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 #ifdef CONFIG_RFS_ACCEL
50 #include <linux/cpu_rmap.h>
51 #endif
52 #include <linux/version.h>
53 #include <net/devlink.h>
54 #include "mlx5_core.h"
55 #include "lib/eq.h"
56 #include "fs_core.h"
57 #include "lib/mpfs.h"
58 #include "eswitch.h"
59 #include "devlink.h"
60 #include "fw_reset.h"
61 #include "lib/mlx5.h"
62 #include "lib/tout.h"
63 #include "fpga/core.h"
64 #include "en_accel/ipsec.h"
65 #include "lib/clock.h"
66 #include "lib/vxlan.h"
67 #include "lib/geneve.h"
68 #include "lib/devcom.h"
69 #include "lib/pci_vsc.h"
70 #include "diag/fw_tracer.h"
71 #include "ecpf.h"
72 #include "lib/hv_vhca.h"
73 #include "diag/rsc_dump.h"
74 #include "sf/vhca_event.h"
75 #include "sf/dev/dev.h"
76 #include "sf/sf.h"
77 #include "mlx5_irq.h"
78 
79 MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
80 MODULE_DESCRIPTION("Mellanox 5th generation network adapters (ConnectX series) core driver");
81 MODULE_LICENSE("Dual BSD/GPL");
82 
83 unsigned int mlx5_core_debug_mask;
84 module_param_named(debug_mask, mlx5_core_debug_mask, uint, 0644);
85 MODULE_PARM_DESC(debug_mask, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0");
86 
87 static unsigned int prof_sel = MLX5_DEFAULT_PROF;
88 module_param_named(prof_sel, prof_sel, uint, 0444);
89 MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
90 
91 static u32 sw_owner_id[4];
92 #define MAX_SW_VHCA_ID (BIT(__mlx5_bit_sz(cmd_hca_cap_2, sw_vhca_id)) - 1)
93 static DEFINE_IDA(sw_vhca_ida);
94 
95 enum {
96 	MLX5_ATOMIC_REQ_MODE_BE = 0x0,
97 	MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1,
98 };
99 
100 #define LOG_MAX_SUPPORTED_QPS 0xff
101 
102 static struct mlx5_profile profile[] = {
103 	[0] = {
104 		.mask           = 0,
105 	},
106 	[1] = {
107 		.mask		= MLX5_PROF_MASK_QP_SIZE,
108 		.log_max_qp	= 12,
109 	},
110 	[2] = {
111 		.mask		= MLX5_PROF_MASK_QP_SIZE |
112 				  MLX5_PROF_MASK_MR_CACHE,
113 		.log_max_qp	= LOG_MAX_SUPPORTED_QPS,
114 		.mr_cache[0]	= {
115 			.size	= 500,
116 			.limit	= 250
117 		},
118 		.mr_cache[1]	= {
119 			.size	= 500,
120 			.limit	= 250
121 		},
122 		.mr_cache[2]	= {
123 			.size	= 500,
124 			.limit	= 250
125 		},
126 		.mr_cache[3]	= {
127 			.size	= 500,
128 			.limit	= 250
129 		},
130 		.mr_cache[4]	= {
131 			.size	= 500,
132 			.limit	= 250
133 		},
134 		.mr_cache[5]	= {
135 			.size	= 500,
136 			.limit	= 250
137 		},
138 		.mr_cache[6]	= {
139 			.size	= 500,
140 			.limit	= 250
141 		},
142 		.mr_cache[7]	= {
143 			.size	= 500,
144 			.limit	= 250
145 		},
146 		.mr_cache[8]	= {
147 			.size	= 500,
148 			.limit	= 250
149 		},
150 		.mr_cache[9]	= {
151 			.size	= 500,
152 			.limit	= 250
153 		},
154 		.mr_cache[10]	= {
155 			.size	= 500,
156 			.limit	= 250
157 		},
158 		.mr_cache[11]	= {
159 			.size	= 500,
160 			.limit	= 250
161 		},
162 		.mr_cache[12]	= {
163 			.size	= 64,
164 			.limit	= 32
165 		},
166 		.mr_cache[13]	= {
167 			.size	= 32,
168 			.limit	= 16
169 		},
170 		.mr_cache[14]	= {
171 			.size	= 16,
172 			.limit	= 8
173 		},
174 		.mr_cache[15]	= {
175 			.size	= 8,
176 			.limit	= 4
177 		},
178 	},
179 };
180 
wait_fw_init(struct mlx5_core_dev * dev,u32 max_wait_mili,u32 warn_time_mili)181 static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili,
182 			u32 warn_time_mili)
183 {
184 	unsigned long warn = jiffies + msecs_to_jiffies(warn_time_mili);
185 	unsigned long end = jiffies + msecs_to_jiffies(max_wait_mili);
186 	u32 fw_initializing;
187 	int err = 0;
188 
189 	do {
190 		fw_initializing = ioread32be(&dev->iseg->initializing);
191 		if (!(fw_initializing >> 31))
192 			break;
193 		if (time_after(jiffies, end) ||
194 		    test_and_clear_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state)) {
195 			err = -EBUSY;
196 			break;
197 		}
198 		if (warn_time_mili && time_after(jiffies, warn)) {
199 			mlx5_core_warn(dev, "Waiting for FW initialization, timeout abort in %ds (0x%x)\n",
200 				       jiffies_to_msecs(end - warn) / 1000, fw_initializing);
201 			warn = jiffies + msecs_to_jiffies(warn_time_mili);
202 		}
203 		msleep(mlx5_tout_ms(dev, FW_PRE_INIT_WAIT));
204 	} while (true);
205 
206 	return err;
207 }
208 
mlx5_set_driver_version(struct mlx5_core_dev * dev)209 static void mlx5_set_driver_version(struct mlx5_core_dev *dev)
210 {
211 	int driver_ver_sz = MLX5_FLD_SZ_BYTES(set_driver_version_in,
212 					      driver_version);
213 	u8 in[MLX5_ST_SZ_BYTES(set_driver_version_in)] = {};
214 	int remaining_size = driver_ver_sz;
215 	char *string;
216 
217 	if (!MLX5_CAP_GEN(dev, driver_version))
218 		return;
219 
220 	string = MLX5_ADDR_OF(set_driver_version_in, in, driver_version);
221 
222 	strncpy(string, "Linux", remaining_size);
223 
224 	remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
225 	strncat(string, ",", remaining_size);
226 
227 	remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
228 	strncat(string, KBUILD_MODNAME, remaining_size);
229 
230 	remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
231 	strncat(string, ",", remaining_size);
232 
233 	remaining_size = max_t(int, 0, driver_ver_sz - strlen(string));
234 
235 	snprintf(string + strlen(string), remaining_size, "%u.%u.%u",
236 		LINUX_VERSION_MAJOR, LINUX_VERSION_PATCHLEVEL,
237 		LINUX_VERSION_SUBLEVEL);
238 
239 	/*Send the command*/
240 	MLX5_SET(set_driver_version_in, in, opcode,
241 		 MLX5_CMD_OP_SET_DRIVER_VERSION);
242 
243 	mlx5_cmd_exec_in(dev, set_driver_version, in);
244 }
245 
set_dma_caps(struct pci_dev * pdev)246 static int set_dma_caps(struct pci_dev *pdev)
247 {
248 	int err;
249 
250 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
251 	if (err) {
252 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask\n");
253 		err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
254 		if (err) {
255 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting\n");
256 			return err;
257 		}
258 	}
259 
260 	dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024);
261 	return err;
262 }
263 
mlx5_pci_enable_device(struct mlx5_core_dev * dev)264 static int mlx5_pci_enable_device(struct mlx5_core_dev *dev)
265 {
266 	struct pci_dev *pdev = dev->pdev;
267 	int err = 0;
268 
269 	mutex_lock(&dev->pci_status_mutex);
270 	if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) {
271 		err = pci_enable_device(pdev);
272 		if (!err)
273 			dev->pci_status = MLX5_PCI_STATUS_ENABLED;
274 	}
275 	mutex_unlock(&dev->pci_status_mutex);
276 
277 	return err;
278 }
279 
mlx5_pci_disable_device(struct mlx5_core_dev * dev)280 static void mlx5_pci_disable_device(struct mlx5_core_dev *dev)
281 {
282 	struct pci_dev *pdev = dev->pdev;
283 
284 	mutex_lock(&dev->pci_status_mutex);
285 	if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) {
286 		pci_disable_device(pdev);
287 		dev->pci_status = MLX5_PCI_STATUS_DISABLED;
288 	}
289 	mutex_unlock(&dev->pci_status_mutex);
290 }
291 
request_bar(struct pci_dev * pdev)292 static int request_bar(struct pci_dev *pdev)
293 {
294 	int err = 0;
295 
296 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) {
297 		dev_err(&pdev->dev, "Missing registers BAR, aborting\n");
298 		return -ENODEV;
299 	}
300 
301 	err = pci_request_regions(pdev, KBUILD_MODNAME);
302 	if (err)
303 		dev_err(&pdev->dev, "Couldn't get PCI resources, aborting\n");
304 
305 	return err;
306 }
307 
release_bar(struct pci_dev * pdev)308 static void release_bar(struct pci_dev *pdev)
309 {
310 	pci_release_regions(pdev);
311 }
312 
313 struct mlx5_reg_host_endianness {
314 	u8	he;
315 	u8      rsvd[15];
316 };
317 
to_fw_pkey_sz(struct mlx5_core_dev * dev,u32 size)318 static u16 to_fw_pkey_sz(struct mlx5_core_dev *dev, u32 size)
319 {
320 	switch (size) {
321 	case 128:
322 		return 0;
323 	case 256:
324 		return 1;
325 	case 512:
326 		return 2;
327 	case 1024:
328 		return 3;
329 	case 2048:
330 		return 4;
331 	case 4096:
332 		return 5;
333 	default:
334 		mlx5_core_warn(dev, "invalid pkey table size %d\n", size);
335 		return 0;
336 	}
337 }
338 
mlx5_core_uplink_netdev_set(struct mlx5_core_dev * dev,struct net_device * netdev)339 void mlx5_core_uplink_netdev_set(struct mlx5_core_dev *dev, struct net_device *netdev)
340 {
341 	mutex_lock(&dev->mlx5e_res.uplink_netdev_lock);
342 	dev->mlx5e_res.uplink_netdev = netdev;
343 	mlx5_blocking_notifier_call_chain(dev, MLX5_DRIVER_EVENT_UPLINK_NETDEV,
344 					  netdev);
345 	mutex_unlock(&dev->mlx5e_res.uplink_netdev_lock);
346 }
347 
mlx5_core_uplink_netdev_event_replay(struct mlx5_core_dev * dev)348 void mlx5_core_uplink_netdev_event_replay(struct mlx5_core_dev *dev)
349 {
350 	mutex_lock(&dev->mlx5e_res.uplink_netdev_lock);
351 	mlx5_blocking_notifier_call_chain(dev, MLX5_DRIVER_EVENT_UPLINK_NETDEV,
352 					  dev->mlx5e_res.uplink_netdev);
353 	mutex_unlock(&dev->mlx5e_res.uplink_netdev_lock);
