_one(dev);
1983 	if (err) {
1984 		mlx5_core_err(dev, "mlx5_init_one failed with error code %d\n",
1985 			      err);
1986 		goto err_init_one;
1987 	}
1988 
1989 	pci_save_state(pdev);
1990 	return 0;
1991 
1992 err_init_one:
1993 	mlx5_pci_close(dev);
1994 pci_init_err:
1995 	mlx5_mdev_uninit(dev);
1996 mdev_init_err:
1997 	mlx5_adev_idx_free(dev->priv.adev_idx);
1998 adev_init_err:
1999 	mlx5_devlink_free(devlink);
2000 
2001 	return err;
2002 }
2003 
remove_one(struct pci_dev * pdev)2004 static void remove_one(struct pci_dev *pdev)
2005 {
2006 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2007 	struct devlink *devlink = priv_to_devlink(dev);
2008 
2009 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2010 	mlx5_drain_fw_reset(dev);
2011 	mlx5_drain_health_wq(dev);
2012 	mlx5_sriov_disable(pdev, false);
2013 	mlx5_uninit_one(dev);
2014 	mlx5_pci_close(dev);
2015 	mlx5_mdev_uninit(dev);
2016 	mlx5_adev_idx_free(dev->priv.adev_idx);
2017 	mlx5_devlink_free(devlink);
2018 }
2019 
2020 #define mlx5_pci_trace(dev, fmt, ...) ({ \
2021 	struct mlx5_core_dev *__dev = (dev); \
2022 	mlx5_core_info(__dev, "%s Device state = %d health sensors: %d pci_status: %d. " fmt, \
2023 		       __func__, __dev->state, mlx5_health_check_fatal_sensors(__dev), \
2024 		       __dev->pci_status, ##__VA_ARGS__); \
2025 })
2026 
result2str(enum pci_ers_result result)2027 static const char *result2str(enum pci_ers_result result)
2028 {
2029 	return  result == PCI_ERS_RESULT_NEED_RESET ? "need reset" :
2030 		result == PCI_ERS_RESULT_DISCONNECT ? "disconnect" :
2031 		result == PCI_ERS_RESULT_RECOVERED  ? "recovered" :
2032 		"unknown";
2033 }
2034 
mlx5_pci_err_detected(struct pci_dev * pdev,pci_channel_state_t state)2035 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
2036 					      pci_channel_state_t state)
2037 {
2038 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2039 	enum pci_ers_result res;
2040 
2041 	mlx5_pci_trace(dev, "Enter, pci channel state = %d\n", state);
2042 
2043 	mlx5_enter_error_state(dev, false);
2044 	mlx5_error_sw_reset(dev);
2045 	mlx5_unload_one(dev, false);
2046 	mlx5_drain_health_wq(dev);
2047 	mlx5_pci_disable_device(dev);
2048 
2049 	res = state == pci_channel_io_perm_failure ?
2050 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
2051 
2052 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, result = %d, %s\n",
2053 		       __func__, dev->state, dev->pci_status, res, result2str(res));
2054 	return res;
2055 }
2056 
2057 /* wait for the device to show vital signs by waiting
2058  * for the health counter to start counting.
