xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/dev.c (revision bc4c2f2e)
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/mlx5/driver.h>
34 #include <linux/mlx5/eswitch.h>
35 #include <linux/mlx5/mlx5_ifc_vdpa.h>
36 #include <linux/mlx5/vport.h>
37 #include "mlx5_core.h"
38 
39 /* intf dev list mutex */
40 static DEFINE_MUTEX(mlx5_intf_mutex);
41 static DEFINE_IDA(mlx5_adev_ida);
42 
43 static bool is_eth_rep_supported(struct mlx5_core_dev *dev)
44 {
45 	if (!IS_ENABLED(CONFIG_MLX5_ESWITCH))
46 		return false;
47 
48 	if (!MLX5_ESWITCH_MANAGER(dev))
49 		return false;
50 
51 	if (!is_mdev_switchdev_mode(dev))
52 		return false;
53 
54 	return true;
55 }
56 
57 bool mlx5_eth_supported(struct mlx5_core_dev *dev)
58 {
59 	if (!IS_ENABLED(CONFIG_MLX5_CORE_EN))
60 		return false;
61 
62 	if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
63 		return false;
64 
65 	if (!MLX5_CAP_GEN(dev, eth_net_offloads)) {
66 		mlx5_core_warn(dev, "Missing eth_net_offloads capability\n");
67 		return false;
68 	}
69 
70 	if (!MLX5_CAP_GEN(dev, nic_flow_table)) {
71 		mlx5_core_warn(dev, "Missing nic_flow_table capability\n");
72 		return false;
73 	}
74 
75 	if (!MLX5_CAP_ETH(dev, csum_cap)) {
76 		mlx5_core_warn(dev, "Missing csum_cap capability\n");
77 		return false;
78 	}
79 
80 	if (!MLX5_CAP_ETH(dev, max_lso_cap)) {
81 		mlx5_core_warn(dev, "Missing max_lso_cap capability\n");
82 		return false;
83 	}
84 
85 	if (!MLX5_CAP_ETH(dev, vlan_cap)) {
86 		mlx5_core_warn(dev, "Missing vlan_cap capability\n");
87 		return false;
88 	}
89 
90 	if (!MLX5_CAP_ETH(dev, rss_ind_tbl_cap)) {
91 		mlx5_core_warn(dev, "Missing rss_ind_tbl_cap capability\n");
92 		return false;
93 	}
94 
95 	if (MLX5_CAP_FLOWTABLE(dev,
96 			       flow_table_properties_nic_receive.max_ft_level) < 3) {
97 		mlx5_core_warn(dev, "max_ft_level < 3\n");
98 		return false;
99 	}
100 
101 	if (!MLX5_CAP_ETH(dev, self_lb_en_modifiable))
102 		mlx5_core_warn(dev, "Self loop back prevention is not supported\n");
103 	if (!MLX5_CAP_GEN(dev, cq_moderation))
104 		mlx5_core_warn(dev, "CQ moderation is not supported\n");
105 
106 	return true;
107 }
108 
109 static bool is_eth_enabled(struct mlx5_core_dev *dev)
110 {
111 	union devlink_param_value val;
112 	int err;
113 
114 	err = devlink_param_driverinit_value_get(priv_to_devlink(dev),
115 						 DEVLINK_PARAM_GENERIC_ID_ENABLE_ETH,
116 						 &val);
117 	return err ? false : val.vbool;
118 }
119 
120 bool mlx5_vnet_supported(struct mlx5_core_dev *dev)
121 {
122 	if (!IS_ENABLED(CONFIG_MLX5_VDPA_NET))
123 		return false;
124 
125 	if (mlx5_core_is_pf(dev))
126 		return false;
127 
128 	if (!(MLX5_CAP_GEN_64(dev, general_obj_types) &
129 	      MLX5_GENERAL_OBJ_TYPES_CAP_VIRTIO_NET_Q))
130 		return false;
131 
132 	if (!(MLX5_CAP_DEV_VDPA_EMULATION(dev, event_mode) &
133 	      MLX5_VIRTIO_Q_EVENT_MODE_QP_MODE))
134 		return false;
135 
136 	if (!MLX5_CAP_DEV_VDPA_EMULATION(dev, eth_frame_offload_type))
137 		return false;
138 