354 }
355 EXPORT_SYMBOL(mlx5_core_uplink_netdev_event_replay);
356 
mlx5_core_get_caps_mode(struct mlx5_core_dev * dev,enum mlx5_cap_type cap_type,enum mlx5_cap_mode cap_mode)357 static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev,
358 				   enum mlx5_cap_type cap_type,
359 				   enum mlx5_cap_mode cap_mode)
360 {
361 	u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)];
362 	int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
363 	void *out, *hca_caps;
364 	u16 opmod = (cap_type << 1) | (cap_mode & 0x01);
365 	int err;
366 
367 	memset(in, 0, sizeof(in));
368 	out = kzalloc(out_sz, GFP_KERNEL);
369 	if (!out)
370 		return -ENOMEM;
371 
372 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
373 	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
374 	err = mlx5_cmd_exec_inout(dev, query_hca_cap, in, out);
375 	if (err) {
376 		mlx5_core_warn(dev,
377 			       "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n",
378 			       cap_type, cap_mode, err);
379 		goto query_ex;
380 	}
381 
382 	hca_caps =  MLX5_ADDR_OF(query_hca_cap_out, out, capability);
383 
384 	switch (cap_mode) {
385 	case HCA_CAP_OPMOD_GET_MAX:
386 		memcpy(dev->caps.hca[cap_type]->max, hca_caps,
387 		       MLX5_UN_SZ_BYTES(hca_cap_union));
388 		break;
389 	case HCA_CAP_OPMOD_GET_CUR:
390 		memcpy(dev->caps.hca[cap_type]->cur, hca_caps,
391 		       MLX5_UN_SZ_BYTES(hca_cap_union));
392 		break;
393 	default:
394 		mlx5_core_warn(dev,
395 			       "Tried to query dev cap type(%x) with wrong opmode(%x)\n",
396 			       cap_type, cap_mode);
397 		err = -EINVAL;
398 		break;
399 	}
400 query_ex:
401 	kfree(out);
402 	return err;
403 }
404 
mlx5_core_get_caps(struct mlx5_core_dev * dev,enum mlx5_cap_type cap_type)405 int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type)
406 {
407 	int ret;
408 
409 	ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR);
410 	if (ret)
411 		return ret;
412 	return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX);
413 }
414 
set_caps(struct mlx5_core_dev * dev,void * in,int opmod)415 static int set_caps(struct mlx5_core_dev *dev, void *in, int opmod)
416 {
417 	MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP);
418 	MLX5_SET(set_hca_cap_in, in, op_mod, opmod << 1);
419 	return mlx5_cmd_exec_in(dev, set_hca_cap, in);
420 }
421 
handle_hca_cap_atomic(struct mlx5_core_dev * dev,void * set_ctx)422 static int handle_hca_cap_atomic(struct mlx5_core_dev *dev, void *set_ctx)
423 {
424 	void *set_hca_cap;
425 	int req_endianness;
426 	int err;
427 
428 	if (!MLX5_CAP_GEN(dev, atomic))
429 		return 0;
430 
431 	err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC);
432 	if (err)
433 		return err;
434 
435 	req_endianness =
436 		MLX5_CAP_ATOMIC(dev,
437 				supported_atomic_req_8B_endianness_mode_1);
438 
439 	if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS)
440 		return 0;
441 
442 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
443 
444 	/* Set requestor to host endianness */
445 	MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianness_mode,
446 		 MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS);
447 
448 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ATOMIC);
449 }
450 
handle_hca_cap_odp(struct mlx5_core_dev * dev,void * set_ctx)451 static int handle_hca_cap_odp(struct mlx5_core_dev *dev, void *set_ctx)
452 {
453 	void *set_hca_cap;
454 	bool do_set = false;
455 	int err;
456 
457 	if (!IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) ||
458 	    !MLX5_CAP_GEN(dev, pg))
459 		return 0;
460 
461 	err = mlx5_core_get_caps(dev, MLX5_CAP_ODP);
462 	if (err)
463 		return err;
464 
465 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
466 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_ODP]->cur,
467 	       MLX5_ST_SZ_BYTES(odp_cap));
468 
469 #define ODP_CAP_SET_MAX(dev, field)                                            \
470 	do {                                                                   \
471 		u32 _res = MLX5_CAP_ODP_MAX(dev, field);                       \
472 		if (_res) {                                                    \
473 			do_set = true;                                         \
474 			MLX5_SET(odp_cap, set_hca_cap, field, _res);           \
475 		}                                                              \
476 	} while (0)
477 
478 	ODP_CAP_SET_MAX(dev, ud_odp_caps.srq_receive);
479 	ODP_CAP_SET_MAX(dev, rc_odp_caps.srq_receive);
480 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.srq_receive);
481 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.send);
482 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.receive);
483 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.write);
484 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.read);
485 	ODP_CAP_SET_MAX(dev, xrc_odp_caps.atomic);
486 	ODP_CAP_SET_MAX(dev, dc_odp_caps.srq_receive);
487 	ODP_CAP_SET_MAX(dev, dc_odp_caps.send);
488 	ODP_CAP_SET_MAX(dev, dc_odp_caps.receive);
489 	ODP_CAP_SET_MAX(dev, dc_odp_caps.write);
490 	ODP_CAP_SET_MAX(dev, dc_odp_caps.read);
491 	ODP_CAP_SET_MAX(dev, dc_odp_caps.atomic);
492 
493 	if (!do_set)
494 		return 0;
495 
496 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ODP);
497 }
498 
max_uc_list_get_devlink_param(struct mlx5_core_dev * dev)499 static int max_uc_list_get_devlink_param(struct mlx5_core_dev *dev)
500 {
501 	struct devlink *devlink = priv_to_devlink(dev);
502 	union devlink_param_value val;
503 	int err;
504 
505 	err = devl_param_driverinit_value_get(devlink,
506 					      DEVLINK_PARAM_GENERIC_ID_MAX_MACS,
507 					      &val);
508 	if (!err)
509 		return val.vu32;
510 	mlx5_core_dbg(dev, "Failed to get param. err = %d\n", err);
511 	return err;
512 }
513 
mlx5_is_roce_on(struct mlx5_core_dev * dev)514 bool mlx5_is_roce_on(struct mlx5_core_dev *dev)
515 {
516 	struct devlink *devlink = priv_to_devlink(dev);
517 	union devlink_param_value val;
518 	int err;
519 
520 	err = devl_param_driverinit_value_get(devlink,
521 					      DEVLINK_PARAM_GENERIC_ID_ENABLE_ROCE,
522 					      &val);
523 
524 	if (!err)
525 		return val.vbool;
526 
527 	mlx5_core_dbg(dev, "Failed to get param. err = %d\n", err);
528 	return MLX5_CAP_GEN(dev, roce);
529 }
530 EXPORT_SYMBOL(mlx5_is_roce_on);
531 
handle_hca_cap_2(struct mlx5_core_dev * dev,void * set_ctx)532 static int handle_hca_cap_2(struct mlx5_core_dev *dev, void *set_ctx)
533 {
534 	void *set_hca_cap;
535 	int err;
536 
537 	if (!MLX5_CAP_GEN_MAX(dev, hca_cap_2))
538 		return 0;
539 
540 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL_2);
541 	if (err)
542 		return err;
543 
544 	if (!MLX5_CAP_GEN_2_MAX(dev, sw_vhca_id_valid) ||
545 	    !(dev->priv.sw_vhca_id > 0))
546 		return 0;
547 
548 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
549 				   capability);
550 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_GENERAL_2]->cur,
551 	       MLX5_ST_SZ_BYTES(cmd_hca_cap_2));
552 	MLX5_SET(cmd_hca_cap_2, set_hca_cap, sw_vhca_id_valid, 1);
553 
554 	return set_caps(dev, set_ctx, MLX5_CAP_GENERAL_2);
555 }
556 
handle_hca_cap(struct mlx5_core_dev * dev,void * set_ctx)557 static int handle_hca_cap(struct mlx5_core_dev *dev, void *set_ctx)
558 {
559 	struct mlx5_profile *prof = &dev->profile;
560 	void *set_hca_cap;
561 	int max_uc_list;
562 	int err;
563 
564 	err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL);
565 	if (err)
566 		return err;
567 
568 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx,
569 				   capability);
570 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_GENERAL]->cur,
571 	       MLX5_ST_SZ_BYTES(cmd_hca_cap));
572 
573 	mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n",
574 		      mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)),
575 		      128);
576 	/* we limit the size of the pkey table to 128 entries for now */
577 	MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size,
578 		 to_fw_pkey_sz(dev, 128));
579 
580 	/* Check log_max_qp from HCA caps to set in current profile */
581 	if (prof->log_max_qp == LOG_MAX_SUPPORTED_QPS) {
582 		prof->log_max_qp = min_t(u8, 18, MLX5_CAP_GEN_MAX(dev, log_max_qp));
583 	} else if (MLX5_CAP_GEN_MAX(dev, log_max_qp) < prof->log_max_qp) {
584 		mlx5_core_warn(dev, "log_max_qp value in current profile is %d, changing it to HCA capability limit (%d)\n",
585 			       prof->log_max_qp,
586 			       MLX5_CAP_GEN_MAX(dev, log_max_qp));
587 		prof->log_max_qp = MLX5_CAP_GEN_MAX(dev, log_max_qp);
588 	}
589 	if (prof->mask & MLX5_PROF_MASK_QP_SIZE)
590 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp,
591 			 prof->log_max_qp);
592 
593 	/* disable cmdif checksum */
594 	MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0);
595 
596 	/* Enable 4K UAR only when HCA supports it and page size is bigger
597 	 * than 4K.