2059  */
wait_vital(struct pci_dev * pdev)2060 static int wait_vital(struct pci_dev *pdev)
2061 {
2062 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2063 	struct mlx5_core_health *health = &dev->priv.health;
2064 	const int niter = 100;
2065 	u32 last_count = 0;
2066 	u32 count;
2067 	int i;
2068 
2069 	for (i = 0; i < niter; i++) {
2070 		count = ioread32be(health->health_counter);
2071 		if (count && count != 0xffffffff) {
2072 			if (last_count && last_count != count) {
2073 				mlx5_core_info(dev,
2074 					       "wait vital counter value 0x%x after %d iterations\n",
2075 					       count, i);
2076 				return 0;
2077 			}
2078 			last_count = count;
2079 		}
2080 		msleep(50);
2081 	}
2082 
2083 	return -ETIMEDOUT;
2084 }
2085 
mlx5_pci_slot_reset(struct pci_dev * pdev)2086 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
2087 {
2088 	enum pci_ers_result res = PCI_ERS_RESULT_DISCONNECT;
2089 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2090 	int err;
2091 
2092 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Enter\n",
2093 		       __func__, dev->state, dev->pci_status);
2094 
2095 	err = mlx5_pci_enable_device(dev);
2096 	if (err) {
2097 		mlx5_core_err(dev, "%s: mlx5_pci_enable_device failed with error code: %d\n",
2098 			      __func__, err);
2099 		goto out;
2100 	}
2101 
2102 	pci_set_master(pdev);
2103 	pci_restore_state(pdev);
2104 	pci_save_state(pdev);
2105 
2106 	err = wait_vital(pdev);
2107 	if (err) {
2108 		mlx5_core_err(dev, "%s: wait vital failed with error code: %d\n",
2109 			      __func__, err);
2110 		goto out;
2111 	}
2112 
2113 	res = PCI_ERS_RESULT_RECOVERED;
2114 out:
2115 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, err = %d, result = %d, %s\n",
2116 		       __func__, dev->state, dev->pci_status, err, res, result2str(res));
2117 	return res;
2118 }
2119 
mlx5_pci_resume(struct pci_dev * pdev)2120 static void mlx5_pci_resume(struct pci_dev *pdev)
2121 {
2122 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2123 	int err;
2124 
2125 	mlx5_pci_trace(dev, "Enter, loading driver..\n");
2126 
2127 	err = mlx5_load_one(dev, false);
2128 
2129 	if (!err)
2130 		devlink_health_reporter_state_update(dev->priv.health.fw_fatal_reporter,
2131 						     DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
2132 
2133 	mlx5_pci_trace(dev, "Done, err = %d, device %s\n", err,
2134 		       !err ? "recovered" : "Failed");
2135 }
2136 
2137 static const struct pci_error_handlers mlx5_err_handler = {
2138 	.error_detected = mlx5_pci_err_detected,
2139 	.slot_reset	= mlx5_pci_slot_reset,
2140 	.resume		= mlx5_pci_resume
2141 };
2142 
mlx5_try_fast_unload(struct mlx5_core_dev * dev)2143 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
2144 {
2145 	bool fast_teardown = false, force_teardown = false;
2146 	int ret = 1;
2147 
2148 	fast_teardown = MLX5_CAP_GEN(dev, fast_teardown);
2149 	force_teardown = MLX5_CAP_GEN(dev, force_teardown);
2150 
2151 	mlx5_core_dbg(dev, "force teardown firmware support=%d\n", force_teardown);
2152 	mlx5_core_dbg(dev, "fast teardown firmware support=%d\n", fast_teardown);
2153 
2154 	if (!fast_teardown && !force_teardown)
2155 		return -EOPNOTSUPP;
2156 
2157 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
2158 		mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
2159 		return -EAGAIN;
2160 	}
2161 
2162 	/* Panic tear down fw command will stop the PCI bus communication
2163 	 * with the HCA, so the health poll is no longer needed.
2164 	 */
2165 	mlx5_stop_health_poll(dev, false);
2166 
2167 	ret = mlx5_cmd_fast_teardown_hca(dev);
2168 	if (!ret)
2169 		goto succeed;
2170 
2171 	ret = mlx5_cmd_force_teardown_hca(dev);
2172 	if (!ret)
2173 		goto succeed;
2174 
2175 	mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", ret);
2176 	mlx5_start_health_poll(dev);
2177 	return ret;
2178 
2179 succeed:
2180 	mlx5_enter_error_state(dev, true);
2181 
2182 	/* Some platforms requiring freeing the IRQ's in the shutdown
2183 	 * flow. If they aren't freed they can't be allocated after
2184 	 * kexec. There is no need to cleanup the mlx5_core software
2185 	 * contexts.