139 	return true;
140 }
141 
142 static bool is_vnet_enabled(struct mlx5_core_dev *dev)
143 {
144 	union devlink_param_value val;
145 	int err;
146 
147 	err = devlink_param_driverinit_value_get(priv_to_devlink(dev),
148 						 DEVLINK_PARAM_GENERIC_ID_ENABLE_VNET,
149 						 &val);
150 	return err ? false : val.vbool;
151 }
152 
153 static bool is_ib_rep_supported(struct mlx5_core_dev *dev)
154 {
155 	if (!IS_ENABLED(CONFIG_MLX5_INFINIBAND))
156 		return false;
157 
158 	if (dev->priv.flags & MLX5_PRIV_FLAGS_DISABLE_IB_ADEV)
159 		return false;
160 
161 	if (!is_eth_rep_supported(dev))
162 		return false;
163 
164 	if (!MLX5_ESWITCH_MANAGER(dev))
165 		return false;
166 
167 	if (!is_mdev_switchdev_mode(dev))
168 		return false;
169 
170 	if (mlx5_core_mp_enabled(dev))
171 		return false;
172 
173 	return true;
174 }
175 
176 static bool is_mp_supported(struct mlx5_core_dev *dev)
177 {
178 	if (!IS_ENABLED(CONFIG_MLX5_INFINIBAND))
179 		return false;
180 
181 	if (dev->priv.flags & MLX5_PRIV_FLAGS_DISABLE_IB_ADEV)
182 		return false;
183 
184 	if (is_ib_rep_supported(dev))
185 		return false;
186 
187 	if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
188 		return false;
189 
190 	if (!mlx5_core_is_mp_slave(dev))
191 		return false;
192 
193 	return true;
194 }
195 
196 bool mlx5_rdma_supported(struct mlx5_core_dev *dev)
197 {
198 	if (!IS_ENABLED(CONFIG_MLX5_INFINIBAND))
199 		return false;
200 
201 	if (dev->priv.flags & MLX5_PRIV_FLAGS_DISABLE_IB_ADEV)
202 		return false;
203 
204 	if (is_ib_rep_supported(dev))
205 		return false;
206 
207 	if (is_mp_supported(dev))
208 		return false;
209 
210 	return true;
211 }
212 
213 static bool is_ib_enabled(struct mlx5_core_dev *dev)
214 {
215 	union devlink_param_value val;
216 	int err;
217 
218 	err = devlink_param_driverinit_value_get(priv_to_devlink(dev),
219 						 DEVLINK_PARAM_GENERIC_ID_ENABLE_RDMA,
220 						 &val);
221 	return err ? false : val.vbool;
222 }
223 
224 enum {
225 	MLX5_INTERFACE_PROTOCOL_ETH,
226 	MLX5_INTERFACE_PROTOCOL_ETH_REP,
227 
228 	MLX5_INTERFACE_PROTOCOL_IB,
229 	MLX5_INTERFACE_PROTOCOL_IB_REP,
230 	MLX5_INTERFACE_PROTOCOL_MPIB,
231 
232 	MLX5_INTERFACE_PROTOCOL_VNET,
233 };
234 
235 static const struct mlx5_adev_device {
236 	const char *suffix;
237 	bool (*is_supported)(struct mlx5_core_dev *dev);
238 	bool (*is_enabled)(struct mlx5_core_dev *dev);
239 } mlx5_adev_devices[] = {
240 	[MLX5_INTERFACE_PROTOCOL_VNET] = { .suffix = "vnet",
241 					   .is_supported = &mlx5_vnet_supported,
242 					   .is_enabled = &is_vnet_enabled },
243 	[MLX5_INTERFACE_PROTOCOL_IB] = { .suffix = "rdma",
244 					 .is_supported = &mlx5_rdma_supported,
245 					 .is_enabled = &is_ib_enabled },
246 	[MLX5_INTERFACE_PROTOCOL_ETH] = { .suffix = "eth",
247 					  .is_supported = &mlx5_eth_supported,
248 					  .is_enabled = &is_eth_enabled },
249 	[MLX5_INTERFACE_PROTOCOL_ETH_REP] = { .suffix = "eth-rep",
250 					   .is_supported = &is_eth_rep_supported },
251 	[MLX5_INTERFACE_PROTOCOL_IB_REP] = { .suffix = "rdma-rep",