598 	 */
599 	if (MLX5_CAP_GEN_MAX(dev, uar_4k) && PAGE_SIZE > 4096)
600 		MLX5_SET(cmd_hca_cap, set_hca_cap, uar_4k, 1);
601 
602 	MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12);
603 
604 	if (MLX5_CAP_GEN_MAX(dev, cache_line_128byte))
605 		MLX5_SET(cmd_hca_cap,
606 			 set_hca_cap,
607 			 cache_line_128byte,
608 			 cache_line_size() >= 128 ? 1 : 0);
609 
610 	if (MLX5_CAP_GEN_MAX(dev, dct))
611 		MLX5_SET(cmd_hca_cap, set_hca_cap, dct, 1);
612 
613 	if (MLX5_CAP_GEN_MAX(dev, pci_sync_for_fw_update_event))
614 		MLX5_SET(cmd_hca_cap, set_hca_cap, pci_sync_for_fw_update_event, 1);
615 
616 	if (MLX5_CAP_GEN_MAX(dev, num_vhca_ports))
617 		MLX5_SET(cmd_hca_cap,
618 			 set_hca_cap,
619 			 num_vhca_ports,
620 			 MLX5_CAP_GEN_MAX(dev, num_vhca_ports));
621 
622 	if (MLX5_CAP_GEN_MAX(dev, release_all_pages))
623 		MLX5_SET(cmd_hca_cap, set_hca_cap, release_all_pages, 1);
624 
625 	if (MLX5_CAP_GEN_MAX(dev, mkey_by_name))
626 		MLX5_SET(cmd_hca_cap, set_hca_cap, mkey_by_name, 1);
627 
628 	mlx5_vhca_state_cap_handle(dev, set_hca_cap);
629 
630 	if (MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix))
631 		MLX5_SET(cmd_hca_cap, set_hca_cap, num_total_dynamic_vf_msix,
632 			 MLX5_CAP_GEN_MAX(dev, num_total_dynamic_vf_msix));
633 
634 	if (MLX5_CAP_GEN(dev, roce_rw_supported) && MLX5_CAP_GEN_MAX(dev, roce))
635 		MLX5_SET(cmd_hca_cap, set_hca_cap, roce,
636 			 mlx5_is_roce_on(dev));
637 
638 	max_uc_list = max_uc_list_get_devlink_param(dev);
639 	if (max_uc_list > 0)
640 		MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_current_uc_list,
641 			 ilog2(max_uc_list));
642 
643 	return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
644 }
645 
646 /* Cached MLX5_CAP_GEN(dev, roce) can be out of sync this early in the
647  * boot process.
648  * In case RoCE cap is writable in FW and user/devlink requested to change the
649  * cap, we are yet to query the final state of the above cap.
650  * Hence, the need for this function.
651  *
652  * Returns
653  * True:
654  * 1) RoCE cap is read only in FW and already disabled
655  * OR:
656  * 2) RoCE cap is writable in FW and user/devlink requested it off.
657  *
658  * In any other case, return False.
659  */
is_roce_fw_disabled(struct mlx5_core_dev * dev)660 static bool is_roce_fw_disabled(struct mlx5_core_dev *dev)
661 {
662 	return (MLX5_CAP_GEN(dev, roce_rw_supported) && !mlx5_is_roce_on(dev)) ||
663 		(!MLX5_CAP_GEN(dev, roce_rw_supported) && !MLX5_CAP_GEN(dev, roce));
664 }
665 
handle_hca_cap_roce(struct mlx5_core_dev * dev,void * set_ctx)666 static int handle_hca_cap_roce(struct mlx5_core_dev *dev, void *set_ctx)
667 {
668 	void *set_hca_cap;
669 	int err;
670 
671 	if (is_roce_fw_disabled(dev))
672 		return 0;
673 
674 	err = mlx5_core_get_caps(dev, MLX5_CAP_ROCE);
675 	if (err)
676 		return err;
677 
678 	if (MLX5_CAP_ROCE(dev, sw_r_roce_src_udp_port) ||
679 	    !MLX5_CAP_ROCE_MAX(dev, sw_r_roce_src_udp_port))
680 		return 0;
681 
682 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
683 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_ROCE]->cur,
684 	       MLX5_ST_SZ_BYTES(roce_cap));
685 	MLX5_SET(roce_cap, set_hca_cap, sw_r_roce_src_udp_port, 1);
686 
687 	err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_ROCE);
688 	return err;
689 }
690 
handle_hca_cap_port_selection(struct mlx5_core_dev * dev,void * set_ctx)691 static int handle_hca_cap_port_selection(struct mlx5_core_dev *dev,
692 					 void *set_ctx)
693 {
694 	void *set_hca_cap;
695 	int err;
696 
697 	if (!MLX5_CAP_GEN(dev, port_selection_cap))
698 		return 0;
699 
700 	err = mlx5_core_get_caps(dev, MLX5_CAP_PORT_SELECTION);
701 	if (err)
702 		return err;
703 
704 	if (MLX5_CAP_PORT_SELECTION(dev, port_select_flow_table_bypass) ||
705 	    !MLX5_CAP_PORT_SELECTION_MAX(dev, port_select_flow_table_bypass))
706 		return 0;
707 
708 	set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability);
709 	memcpy(set_hca_cap, dev->caps.hca[MLX5_CAP_PORT_SELECTION]->cur,
710 	       MLX5_ST_SZ_BYTES(port_selection_cap));
711 	MLX5_SET(port_selection_cap, set_hca_cap, port_select_flow_table_bypass, 1);
712 
713 	err = set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MODE_PORT_SELECTION);
714 
715 	return err;
716 }
717 
set_hca_cap(struct mlx5_core_dev * dev)718 static int set_hca_cap(struct mlx5_core_dev *dev)
719 {
720 	int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in);
721 	void *set_ctx;
722 	int err;
723 
724 	set_ctx = kzalloc(set_sz, GFP_KERNEL);
725 	if (!set_ctx)
726 		return -ENOMEM;
727 
728 	err = handle_hca_cap(dev, set_ctx);
729 	if (err) {
730 		mlx5_core_err(dev, "handle_hca_cap failed\n");
731 		goto out;
732 	}
733 
734 	memset(set_ctx, 0, set_sz);
735 	err = handle_hca_cap_atomic(dev, set_ctx);
736 	if (err) {
737 		mlx5_core_err(dev, "handle_hca_cap_atomic failed\n");
738 		goto out;
739 	}
740 
741 	memset(set_ctx, 0, set_sz);
742 	err = handle_hca_cap_odp(dev, set_ctx);
743 	if (err) {
744 		mlx5_core_err(dev, "handle_hca_cap_odp failed\n");
745 		goto out;
746 	}
747 
748 	memset(set_ctx, 0, set_sz);
749 	err = handle_hca_cap_roce(dev, set_ctx);
750 	if (err) {
751 		mlx5_core_err(dev, "handle_hca_cap_roce failed\n");
752 		goto out;
753 	}
754 
755 	memset(set_ctx, 0, set_sz);
756 	err = handle_hca_cap_2(dev, set_ctx);
757 	if (err) {
758 		mlx5_core_err(dev, "handle_hca_cap_2 failed\n");
759 		goto out;
760 	}
761 
762 	memset(set_ctx, 0, set_sz);
763 	err = handle_hca_cap_port_selection(dev, set_ctx);
764 	if (err) {
765 		mlx5_core_err(dev, "handle_hca_cap_port_selection failed\n");
766 		goto out;
767 	}
768 
769 out:
770 	kfree(set_ctx);
771 	return err;
772 }
773 
set_hca_ctrl(struct mlx5_core_dev * dev)774 static int set_hca_ctrl(struct mlx5_core_dev *dev)
775 {
776 	struct mlx5_reg_host_endianness he_in;
777 	struct mlx5_reg_host_endianness he_out;
778 	int err;
779 
780 	if (!mlx5_core_is_pf(dev))
781 		return 0;
782 
783 	memset(&he_in, 0, sizeof(he_in));
784 	he_in.he = MLX5_SET_HOST_ENDIANNESS;
785 	err = mlx5_core_access_reg(dev, &he_in,  sizeof(he_in),
786 					&he_out, sizeof(he_out),
787 					MLX5_REG_HOST_ENDIANNESS, 0, 1);
788 	return err;
789 }
790 
mlx5_core_set_hca_defaults(struct mlx5_core_dev * dev)791 static int mlx5_core_set_hca_defaults(struct mlx5_core_dev *dev)
792 {
793 	int ret = 0;
794 
795 	/* Disable local_lb by default */
796 	if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH)
797 		ret = mlx5_nic_vport_update_local_lb(dev, false);
798 
799 	return ret;
800 }
801 
mlx5_core_enable_hca(struct mlx5_core_dev * dev,u16 func_id)802 int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id)
803 {
804 	u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {};
805 
806 	MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA);
807 	MLX5_SET(enable_hca_in, in, function_id, func_id);
808 	MLX5_SET(enable_hca_in, in, embedded_cpu_function,
809 		 dev->caps.embedded_cpu);
810 	return mlx5_cmd_exec_in(dev, enable_hca, in);
811 }
812 
mlx5_core_disable_hca(struct mlx5_core_dev * dev,u16 func_id)813 int mlx5_core_disable_hca(struct mlx5_core_dev *dev, u16 func_id)
814 {
815 	u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {};
816 
817 	MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA);
818 	MLX5_SET(disable_hca_in, in, function_id, func_id);
819 	MLX5_SET(enable_hca_in, in, embedded_cpu_function,
820 		 dev->caps.embedded_cpu);
821 	return mlx5_cmd_exec_in(dev, disable_hca, in);
822 }
823 
mlx5_core_set_issi(struct mlx5_core_dev * dev)824 static int mlx5_core_set_issi(struct mlx5_core_dev *dev)
825 {
826 	u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {};
827 	u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {};
828 	u32 sup_issi;
829 	int err;
830 
831 	MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI);
832 	err = mlx5_cmd_exec_inout(dev, query_issi, query_in, query_out);
833 	if (err) {
834 		u32 syndrome = MLX5_GET(query_issi_out, query_out, syndrome);
835 		u8 status = MLX5_GET(query_issi_out, query_out, status);
836 
837 		if (!status || syndrome == MLX5_DRIVER_SYND) {
838 			mlx5_core_err(dev, "Failed to query ISSI err(%d) status(%d) synd(%d)\n",