2186 	 */
2187 	mlx5_core_eq_free_irqs(dev);
2188 
2189 	return 0;
2190 }
2191 
shutdown(struct pci_dev * pdev)2192 static void shutdown(struct pci_dev *pdev)
2193 {
2194 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
2195 	int err;
2196 
2197 	mlx5_core_info(dev, "Shutdown was called\n");
2198 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
2199 	mlx5_drain_health_wq(dev);
2200 	err = mlx5_try_fast_unload(dev);
2201 	if (err)
2202 		mlx5_unload_one(dev, false);
2203 	mlx5_pci_disable_device(dev);
2204 }
2205 
mlx5_suspend(struct pci_dev * pdev,pm_message_t state)2206 static int mlx5_suspend(struct pci_dev *pdev, pm_message_t state)
2207 {
2208 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2209 
2210 	mlx5_unload_one(dev, true);
2211 
2212 	return 0;
2213 }
2214 
mlx5_resume(struct pci_dev * pdev)2215 static int mlx5_resume(struct pci_dev *pdev)
2216 {
2217 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2218 
2219 	return mlx5_load_one(dev, false);
2220 }
2221 
2222 static const struct pci_device_id mlx5_core_pci_table[] = {
2223 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTIB) },
2224 	{ PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF},	/* Connect-IB VF */
2225 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4) },
2226 	{ PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4 VF */
2227 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4_LX) },
2228 	{ PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4LX VF */
2229 	{ PCI_VDEVICE(MELLANOX, 0x1017) },			/* ConnectX-5, PCIe 3.0 */
2230 	{ PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 VF */
2231 	{ PCI_VDEVICE(MELLANOX, 0x1019) },			/* ConnectX-5 Ex */
2232 	{ PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 Ex VF */
2233 	{ PCI_VDEVICE(MELLANOX, 0x101b) },			/* ConnectX-6 */
2234 	{ PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF},	/* ConnectX-6 VF */
2235 	{ PCI_VDEVICE(MELLANOX, 0x101d) },			/* ConnectX-6 Dx */
2236 	{ PCI_VDEVICE(MELLANOX, 0x101e), MLX5_PCI_DEV_IS_VF},	/* ConnectX Family mlx5Gen Virtual Function */
2237 	{ PCI_VDEVICE(MELLANOX, 0x101f) },			/* ConnectX-6 LX */
2238 	{ PCI_VDEVICE(MELLANOX, 0x1021) },			/* ConnectX-7 */
2239 	{ PCI_VDEVICE(MELLANOX, 0x1023) },			/* ConnectX-8 */
2240 	{ PCI_VDEVICE(MELLANOX, 0x1025) },			/* ConnectX-9 */
2241 	{ PCI_VDEVICE(MELLANOX, 0x1027) },			/* ConnectX-10 */
2242 	{ PCI_VDEVICE(MELLANOX, 0xa2d2) },			/* BlueField integrated ConnectX-5 network controller */
2243 	{ PCI_VDEVICE(MELLANOX, 0xa2d3), MLX5_PCI_DEV_IS_VF},	/* BlueField integrated ConnectX-5 network controller VF */
2244 	{ PCI_VDEVICE(MELLANOX, 0xa2d6) },			/* BlueField-2 integrated ConnectX-6 Dx network controller */
2245 	{ PCI_VDEVICE(MELLANOX, 0xa2dc) },			/* BlueField-3 integrated ConnectX-7 network controller */
2246 	{ PCI_VDEVICE(MELLANOX, 0xa2df) },			/* BlueField-4 integrated ConnectX-8 network controller */
2247 	{ 0, }
2248 };
2249 
2250 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
2251 
mlx5_disable_device(struct mlx5_core_dev * dev)2252 void mlx5_disable_device(struct mlx5_core_dev *dev)
2253 {
2254 	mlx5_error_sw_reset(dev);
2255 	mlx5_unload_one_devl_locked(dev, false);
2256 }
2257 
mlx5_recover_device(struct mlx5_core_dev * dev)2258 int mlx5_recover_device(struct mlx5_core_dev *dev)
2259 {
2260 	if (!mlx5_core_is_sf(dev)) {
2261 		mlx5_pci_disable_device(dev);
2262 		if (mlx5_pci_slot_reset(dev->pdev) != PCI_ERS_RESULT_RECOVERED)
2263 			return -EIO;
2264 	}
2265 
2266 	return mlx5_load_one_devl_locked(dev, true);
2267 }
2268 
2269 static struct pci_driver mlx5_core_driver = {
2270 	.name           = KBUILD_MODNAME,
2271 	.id_table       = mlx5_core_pci_table,
2272 	.probe          = probe_one,
2273 	.remove         = remove_one,
2274 	.suspend        = mlx5_suspend,
2275 	.resume         = mlx5_resume,
2276 	.shutdown	= shutdown,
2277 	.err_handler	= &mlx5_err_handler,
2278 	.sriov_configure   = mlx5_core_sriov_configure,
2279 	.sriov_get_vf_total_msix = mlx5_sriov_get_vf_total_msix,
2280 	.sriov_set_msix_vec_count = mlx5_core_sriov_set_msix_vec_count,
2281 };
2282 
2283 /**
2284  * mlx5_vf_get_core_dev - Get the mlx5 core device from a given VF PCI device if
2285  *                     mlx5_core is its driver.