839 				      err, status, syndrome);
840 			return err;
841 		}
842 
843 		mlx5_core_warn(dev, "Query ISSI is not supported by FW, ISSI is 0\n");
844 		dev->issi = 0;
845 		return 0;
846 	}
847 
848 	sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0);
849 
850 	if (sup_issi & (1 << 1)) {
851 		u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {};
852 
853 		MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI);
854 		MLX5_SET(set_issi_in, set_in, current_issi, 1);
855 		err = mlx5_cmd_exec_in(dev, set_issi, set_in);
856 		if (err) {
857 			mlx5_core_err(dev, "Failed to set ISSI to 1 err(%d)\n",
858 				      err);
859 			return err;
860 		}
861 
862 		dev->issi = 1;
863 
864 		return 0;
865 	} else if (sup_issi & (1 << 0) || !sup_issi) {
866 		return 0;
867 	}
868 
869 	return -EOPNOTSUPP;
870 }
871 
mlx5_pci_init(struct mlx5_core_dev * dev,struct pci_dev * pdev,const struct pci_device_id * id)872 static int mlx5_pci_init(struct mlx5_core_dev *dev, struct pci_dev *pdev,
873 			 const struct pci_device_id *id)
874 {
875 	int err = 0;
876 
877 	mutex_init(&dev->pci_status_mutex);
878 	pci_set_drvdata(dev->pdev, dev);
879 
880 	dev->bar_addr = pci_resource_start(pdev, 0);
881 
882 	err = mlx5_pci_enable_device(dev);
883 	if (err) {
884 		mlx5_core_err(dev, "Cannot enable PCI device, aborting\n");
885 		return err;
886 	}
887 
888 	err = request_bar(pdev);
889 	if (err) {
890 		mlx5_core_err(dev, "error requesting BARs, aborting\n");
891 		goto err_disable;
892 	}
893 
894 	pci_set_master(pdev);
895 
896 	err = set_dma_caps(pdev);
897 	if (err) {
898 		mlx5_core_err(dev, "Failed setting DMA capabilities mask, aborting\n");
899 		goto err_clr_master;
900 	}
901 
902 	if (pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP32) &&
903 	    pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP64) &&
904 	    pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP128))
905 		mlx5_core_dbg(dev, "Enabling pci atomics failed\n");
906 
907 	dev->iseg_base = dev->bar_addr;
908 	dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg));
909 	if (!dev->iseg) {
910 		err = -ENOMEM;
911 		mlx5_core_err(dev, "Failed mapping initialization segment, aborting\n");
912 		goto err_clr_master;
913 	}
914 
915 	mlx5_pci_vsc_init(dev);
916 	dev->caps.embedded_cpu = mlx5_read_embedded_cpu(dev);
917 	return 0;
918 
919 err_clr_master:
920 	pci_clear_master(dev->pdev);
921 	release_bar(dev->pdev);
922 err_disable:
923 	mlx5_pci_disable_device(dev);
924 	return err;
925 }
926 
mlx5_pci_close(struct mlx5_core_dev * dev)927 static void mlx5_pci_close(struct mlx5_core_dev *dev)
928 {
929 	/* health work might still be active, and it needs pci bar in
930 	 * order to know the NIC state. Therefore, drain the health WQ
931 	 * before removing the pci bars
932 	 */
933 	mlx5_drain_health_wq(dev);
934 	iounmap(dev->iseg);
935 	pci_clear_master(dev->pdev);
936 	release_bar(dev->pdev);
937 	mlx5_pci_disable_device(dev);
938 }
939 
mlx5_init_once(struct mlx5_core_dev * dev)940 static int mlx5_init_once(struct mlx5_core_dev *dev)
941 {
942 	int err;
943 
944 	dev->priv.devcom = mlx5_devcom_register_device(dev);
945 	if (IS_ERR(dev->priv.devcom))
946 		mlx5_core_err(dev, "failed to register with devcom (0x%p)\n",
947 			      dev->priv.devcom);
948 
949 	err = mlx5_query_board_id(dev);
950 	if (err) {
951 		mlx5_core_err(dev, "query board id failed\n");
952 		goto err_devcom;
953 	}
954 
955 	err = mlx5_irq_table_init(dev);
956 	if (err) {
957 		mlx5_core_err(dev, "failed to initialize irq table\n");
958 		goto err_devcom;
959 	}
960 
961 	err = mlx5_eq_table_init(dev);
962 	if (err) {
963 		mlx5_core_err(dev, "failed to initialize eq\n");
964 		goto err_irq_cleanup;
965 	}
966 
967 	err = mlx5_events_init(dev);
968 	if (err) {
969 		mlx5_core_err(dev, "failed to initialize events\n");
970 		goto err_eq_cleanup;
971 	}
972 
973 	err = mlx5_fw_reset_init(dev);
974 	if (err) {
975 		mlx5_core_err(dev, "failed to initialize fw reset events\n");
976 		goto err_events_cleanup;
977 	}
978 
979 	mlx5_cq_debugfs_init(dev);
980 
981 	mlx5_init_reserved_gids(dev);
982 
983 	mlx5_init_clock(dev);
984 
985 	dev->vxlan = mlx5_vxlan_create(dev);
986 	dev->geneve = mlx5_geneve_create(dev);
987 
988 	err = mlx5_init_rl_table(dev);
989 	if (err) {
990 		mlx5_core_err(dev, "Failed to init rate limiting\n");
991 		goto err_tables_cleanup;
992 	}
993 
994 	err = mlx5_mpfs_init(dev);
995 	if (err) {
996 		mlx5_core_err(dev, "Failed to init l2 table %d\n", err);
997 		goto err_rl_cleanup;
998 	}
999 
1000 	err = mlx5_sriov_init(dev);
1001 	if (err) {
1002 		mlx5_core_err(dev, "Failed to init sriov %d\n", err);
1003 		goto err_mpfs_cleanup;
1004 	}
1005 
1006 	err = mlx5_eswitch_init(dev);
1007 	if (err) {
1008 		mlx5_core_err(dev, "Failed to init eswitch %d\n", err);
1009 		goto err_sriov_cleanup;
1010 	}
1011 
1012 	err = mlx5_fpga_init(dev);
1013 	if (err) {
1014 		mlx5_core_err(dev, "Failed to init fpga device %d\n", err);
1015 		goto err_eswitch_cleanup;
1016 	}
1017 
1018 	err = mlx5_vhca_event_init(dev);
1019 	if (err) {
1020 		mlx5_core_err(dev, "Failed to init vhca event notifier %d\n", err);
1021 		goto err_fpga_cleanup;
1022 	}
1023 
1024 	err = mlx5_sf_hw_table_init(dev);
1025 	if (err) {
1026 		mlx5_core_err(dev, "Failed to init SF HW table %d\n", err);
1027 		goto err_sf_hw_table_cleanup;
1028 	}
1029 
1030 	err = mlx5_sf_table_init(dev);
1031 	if (err) {
1032 		mlx5_core_err(dev, "Failed to init SF table %d\n", err);
1033 		goto err_sf_table_cleanup;
1034 	}
1035 
1036 	err = mlx5_fs_core_alloc(dev);
1037 	if (err) {
1038 		mlx5_core_err(dev, "Failed to alloc flow steering\n");
1039 		goto err_fs;
1040 	}
1041 
1042 	dev->dm = mlx5_dm_create(dev);
1043 	if (IS_ERR(dev->dm))
1044 		mlx5_core_warn(dev, "Failed to init device memory%d\n", err);
1045 
1046 	dev->tracer = mlx5_fw_tracer_create(dev);
1047 	dev->hv_vhca = mlx5_hv_vhca_create(dev);
1048 	dev->rsc_dump = mlx5_rsc_dump_create(dev);
1049 
1050 	return 0;
1051 
1052 err_fs:
1053 	mlx5_sf_table_cleanup(dev);
1054 err_sf_table_cleanup:
1055 	mlx5_sf_hw_table_cleanup(dev);
1056 err_sf_hw_table_cleanup:
1057 	mlx5_vhca_event_cleanup(dev);
1058 err_fpga_cleanup:
1059 	mlx5_fpga_cleanup(dev);
1060 err_eswitch_cleanup:
1061 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1062 err_sriov_cleanup:
1063 	mlx5_sriov_cleanup(dev);
1064 err_mpfs_cleanup:
1065 	mlx5_mpfs_cleanup(dev);
1066 err_rl_cleanup:
1067 	mlx5_cleanup_rl_table(dev);
1068 err_tables_cleanup:
1069 	mlx5_geneve_destroy(dev->geneve);
1070 	mlx5_vxlan_destroy(dev->vxlan);
1071 	mlx5_cleanup_clock(dev);
1072 	mlx5_cleanup_reserved_gids(dev);
1073 	mlx5_cq_debugfs_cleanup(dev);
1074 	mlx5_fw_reset_cleanup(dev);
1075 err_events_cleanup:
1076 	mlx5_events_cleanup(dev);
1077 err_eq_cleanup:
1078 	mlx5_eq_table_cleanup(dev);
1079 err_irq_cleanup:
1080 	mlx5_irq_table_cleanup(dev);
1081 err_devcom:
1082 	mlx5_devcom_unregister_device(dev->priv.devcom);
1083 
1084 	return err;
1085 }
1086 
mlx5_cleanup_once(struct mlx5_core_dev * dev)1087 static void mlx5_cleanup_once(struct mlx5_core_dev *dev)
1088 {
1089 	mlx5_rsc_dump_destroy(dev);
1090 	mlx5_hv_vhca_destroy(dev->hv_vhca);
1091 	mlx5_fw_tracer_destroy(dev->tracer);
1092 	mlx5_dm_cleanup(dev);
1093 	mlx5_fs_core_free(dev);
1094 	mlx5_sf_table_cleanup(dev);
1095 	mlx5_sf_hw_table_cleanup(dev);
1096 	mlx5_vhca_event_cleanup(dev);
1097 	mlx5_fpga_cleanup(dev);
1098 	mlx5_eswitch_cleanup(dev->priv.eswitch);
1099 	mlx5_sriov_cleanup(dev);
1100 	mlx5_mpfs_cleanup(dev);
1101 	mlx5_cleanup_rl_table(dev);
1102 	mlx5_geneve_destroy(dev->geneve);
1103 	mlx5_vxlan_destroy(dev->vxlan);
1104 	mlx5_cleanup_clock(dev);
1105 	mlx5_cleanup_reserved_gids(dev);
1106 	mlx5_cq_debugfs_cleanup(dev);
1107 	mlx5_fw_reset_cleanup(dev);
1108 	mlx5_events_cleanup(dev);
1109 	mlx5_eq_table_cleanup(dev);
1110 	mlx5_irq_table_cleanup(dev);
1111 	mlx5_devcom_unregister_device(dev->priv.devcom);
1112 }
1113 
mlx5_function_setup(struct mlx5_core_dev * dev,bool boot,u64 timeout)1114 static int mlx5_function_setup(struct mlx5_core_dev *dev, bool boot, u64 timeout)
1115 {
1116 	int err;
1117 
1118 	mlx5_core_info(dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev),
1119 		       fw_rev_min(dev), fw_rev_sub(dev));
1120 
1121 	/* Only PFs hold the relevant PCIe information for this query */
1122 	if (mlx5_core_is_pf(dev))
1123 		pcie_print_link_status(dev->pdev);
1124 
1125 	/* wait for firmware to accept initialization segments configurations
1126 	 */
1127 	err = wait_fw_init(dev, timeout,
1128 			   mlx5_tout_ms(dev, FW_PRE_INIT_WARN_MESSAGE_INTERVAL));
1129 	if (err) {
1130 		mlx5_core_err(dev, "Firmware over %llu MS in pre-initializing state, aborting\n",
1131 			      timeout);
1132 		return err;
1133 	}
1134 
1135 	err = mlx5_cmd_init(dev);
1136 	if (err) {
1137 		mlx5_core_err(dev, "Failed initializing command interface, aborting\n");
1138 		return err;
1139 	}
1140 
1141 	mlx5_tout_query_iseg(dev);
1142 
1143 	err = wait_fw_init(dev, mlx5_tout_ms(dev, FW_INIT), 0);
1144 	if (err) {
1145 		mlx5_core_err(dev, "Firmware over %llu MS in initializing state, aborting\n",
1146 			      mlx5_tout_ms(dev, FW_INIT));
1147 		goto err_cmd_cleanup;
1148 	}
1149 
1150 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_UP);
1151 
1152 	mlx5_start_health_poll(dev);
1153 
1154 	err = mlx5_core_enable_hca(dev, 0);
1155 	if (err) {
1156 		mlx5_core_err(dev, "enable hca failed\n");
1157 		goto stop_health_poll;
1158 	}
1159 
1160 	err = mlx5_core_set_issi(dev);
1161 	if (err) {
1162 		mlx5_core_err(dev, "failed to set issi\n");
1163 		goto err_disable_hca;
1164 	}
1165 
1166 	err = mlx5_satisfy_startup_pages(dev, 1);
1167 	if (err) {
1168 		mlx5_core_err(dev, "failed to allocate boot pages\n");
1169 		goto err_disable_hca;
1170 	}
1171 
1172 	err = mlx5_tout_query_dtor(dev);
1173 	if (err) {
1174 		mlx5_core_err(dev, "failed to read dtor\n");
1175 		goto reclaim_boot_pages;
1176 	}
1177 
1178 	err = set_hca_ctrl(dev);
1179 	if (err) {
1180 		mlx5_core_err(dev, "set_hca_ctrl failed\n");
1181 		goto reclaim_boot_pages;
1182 	}
1183 
1184 	err = set_hca_cap(dev);
1185 	if (err) {
1186 		mlx5_core_err(dev, "set_hca_cap failed\n");
1187 		goto reclaim_boot_pages;
1188 	}
1189 
1190 	err = mlx5_satisfy_startup_pages(dev, 0);
1191 	if (err) {
1192 		mlx5_core_err(dev, "failed to allocate init pages\n");
1193 		goto reclaim_boot_pages;
1194 	}
1195 
1196 	err = mlx5_cmd_init_hca(dev, sw_owner_id);
1197 	if (err) {
1198 		mlx5_core_err(dev, "init hca failed\n");
1199 		goto reclaim_boot_pages;
1200 	}
1201 
1202 	mlx5_set_driver_version(dev);
1203 
1204 	err = mlx5_query_hca_caps(dev);
1205 	if (err) {
1206 		mlx5_core_err(dev, "query hca failed\n");
1207 		goto reclaim_boot_pages;
1208 	}
1209 	mlx5_start_health_fw_log_up(dev);
1210 
1211 	return 0;
1212 
1213 reclaim_boot_pages:
1214 	mlx5_reclaim_startup_pages(dev);
1215 err_disable_hca:
1216 	mlx5_core_disable_hca(dev, 0);
1217 stop_health_poll:
1218 	mlx5_stop_health_poll(dev, boot);
1219 err_cmd_cleanup:
1220 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1221 	mlx5_cmd_cleanup(dev);
1222 
1223 	return err;
1224 }
1225 
mlx5_function_teardown(struct mlx5_core_dev * dev,bool boot)1226 static int mlx5_function_teardown(struct mlx5_core_dev *dev, bool boot)
1227 {
1228 	int err;
1229 
1230 	err = mlx5_cmd_teardown_hca(dev);
1231 	if (err) {
1232 		mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n");
1233 		return err;
1234 	}
1235 	mlx5_reclaim_startup_pages(dev);
1236 	mlx5_core_disable_hca(dev, 0);
1237 	mlx5_stop_health_poll(dev, boot);
1238 	mlx5_cmd_set_state(dev, MLX5_CMDIF_STATE_DOWN);
1239 	mlx5_cmd_cleanup(dev);
1240 
1241 	return 0;
1242 }
1243 
mlx5_load(struct mlx5_core_dev * dev)1244 static int mlx5_load(struct mlx5_core_dev *dev)
1245 {
1246 	int err;
1247 
1248 	dev->priv.uar = mlx5_get_uars_page(dev);
1249 	if (IS_ERR(dev->priv.uar)) {
1250 		mlx5_core_err(dev, "Failed allocating uar, aborting\n");
1251 		err = PTR_ERR(dev->priv.uar);
1252 		return err;
1253 	}
1254 
1255 	mlx5_events_start(dev);
1256 	mlx5_pagealloc_start(dev);
1257 
1258 	err = mlx5_irq_table_create(dev);
1259 	if (err) {
1260 		mlx5_core_err(dev, "Failed to alloc IRQs\n");
1261 		goto err_irq_table;
1262 	}
1263 
1264 	err = mlx5_eq_table_create(dev);
1265 	if (err) {
1266 		mlx5_core_err(dev, "Failed to create EQs\n");
1267 		goto err_eq_table;
1268 	}
1269 
1270 	err = mlx5_fw_tracer_init(dev->tracer);
1271 	if (err) {
1272 		mlx5_core_err(dev, "Failed to init FW tracer %d\n", err);
1273 		mlx5_fw_tracer_destroy(dev->tracer);
1274 		dev->tracer = NULL;
1275 	}
1276 
1277 	mlx5_fw_reset_events_start(dev);
1278 	mlx5_hv_vhca_init(dev->hv_vhca);
1279 
1280 	err = mlx5_rsc_dump_init(dev);
1281 	if (err) {
1282 		mlx5_core_err(dev, "Failed to init Resource dump %d\n", err);
1283 		mlx5_rsc_dump_destroy(dev);
1284 		dev->rsc_dump = NULL;
1285 	}
1286 
1287 	err = mlx5_fpga_device_start(dev);
1288 	if (err) {
1289 		mlx5_core_err(dev, "fpga device start failed %d\n", err);
1290 		goto err_fpga_start;
1291 	}
1292 
1293 	err = mlx5_fs_core_init(dev);
1294 	if (err) {
1295 		mlx5_core_err(dev, "Failed to init flow steering\n");
1296 		goto err_fs;
1297 	}
1298 
1299 	err = mlx5_core_set_hca_defaults(dev);
1300 	if (err) {
1301 		mlx5_core_err(dev, "Failed to set hca defaults\n");
1302 		goto err_set_hca;
1303 	}
1304 
1305 	mlx5_vhca_event_start(dev);
1306 
1307 	err = mlx5_sf_hw_table_create(dev);
1308 	if (err) {
1309 		mlx5_core_err(dev, "sf table create failed %d\n", err);
1310 		goto err_vhca;
1311 	}
1312 
1313 	err = mlx5_ec_init(dev);
1314 	if (err) {
1315 		mlx5_core_err(dev, "Failed to init embedded CPU\n");
1316 		goto err_ec;
1317 	}
1318 
1319 	mlx5_lag_add_mdev(dev);
1320 	err = mlx5_sriov_attach(dev);
1321 	if (err) {
1322 		mlx5_core_err(dev, "sriov init failed %d\n", err);
1323 		goto err_sriov;
1324 	}
1325 
1326 	mlx5_sf_dev_table_create(dev);
1327 
1328 	err = mlx5_devlink_traps_register(priv_to_devlink(dev));
1329 	if (err)
1330 		goto err_traps_reg;
1331 
1332 	return 0;
1333 
1334 err_traps_reg:
1335 	mlx5_sf_dev_table_destroy(dev);
1336 	mlx5_sriov_detach(dev);
1337 err_sriov:
1338 	mlx5_lag_remove_mdev(dev);
1339 	mlx5_ec_cleanup(dev);
1340 err_ec:
1341 	mlx5_sf_hw_table_destroy(dev);
1342 err_vhca:
1343 	mlx5_vhca_event_stop(dev);
1344 err_set_hca:
1345 	mlx5_fs_core_cleanup(dev);
1346 err_fs:
1347 	mlx5_fpga_device_stop(dev);
1348 err_fpga_start:
1349 	mlx5_rsc_dump_cleanup(dev);
1350 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1351 	mlx5_fw_reset_events_stop(dev);
1352 	mlx5_fw_tracer_cleanup(dev->tracer);
1353 	mlx5_eq_table_destroy(dev);
1354 err_eq_table:
1355 	mlx5_irq_table_destroy(dev);
1356 err_irq_table:
1357 	mlx5_pagealloc_stop(dev);
1358 	mlx5_events_stop(dev);
1359 	mlx5_put_uars_page(dev, dev->priv.uar);
1360 	return err;
1361 }
1362 
mlx5_unload(struct mlx5_core_dev * dev)1363 static void mlx5_unload(struct mlx5_core_dev *dev)
1364 {
1365 	mlx5_devlink_traps_unregister(priv_to_devlink(dev));
1366 	mlx5_sf_dev_table_destroy(dev);
1367 	mlx5_sriov_detach(dev);
1368 	mlx5_eswitch_disable(dev->priv.eswitch);
1369 	mlx5_lag_remove_mdev(dev);
1370 	mlx5_ec_cleanup(dev);
1371 	mlx5_sf_hw_table_destroy(dev);
1372 	mlx5_vhca_event_stop(dev);
1373 	mlx5_fs_core_cleanup(dev);
1374 	mlx5_fpga_device_stop(dev);
1375 	mlx5_rsc_dump_cleanup(dev);
1376 	mlx5_hv_vhca_cleanup(dev->hv_vhca);
1377 	mlx5_fw_reset_events_stop(dev);
1378 	mlx5_fw_tracer_cleanup(dev->tracer);
1379 	mlx5_eq_table_destroy(dev);
1380 	mlx5_irq_table_destroy(dev);
1381 	mlx5_pagealloc_stop(dev);
1382 	mlx5_events_stop(dev);
1383 	mlx5_put_uars_page(dev, dev->priv.uar);
1384 }
1385 
mlx5_init_one(struct mlx5_core_dev * dev)1386 int mlx5_init_one(struct mlx5_core_dev *dev)
1387 {
1388 	struct devlink *devlink = priv_to_devlink(dev);
1389 	int err = 0;
1390 
1391 	devl_lock(devlink);
1392 	mutex_lock(&dev->intf_state_mutex);
1393 	dev->state = MLX5_DEVICE_STATE_UP;
1394 
1395 	err = mlx5_function_setup(dev, true, mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT));
1396 	if (err)
1397 		goto err_function;
1398 
1399 	err = mlx5_init_once(dev);
1400 	if (err) {