2286  * @pdev: The associated PCI device.
2287  *
2288  * Upon return the interface state lock stay held to let caller uses it safely.
2289  * Caller must ensure to use the returned mlx5 device for a narrow window
2290  * and put it back with mlx5_vf_put_core_dev() immediately once usage was over.
2291  *
2292  * Return: Pointer to the associated mlx5_core_dev or NULL.
2293  */
mlx5_vf_get_core_dev(struct pci_dev * pdev)2294 struct mlx5_core_dev *mlx5_vf_get_core_dev(struct pci_dev *pdev)
2295 {
2296 	struct mlx5_core_dev *mdev;
2297 
2298 	mdev = pci_iov_get_pf_drvdata(pdev, &mlx5_core_driver);
2299 	if (IS_ERR(mdev))
2300 		return NULL;
2301 
2302 	mutex_lock(&mdev->intf_state_mutex);
2303 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &mdev->intf_state)) {
2304 		mutex_unlock(&mdev->intf_state_mutex);
2305 		return NULL;
2306 	}
2307 
2308 	return mdev;
2309 }
2310 EXPORT_SYMBOL(mlx5_vf_get_core_dev);
2311 
2312 /**
2313  * mlx5_vf_put_core_dev - Put the mlx5 core device back.
2314  * @mdev: The mlx5 core device.
2315  *
2316  * Upon return the interface state lock is unlocked and caller should not
2317  * access the mdev any more.
2318  */
mlx5_vf_put_core_dev(struct mlx5_core_dev * mdev)2319 void mlx5_vf_put_core_dev(struct mlx5_core_dev *mdev)
2320 {
2321 	mutex_unlock(&mdev->intf_state_mutex);
2322 }
2323 EXPORT_SYMBOL(mlx5_vf_put_core_dev);
2324 
mlx5_core_verify_params(void)2325 static void mlx5_core_verify_params(void)
2326 {
2327 	if (prof_sel >= ARRAY_SIZE(profile)) {
2328 		pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
2329 			prof_sel,
2330 			ARRAY_SIZE(profile) - 1,
2331 			MLX5_DEFAULT_PROF);
2332 		prof_sel = MLX5_DEFAULT_PROF;
2333 	}
2334 }
2335 
mlx5_init(void)2336 static int __init mlx5_init(void)
2337 {
2338 	int err;
2339 
2340 	WARN_ONCE(strcmp(MLX5_ADEV_NAME, KBUILD_MODNAME),
2341 		  "mlx5_core name not in sync with kernel module name");
2342 
2343 	get_random_bytes(&sw_owner_id, sizeof(sw_owner_id));
2344 
2345 	mlx5_core_verify_params();
2346 	mlx5_register_debugfs();
2347 
2348 	err = mlx5e_init();
2349 	if (err)
2350 		goto err_debug;
2351 
2352 	err = mlx5_sf_driver_register();
2353 	if (err)
2354 		goto err_sf;
2355 
2356 	err = pci_register_driver(&mlx5_core_driver);
2357 	if (err)
2358 		goto err_pci;
2359 
2360 	return 0;
2361 
2362 err_pci:
2363 	mlx5_sf_driver_unregister();
2364 err_sf:
2365 	mlx5e_cleanup();
2366 err_debug:
2367 	mlx5_unregister_debugfs();
2368 	return err;
2369 }
2370 
mlx5_cleanup(void)2371 static void __exit mlx5_cleanup(void)
2372 {
2373 	pci_unregister_driver(&mlx5_core_driver);
2374 	mlx5_sf_driver_unregister();
2375 	mlx5e_cleanup();
2376 	mlx5_unregister_debugfs();
2377 }
2378 
2379 module_init(mlx5_init);
2380 module_exit(mlx5_cleanup);
2381