1401 		mlx5_core_err(dev, "sw objs init failed\n");
1402 		goto function_teardown;
1403 	}
1404 
1405 	err = mlx5_load(dev);
1406 	if (err)
1407 		goto err_load;
1408 
1409 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1410 
1411 	err = mlx5_devlink_params_register(priv_to_devlink(dev));
1412 	if (err)
1413 		goto err_devlink_params_reg;
1414 
1415 	err = mlx5_register_device(dev);
1416 	if (err)
1417 		goto err_register;
1418 
1419 	mutex_unlock(&dev->intf_state_mutex);
1420 	devl_unlock(devlink);
1421 	return 0;
1422 
1423 err_register:
1424 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1425 err_devlink_params_reg:
1426 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1427 	mlx5_unload(dev);
1428 err_load:
1429 	mlx5_cleanup_once(dev);
1430 function_teardown:
1431 	mlx5_function_teardown(dev, true);
1432 err_function:
1433 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1434 	mutex_unlock(&dev->intf_state_mutex);
1435 	devl_unlock(devlink);
1436 	return err;
1437 }
1438 
mlx5_uninit_one(struct mlx5_core_dev * dev)1439 void mlx5_uninit_one(struct mlx5_core_dev *dev)
1440 {
1441 	struct devlink *devlink = priv_to_devlink(dev);
1442 
1443 	devl_lock(devlink);
1444 	mutex_lock(&dev->intf_state_mutex);
1445 
1446 	mlx5_unregister_device(dev);
1447 	mlx5_devlink_params_unregister(priv_to_devlink(dev));
1448 
1449 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1450 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1451 			       __func__);
1452 		mlx5_cleanup_once(dev);
1453 		goto out;
1454 	}
1455 
1456 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1457 	mlx5_unload(dev);
1458 	mlx5_cleanup_once(dev);
1459 	mlx5_function_teardown(dev, true);
1460 out:
1461 	mutex_unlock(&dev->intf_state_mutex);
1462 	devl_unlock(devlink);
1463 }
1464 
mlx5_load_one_devl_locked(struct mlx5_core_dev * dev,bool recovery)1465 int mlx5_load_one_devl_locked(struct mlx5_core_dev *dev, bool recovery)
1466 {
1467 	int err = 0;
1468 	u64 timeout;
1469 
1470 	devl_assert_locked(priv_to_devlink(dev));
1471 	mutex_lock(&dev->intf_state_mutex);
1472 	if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1473 		mlx5_core_warn(dev, "interface is up, NOP\n");
1474 		goto out;
1475 	}
1476 	/* remove any previous indication of internal error */
1477 	dev->state = MLX5_DEVICE_STATE_UP;
1478 
1479 	if (recovery)
1480 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_ON_RECOVERY_TIMEOUT);
1481 	else
1482 		timeout = mlx5_tout_ms(dev, FW_PRE_INIT_TIMEOUT);
1483 	err = mlx5_function_setup(dev, false, timeout);
1484 	if (err)
1485 		goto err_function;
1486 
1487 	err = mlx5_load(dev);
1488 	if (err)
1489 		goto err_load;
1490 
1491 	set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1492 
1493 	err = mlx5_attach_device(dev);
1494 	if (err)
1495 		goto err_attach;
1496 
1497 	mutex_unlock(&dev->intf_state_mutex);
1498 	return 0;
1499 
1500 err_attach:
1501 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1502 	mlx5_unload(dev);
1503 err_load:
1504 	mlx5_function_teardown(dev, false);
1505 err_function:
1506 	dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR;
1507 out:
1508 	mutex_unlock(&dev->intf_state_mutex);
1509 	return err;
1510 }
1511 
mlx5_load_one(struct mlx5_core_dev * dev)1512 int mlx5_load_one(struct mlx5_core_dev *dev)
1513 {
1514 	struct devlink *devlink = priv_to_devlink(dev);
1515 	int ret;
1516 
1517 	devl_lock(devlink);
1518 	ret = mlx5_load_one_devl_locked(dev, false);
1519 	devl_unlock(devlink);
1520 	return ret;
1521 }
1522 
mlx5_unload_one_devl_locked(struct mlx5_core_dev * dev,bool suspend)1523 void mlx5_unload_one_devl_locked(struct mlx5_core_dev *dev, bool suspend)
1524 {
1525 	devl_assert_locked(priv_to_devlink(dev));
1526 	mutex_lock(&dev->intf_state_mutex);
1527 
1528 	mlx5_detach_device(dev, suspend);
1529 
1530 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) {
1531 		mlx5_core_warn(dev, "%s: interface is down, NOP\n",
1532 			       __func__);
1533 		goto out;
1534 	}
1535 
1536 	clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state);
1537 	mlx5_unload(dev);
1538 	mlx5_function_teardown(dev, false);
1539 out:
1540 	mutex_unlock(&dev->intf_state_mutex);
1541 }
1542 
mlx5_unload_one(struct mlx5_core_dev * dev,bool suspend)1543 void mlx5_unload_one(struct mlx5_core_dev *dev, bool suspend)
1544 {
1545 	struct devlink *devlink = priv_to_devlink(dev);
1546 
1547 	devl_lock(devlink);
1548 	mlx5_unload_one_devl_locked(dev, suspend);
1549 	devl_unlock(devlink);
1550 }
1551 
1552 static const int types[] = {
1553 	MLX5_CAP_GENERAL,
1554 	MLX5_CAP_GENERAL_2,
1555 	MLX5_CAP_ETHERNET_OFFLOADS,
1556 	MLX5_CAP_IPOIB_ENHANCED_OFFLOADS,
1557 	MLX5_CAP_ODP,
1558 	MLX5_CAP_ATOMIC,
1559 	MLX5_CAP_ROCE,
1560 	MLX5_CAP_IPOIB_OFFLOADS,
1561 	MLX5_CAP_FLOW_TABLE,
1562 	MLX5_CAP_ESWITCH_FLOW_TABLE,
1563 	MLX5_CAP_ESWITCH,
1564 	MLX5_CAP_VECTOR_CALC,
1565 	MLX5_CAP_QOS,
1566 	MLX5_CAP_DEBUG,
1567 	MLX5_CAP_DEV_MEM,
1568 	MLX5_CAP_DEV_EVENT,
1569 	MLX5_CAP_TLS,
1570 	MLX5_CAP_VDPA_EMULATION,
1571 	MLX5_CAP_IPSEC,
1572 	MLX5_CAP_PORT_SELECTION,
1573 	MLX5_CAP_DEV_SHAMPO,
1574 	MLX5_CAP_MACSEC,
1575 	MLX5_CAP_ADV_VIRTUALIZATION,
1576 	MLX5_CAP_CRYPTO,
1577 };
1578 
mlx5_hca_caps_free(struct mlx5_core_dev * dev)1579 static void mlx5_hca_caps_free(struct mlx5_core_dev *dev)
1580 {
1581 	int type;
1582 	int i;
1583 
1584 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1585 		type = types[i];
1586 		kfree(dev->caps.hca[type]);
1587 	}
1588 }
1589 
mlx5_hca_caps_alloc(struct mlx5_core_dev * dev)1590 static int mlx5_hca_caps_alloc(struct mlx5_core_dev *dev)
1591 {
1592 	struct mlx5_hca_cap *cap;
1593 	int type;
1594 	int i;
1595 
1596 	for (i = 0; i < ARRAY_SIZE(types); i++) {
1597 		cap = kzalloc(sizeof(*cap), GFP_KERNEL);
1598 		if (!cap)
1599 			goto err;
1600 		type = types[i];
1601 		dev->caps.hca[type] = cap;
1602 	}
1603 
1604 	return 0;
1605 
1606 err:
1607 	mlx5_hca_caps_free(dev);
1608 	return -ENOMEM;
1609 }
1610 
vhca_id_show(struct seq_file * file,void * priv)1611 static int vhca_id_show(struct seq_file *file, void *priv)
1612 {
1613 	struct mlx5_core_dev *dev = file->private;
1614 
1615 	seq_printf(file, "0x%x\n", MLX5_CAP_GEN(dev, vhca_id));
1616 	return 0;
1617 }
1618 
1619 DEFINE_SHOW_ATTRIBUTE(vhca_id);
1620 
mlx5_mdev_init(struct mlx5_core_dev * dev,int profile_idx)1621 int mlx5_mdev_init(struct mlx5_core_dev *dev, int profile_idx)
1622 {
1623 	struct mlx5_priv *priv = &dev->priv;
1624 	int err;
1625 
1626 	memcpy(&dev->profile, &profile[profile_idx], sizeof(dev->profile));
1627 	lockdep_register_key(&dev->lock_key);
1628 	mutex_init(&dev->intf_state_mutex);
1629 	lockdep_set_class(&dev->intf_state_mutex, &dev->lock_key);
1630 	mutex_init(&dev->mlx5e_res.uplink_netdev_lock);
1631 
1632 	mutex_init(&priv->bfregs.reg_head.lock);
1633 	mutex_init(&priv->bfregs.wc_head.lock);
1634 	INIT_LIST_HEAD(&priv->bfregs.reg_head.list);
1635 	INIT_LIST_HEAD(&priv->bfregs.wc_head.list);
1636 
1637 	mutex_init(&priv->alloc_mutex);
1638 	mutex_init(&priv->pgdir_mutex);
1639 	INIT_LIST_HEAD(&priv->pgdir_list);
1640 
1641 	priv->numa_node = dev_to_node(mlx5_core_dma_dev(dev));
1642 	priv->dbg.dbg_root = debugfs_create_dir(dev_name(dev->device),
1643 						mlx5_debugfs_root);
1644 	debugfs_create_file("vhca_id", 0400, priv->dbg.dbg_root, dev, &vhca_id_fops);
1645 	INIT_LIST_HEAD(&priv->traps);
1646 
1647 	err = mlx5_tout_init(dev);
1648 	if (err) {
1649 		mlx5_core_err(dev, "Failed initializing timeouts, aborting\n");
1650 		goto err_timeout_init;
1651 	}
1652 
1653 	err = mlx5_health_init(dev);
1654 	if (err)
1655 		goto err_health_init;
1656 
1657 	err = mlx5_pagealloc_init(dev);
1658 	if (err)
1659 		goto err_pagealloc_init;
1660 
1661 	err = mlx5_adev_init(dev);
1662 	if (err)
1663 		goto err_adev_init;
1664 
1665 	err = mlx5_hca_caps_alloc(dev);
1666 	if (err)
1667 		goto err_hca_caps;
1668 
1669 	/* The conjunction of sw_vhca_id with sw_owner_id will be a global
1670 	 * unique id per function which uses mlx5_core.
1671 	 * Those values are supplied to FW as part of the init HCA command to
1672 	 * be used by both driver and FW when it's applicable.
1673 	 */
1674 	dev->priv.sw_vhca_id = ida_alloc_range(&sw_vhca_ida, 1,
1675 					       MAX_SW_VHCA_ID,
1676 					       GFP_KERNEL);
1677 	if (dev->priv.sw_vhca_id < 0)
1678 		mlx5_core_err(dev, "failed to allocate sw_vhca_id, err=%d\n",
1679 			      dev->priv.sw_vhca_id);
1680 
1681 	return 0;
1682 
1683 err_hca_caps:
1684 	mlx5_adev_cleanup(dev);
1685 err_adev_init:
1686 	mlx5_pagealloc_cleanup(dev);
1687 err_pagealloc_init:
1688 	mlx5_health_cleanup(dev);
1689 err_health_init:
1690 	mlx5_tout_cleanup(dev);
1691 err_timeout_init:
1692 	debugfs_remove(dev->priv.dbg.dbg_root);
1693 	mutex_destroy(&priv->pgdir_mutex);
1694 	mutex_destroy(&priv->alloc_mutex);
1695 	mutex_destroy(&priv->bfregs.wc_head.lock);
1696 	mutex_destroy(&priv->bfregs.reg_head.lock);
1697 	mutex_destroy(&dev->intf_state_mutex);
1698 	lockdep_unregister_key(&dev->lock_key);
1699 	return err;
1700 }
1701 
mlx5_mdev_uninit(struct mlx5_core_dev * dev)1702 void mlx5_mdev_uninit(struct mlx5_core_dev *dev)
1703 {
1704 	struct mlx5_priv *priv = &dev->priv;
1705 
1706 	if (priv->sw_vhca_id > 0)
1707 		ida_free(&sw_vhca_ida, dev->priv.sw_vhca_id);
1708 
1709 	mlx5_hca_caps_free(dev);
1710 	mlx5_adev_cleanup(dev);
1711 	mlx5_pagealloc_cleanup(dev);
1712 	mlx5_health_cleanup(dev);
1713 	mlx5_tout_cleanup(dev);
1714 	debugfs_remove_recursive(dev->priv.dbg.dbg_root);
1715 	mutex_destroy(&priv->pgdir_mutex);
1716 	mutex_destroy(&priv->alloc_mutex);
1717 	mutex_destroy(&priv->bfregs.wc_head.lock);
1718 	mutex_destroy(&priv->bfregs.reg_head.lock);
1719 	mutex_destroy(&dev->mlx5e_res.uplink_netdev_lock);
1720 	mutex_destroy(&dev->intf_state_mutex);
1721 	lockdep_unregister_key(&dev->lock_key);
1722 }
1723 
probe_one(struct pci_dev * pdev,const struct pci_device_id * id)1724 static int probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
1725 {
1726 	struct mlx5_core_dev *dev;
1727 	struct devlink *devlink;
1728 	int err;
1729 
1730 	devlink = mlx5_devlink_alloc(&pdev->dev);
1731 	if (!devlink) {
1732 		dev_err(&pdev->dev, "devlink alloc failed\n");
1733 		return -ENOMEM;
1734 	}
1735 
1736 	dev = devlink_priv(devlink);
1737 	dev->device = &pdev->dev;
1738 	dev->pdev = pdev;
1739 
1740 	dev->coredev_type = id->driver_data & MLX5_PCI_DEV_IS_VF ?
1741 			 MLX5_COREDEV_VF : MLX5_COREDEV_PF;
1742 
1743 	dev->priv.adev_idx = mlx5_adev_idx_alloc();
1744 	if (dev->priv.adev_idx < 0) {
1745 		err = dev->priv.adev_idx;
1746 		goto adev_init_err;
1747 	}
1748 
1749 	err = mlx5_mdev_init(dev, prof_sel);
1750 	if (err)
1751 		goto mdev_init_err;
1752 
1753 	err = mlx5_pci_init(dev, pdev, id);
1754 	if (err) {
1755 		mlx5_core_err(dev, "mlx5_pci_init failed with error code %d\n",
1756 			      err);
1757 		goto pci_init_err;
1758 	}
1759 
1760 	err = mlx5_init_one(dev);
1761 	if (err) {
1762 		mlx5_core_err(dev, "mlx5_init_one failed with error code %d\n",
1763 			      err);
1764 		goto err_init_one;
1765 	}
1766 
1767 	err = mlx5_crdump_enable(dev);
1768 	if (err)
1769 		dev_err(&pdev->dev, "mlx5_crdump_enable failed with error code %d\n", err);
1770 
1771 	pci_save_state(pdev);
1772 	devlink_register(devlink);
1773 	return 0;
1774 
1775 err_init_one:
1776 	mlx5_pci_close(dev);
1777 pci_init_err:
1778 	mlx5_mdev_uninit(dev);
1779 mdev_init_err:
1780 	mlx5_adev_idx_free(dev->priv.adev_idx);
1781 adev_init_err:
1782 	mlx5_devlink_free(devlink);
1783 
1784 	return err;
1785 }
1786 
remove_one(struct pci_dev * pdev)1787 static void remove_one(struct pci_dev *pdev)
1788 {
1789 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1790 	struct devlink *devlink = priv_to_devlink(dev);
1791 
1792 	/* mlx5_drain_fw_reset() is using devlink APIs. Hence, we must drain
1793 	 * fw_reset before unregistering the devlink.
1794 	 */
1795 	mlx5_drain_fw_reset(dev);
1796 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
1797 	devlink_unregister(devlink);
1798 	mlx5_sriov_disable(pdev);
1799 	mlx5_crdump_disable(dev);
1800 	mlx5_drain_health_wq(dev);
1801 	mlx5_uninit_one(dev);
1802 	mlx5_pci_close(dev);
1803 	mlx5_mdev_uninit(dev);
1804 	mlx5_adev_idx_free(dev->priv.adev_idx);
1805 	mlx5_devlink_free(devlink);
1806 }
1807 
1808 #define mlx5_pci_trace(dev, fmt, ...) ({ \
1809 	struct mlx5_core_dev *__dev = (dev); \
1810 	mlx5_core_info(__dev, "%s Device state = %d health sensors: %d pci_status: %d. " fmt, \
1811 		       __func__, __dev->state, mlx5_health_check_fatal_sensors(__dev), \
1812 		       __dev->pci_status, ##__VA_ARGS__); \
1813 })
1814 
result2str(enum pci_ers_result result)1815 static const char *result2str(enum pci_ers_result result)
1816 {
1817 	return  result == PCI_ERS_RESULT_NEED_RESET ? "need reset" :
1818 		result == PCI_ERS_RESULT_DISCONNECT ? "disconnect" :
1819 		result == PCI_ERS_RESULT_RECOVERED  ? "recovered" :
1820 		"unknown";
1821 }
1822 
mlx5_pci_err_detected(struct pci_dev * pdev,pci_channel_state_t state)1823 static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev,
1824 					      pci_channel_state_t state)
1825 {
1826 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1827 	enum pci_ers_result res;
1828 
1829 	mlx5_pci_trace(dev, "Enter, pci channel state = %d\n", state);
1830 
1831 	mlx5_enter_error_state(dev, false);
1832 	mlx5_error_sw_reset(dev);
1833 	mlx5_unload_one(dev, true);
1834 	mlx5_drain_health_wq(dev);
1835 	mlx5_pci_disable_device(dev);
1836 
1837 	res = state == pci_channel_io_perm_failure ?
1838 		PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET;
1839 
1840 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, result = %d, %s\n",
1841 		       __func__, dev->state, dev->pci_status, res, result2str(res));
1842 	return res;
1843 }
1844 
1845 /* wait for the device to show vital signs by waiting
1846  * for the health counter to start counting.
1847  */
wait_vital(struct pci_dev * pdev)1848 static int wait_vital(struct pci_dev *pdev)
1849 {
1850 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1851 	struct mlx5_core_health *health = &dev->priv.health;
1852 	const int niter = 100;
1853 	u32 last_count = 0;
1854 	u32 count;
1855 	int i;
1856 
1857 	for (i = 0; i < niter; i++) {
1858 		count = ioread32be(health->health_counter);
1859 		if (count && count != 0xffffffff) {
1860 			if (last_count && last_count != count) {
1861 				mlx5_core_info(dev,
1862 					       "wait vital counter value 0x%x after %d iterations\n",
1863 					       count, i);
1864 				return 0;
1865 			}
1866 			last_count = count;
1867 		}
1868 		msleep(50);
1869 	}
1870 
1871 	return -ETIMEDOUT;
1872 }
1873 
mlx5_pci_slot_reset(struct pci_dev * pdev)1874 static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev)
1875 {
1876 	enum pci_ers_result res = PCI_ERS_RESULT_DISCONNECT;
1877 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1878 	int err;
1879 
1880 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Enter\n",
1881 		       __func__, dev->state, dev->pci_status);
1882 
1883 	err = mlx5_pci_enable_device(dev);
1884 	if (err) {
1885 		mlx5_core_err(dev, "%s: mlx5_pci_enable_device failed with error code: %d\n",
1886 			      __func__, err);
1887 		goto out;
1888 	}
1889 
1890 	pci_set_master(pdev);
1891 	pci_restore_state(pdev);
1892 	pci_save_state(pdev);
1893 
1894 	err = wait_vital(pdev);
1895 	if (err) {
1896 		mlx5_core_err(dev, "%s: wait vital failed with error code: %d\n",
1897 			      __func__, err);
1898 		goto out;
1899 	}
1900 
1901 	res = PCI_ERS_RESULT_RECOVERED;
1902 out:
1903 	mlx5_core_info(dev, "%s Device state = %d pci_status: %d. Exit, err = %d, result = %d, %s\n",
1904 		       __func__, dev->state, dev->pci_status, err, res, result2str(res));
1905 	return res;
1906 }
1907 
mlx5_pci_resume(struct pci_dev * pdev)1908 static void mlx5_pci_resume(struct pci_dev *pdev)
1909 {
1910 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1911 	int err;
1912 
1913 	mlx5_pci_trace(dev, "Enter, loading driver..\n");
1914 
1915 	err = mlx5_load_one(dev);
1916 	if (!err)
1917 		devlink_health_reporter_state_update(dev->priv.health.fw_fatal_reporter,
1918 						     DEVLINK_HEALTH_REPORTER_STATE_HEALTHY);
1919 
1920 	mlx5_pci_trace(dev, "Done, err = %d, device %s\n", err,
1921 		       !err ? "recovered" : "Failed");
1922 }
1923 
1924 static const struct pci_error_handlers mlx5_err_handler = {
1925 	.error_detected = mlx5_pci_err_detected,
1926 	.slot_reset	= mlx5_pci_slot_reset,
1927 	.resume		= mlx5_pci_resume
1928 };
1929 
mlx5_try_fast_unload(struct mlx5_core_dev * dev)1930 static int mlx5_try_fast_unload(struct mlx5_core_dev *dev)
1931 {
1932 	bool fast_teardown = false, force_teardown = false;
1933 	int ret = 1;
1934 
1935 	fast_teardown = MLX5_CAP_GEN(dev, fast_teardown);
1936 	force_teardown = MLX5_CAP_GEN(dev, force_teardown);
1937 
1938 	mlx5_core_dbg(dev, "force teardown firmware support=%d\n", force_teardown);
1939 	mlx5_core_dbg(dev, "fast teardown firmware support=%d\n", fast_teardown);
1940 
1941 	if (!fast_teardown && !force_teardown)
1942 		return -EOPNOTSUPP;
1943 
1944 	if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
1945 		mlx5_core_dbg(dev, "Device in internal error state, giving up\n");
1946 		return -EAGAIN;
1947 	}
1948 
1949 	/* Panic tear down fw command will stop the PCI bus communication
1950 	 * with the HCA, so the health poll is no longer needed.
1951 	 */
1952 	mlx5_drain_health_wq(dev);
1953 	mlx5_stop_health_poll(dev, false);
1954 
1955 	ret = mlx5_cmd_fast_teardown_hca(dev);
1956 	if (!ret)
1957 		goto succeed;
1958 
1959 	ret = mlx5_cmd_force_teardown_hca(dev);
1960 	if (!ret)
1961 		goto succeed;
1962 
1963 	mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", ret);
1964 	mlx5_start_health_poll(dev);
1965 	return ret;
1966 
1967 succeed:
1968 	mlx5_enter_error_state(dev, true);
1969 
1970 	/* Some platforms requiring freeing the IRQ's in the shutdown
1971 	 * flow. If they aren't freed they can't be allocated after
1972 	 * kexec. There is no need to cleanup the mlx5_core software
1973 	 * contexts.
1974 	 */
1975 	mlx5_core_eq_free_irqs(dev);
1976 
1977 	return 0;
1978 }
1979 
shutdown(struct pci_dev * pdev)1980 static void shutdown(struct pci_dev *pdev)
1981 {
1982 	struct mlx5_core_dev *dev  = pci_get_drvdata(pdev);
1983 	int err;
1984 
1985 	mlx5_core_info(dev, "Shutdown was called\n");
1986 	set_bit(MLX5_BREAK_FW_WAIT, &dev->intf_state);
1987 	err = mlx5_try_fast_unload(dev);
1988 	if (err)
1989 		mlx5_unload_one(dev, false);
1990 	mlx5_pci_disable_device(dev);
1991 }
1992 
mlx5_suspend(struct pci_dev * pdev,pm_message_t state)1993 static int mlx5_suspend(struct pci_dev *pdev, pm_message_t state)
1994 {
1995 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
1996 
1997 	mlx5_unload_one(dev, true);
1998 
1999 	return 0;
2000 }
2001 
mlx5_resume(struct pci_dev * pdev)2002 static int mlx5_resume(struct pci_dev *pdev)
2003 {
2004 	struct mlx5_core_dev *dev = pci_get_drvdata(pdev);
2005 
2006 	return mlx5_load_one(dev);
2007 }
2008 
2009 static const struct pci_device_id mlx5_core_pci_table[] = {
2010 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTIB) },
2011 	{ PCI_VDEVICE(MELLANOX, 0x1012), MLX5_PCI_DEV_IS_VF},	/* Connect-IB VF */
2012 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4) },
2013 	{ PCI_VDEVICE(MELLANOX, 0x1014), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4 VF */
2014 	{ PCI_VDEVICE(MELLANOX, PCI_DEVICE_ID_MELLANOX_CONNECTX4_LX) },
2015 	{ PCI_VDEVICE(MELLANOX, 0x1016), MLX5_PCI_DEV_IS_VF},	/* ConnectX-4LX VF */
2016 	{ PCI_VDEVICE(MELLANOX, 0x1017) },			/* ConnectX-5, PCIe 3.0 */
2017 	{ PCI_VDEVICE(MELLANOX, 0x1018), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 VF */
2018 	{ PCI_VDEVICE(MELLANOX, 0x1019) },			/* ConnectX-5 Ex */
2019 	{ PCI_VDEVICE(MELLANOX, 0x101a), MLX5_PCI_DEV_IS_VF},	/* ConnectX-5 Ex VF */
2020 	{ PCI_VDEVICE(MELLANOX, 0x101b) },			/* ConnectX-6 */
2021 	{ PCI_VDEVICE(MELLANOX, 0x101c), MLX5_PCI_DEV_IS_VF},	/* ConnectX-6 VF */
2022 	{ PCI_VDEVICE(MELLANOX, 0x101d) },			/* ConnectX-6 Dx */
2023 	{ PCI_VDEVICE(MELLANOX, 0x101e), MLX5_PCI_DEV_IS_VF},	/* ConnectX Family mlx5Gen Virtual Function */
2024 	{ PCI_VDEVICE(MELLANOX, 0x101f) },			/* ConnectX-6 LX */
2025 	{ PCI_VDEVICE(MELLANOX, 0x1021) },			/* ConnectX-7 */
2026 	{ PCI_VDEVICE(MELLANOX, 0x1023) },			/* ConnectX-8 */
2027 	{ PCI_VDEVICE(MELLANOX, 0xa2d2) },			/* BlueField integrated ConnectX-5 network controller */
2028 	{ PCI_VDEVICE(MELLANOX, 0xa2d3), MLX5_PCI_DEV_IS_VF},	/* BlueField integrated ConnectX-5 network controller VF */
2029 	{ PCI_VDEVICE(MELLANOX, 0xa2d6) },			/* BlueField-2 integrated ConnectX-6 Dx network controller */
2030 	{ PCI_VDEVICE(MELLANOX, 0xa2dc) },			/* BlueField-3 integrated ConnectX-7 network controller */
2031 	{ PCI_VDEVICE(MELLANOX, 0xa2df) },			/* BlueField-4 integrated ConnectX-8 network controller */
2032 	{ 0, }
2033 };
2034 
2035 MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table);
2036 
mlx5_disable_device(struct mlx5_core_dev * dev)2037 void mlx5_disable_device(struct mlx5_core_dev *dev)
2038 {
2039 	mlx5_error_sw_reset(dev);
2040 	mlx5_unload_one_devl_locked(dev, false);
2041 }
2042 
mlx5_recover_device(struct mlx5_core_dev * dev)2043 int mlx5_recover_device(struct mlx5_core_dev *dev)
2044 {
2045 	if (!mlx5_core_is_sf(dev)) {
2046 		mlx5_pci_disable_device(dev);
2047 		if (mlx5_pci_slot_reset(dev->pdev) != PCI_ERS_RESULT_RECOVERED)
2048 			return -EIO;
2049 	}
2050 
2051 	return mlx5_load_one_devl_locked(dev, true);
2052 }
2053 
2054 static struct pci_driver mlx5_core_driver = {
2055 	.name           = KBUILD_MODNAME,
2056 	.id_table       = mlx5_core_pci_table,
2057 	.probe          = probe_one,
2058 	.remove         = remove_one,
2059 	.suspend        = mlx5_suspend,
2060 	.resume         = mlx5_resume,
2061 	.shutdown	= shutdown,
2062 	.err_handler	= &mlx5_err_handler,
2063 	.sriov_configure   = mlx5_core_sriov_configure,
2064 	.sriov_get_vf_total_msix = mlx5_sriov_get_vf_total_msix,
2065 	.sriov_set_msix_vec_count = mlx5_core_sriov_set_msix_vec_count,
2066 };
2067 
2068 /**
2069  * mlx5_vf_get_core_dev - Get the mlx5 core device from a given VF PCI device if
2070  *                     mlx5_core is its driver.
2071  * @pdev: The associated PCI device.
2072  *
2073  * Upon return the interface state lock stay held to let caller uses it safely.
2074  * Caller must ensure to use the returned mlx5 device for a narrow window
2075  * and put it back with mlx5_vf_put_core_dev() immediately once usage was over.
2076  *
2077  * Return: Pointer to the associated mlx5_core_dev or NULL.
2078  */
mlx5_vf_get_core_dev(struct pci_dev * pdev)2079 struct mlx5_core_dev *mlx5_vf_get_core_dev(struct pci_dev *pdev)
2080 {
2081 	struct mlx5_core_dev *mdev;
2082 
2083 	mdev = pci_iov_get_pf_drvdata(pdev, &mlx5_core_driver);
2084 	if (IS_ERR(mdev))
2085 		return NULL;
2086 
2087 	mutex_lock(&mdev->intf_state_mutex);
2088 	if (!test_bit(MLX5_INTERFACE_STATE_UP, &mdev->intf_state)) {
2089 		mutex_unlock(&mdev->intf_state_mutex);
2090 		return NULL;
2091 	}
2092 
2093 	return mdev;
2094 }
2095 EXPORT_SYMBOL(mlx5_vf_get_core_dev);
2096 
2097 /**
2098  * mlx5_vf_put_core_dev - Put the mlx5 core device back.
2099  * @mdev: The mlx5 core device.
2100  *
2101  * Upon return the interface state lock is unlocked and caller should not
2102  * access the mdev any more.
2103  */
mlx5_vf_put_core_dev(struct mlx5_core_dev * mdev)2104 void mlx5_vf_put_core_dev(struct mlx5_core_dev *mdev)
2105 {
2106 	mutex_unlock(&mdev->intf_state_mutex);
2107 }
2108 EXPORT_SYMBOL(mlx5_vf_put_core_dev);
2109 
mlx5_core_verify_params(void)2110 static void mlx5_core_verify_params(void)
2111 {
2112 	if (prof_sel >= ARRAY_SIZE(profile)) {
2113 		pr_warn("mlx5_core: WARNING: Invalid module parameter prof_sel %d, valid range 0-%zu, changing back to default(%d)\n",
2114 			prof_sel,
2115 			ARRAY_SIZE(profile) - 1,
2116 			MLX5_DEFAULT_PROF);
2117 		prof_sel = MLX5_DEFAULT_PROF;
2118 	}
2119 }
2120 
mlx5_init(void)2121 static int __init mlx5_init(void)
2122 {
2123 	int err;
2124 
2125 	WARN_ONCE(strcmp(MLX5_ADEV_NAME, KBUILD_MODNAME),
2126 		  "mlx5_core name not in sync with kernel module name");
2127 
2128 	get_random_bytes(&sw_owner_id, sizeof(sw_owner_id));
2129 
2130 	mlx5_core_verify_params();
2131 	mlx5_register_debugfs();
2132 
2133 	err = mlx5e_init();
2134 	if (err)
2135 		goto err_debug;
2136 
2137 	err = mlx5_sf_driver_register();
2138 	if (err)
2139 		goto err_sf;
2140 
2141 	err = pci_register_driver(&mlx5_core_driver);
2142 	if (err)
2143 		goto err_pci;
2144 
2145 	return 0;
2146 
2147 err_pci:
2148 	mlx5_sf_driver_unregister();
2149 err_sf:
2150 	mlx5e_cleanup();
2151 err_debug:
2152 	mlx5_unregister_debugfs();
2153 	return err;
2154 }
2155 
mlx5_cleanup(void)2156 static void __exit mlx5_cleanup(void)
2157 {
2158 	pci_unregister_driver(&mlx5_core_driver);
2159 	mlx5_sf_driver_unregister();
2160 	mlx5e_cleanup();
2161 	mlx5_unregister_debugfs();
2162 }
2163 
2164 module_init(mlx5_init);
2165 module_exit(mlx5_cleanup);
2166