252 					   .is_supported = &is_ib_rep_supported },
253 	[MLX5_INTERFACE_PROTOCOL_MPIB] = { .suffix = "multiport",
254 					   .is_supported = &is_mp_supported },
255 };
256 
257 int mlx5_adev_idx_alloc(void)
258 {
259 	return ida_alloc(&mlx5_adev_ida, GFP_KERNEL);
260 }
261 
262 void mlx5_adev_idx_free(int idx)
263 {
264 	ida_free(&mlx5_adev_ida, idx);
265 }
266 
267 int mlx5_adev_init(struct mlx5_core_dev *dev)
268 {
269 	struct mlx5_priv *priv = &dev->priv;
270 
271 	priv->adev = kcalloc(ARRAY_SIZE(mlx5_adev_devices),
272 			     sizeof(struct mlx5_adev *), GFP_KERNEL);
273 	if (!priv->adev)
274 		return -ENOMEM;
275 
276 	return 0;
277 }
278 
279 void mlx5_adev_cleanup(struct mlx5_core_dev *dev)
280 {
281 	struct mlx5_priv *priv = &dev->priv;
282 
283 	kfree(priv->adev);
284 }
285 
286 static void adev_release(struct device *dev)
287 {
288 	struct mlx5_adev *mlx5_adev =
289 		container_of(dev, struct mlx5_adev, adev.dev);
290 	struct mlx5_priv *priv = &mlx5_adev->mdev->priv;
291 	int idx = mlx5_adev->idx;
292 
293 	kfree(mlx5_adev);
294 	priv->adev[idx] = NULL;
295 }
296 
297 static struct mlx5_adev *add_adev(struct mlx5_core_dev *dev, int idx)
298 {
299 	const char *suffix = mlx5_adev_devices[idx].suffix;
300 	struct auxiliary_device *adev;
301 	struct mlx5_adev *madev;
302 	int ret;
303 
304 	madev = kzalloc(sizeof(*madev), GFP_KERNEL);
305 	if (!madev)
306 		return ERR_PTR(-ENOMEM);
307 
308 	adev = &madev->adev;
309 	adev->id = dev->priv.adev_idx;
310 	adev->name = suffix;
311 	adev->dev.parent = dev->device;
312 	adev->dev.release = adev_release;
313 	madev->mdev = dev;
314 	madev->idx = idx;
315 
316 	ret = auxiliary_device_init(adev);
317 	if (ret) {
318 		kfree(madev);
319 		return ERR_PTR(ret);
320 	}
321 
322 	ret = auxiliary_device_add(adev);
323 	if (ret) {
324 		auxiliary_device_uninit(adev);
325 		return ERR_PTR(ret);
326 	}
327 	return madev;
328 }
329 
330 static void del_adev(struct auxiliary_device *adev)
331 {
332 	auxiliary_device_delete(adev);
333 	auxiliary_device_uninit(adev);
334 }
335 
336 int mlx5_attach_device(struct mlx5_core_dev *dev)
337 {
338 	struct mlx5_priv *priv = &dev->priv;
339 	struct auxiliary_device *adev;
340 	struct auxiliary_driver *adrv;
341 	int ret = 0, i;
342 
343 	mutex_lock(&mlx5_intf_mutex);
344 	priv->flags &= ~MLX5_PRIV_FLAGS_DETACH;
345 	for (i = 0; i < ARRAY_SIZE(mlx5_adev_devices); i++) {
346 		if (!priv->adev[i]) {
347 			bool is_supported = false;
348 
349 			if (mlx5_adev_devices[i].is_enabled) {
350 				bool enabled;
351 
352 				enabled = mlx5_adev_devices[i].is_enabled(dev);
353 				if (!enabled)
354 					continue;
355 			}
356 
357 			if (mlx5_adev_devices[i].is_supported)
358 				is_supported = mlx5_adev_devices[i].is_supported(dev);
359 
360 			if (!is_supported)
361 				continue;
362 
363 			priv->adev[i] = add_adev(dev, i);
364 			if (IS_ERR(priv->adev[i])) {
365 				ret = PTR_ERR(priv->adev[i]);
366 				priv->adev[i] = NULL;
367 			}
368 		} else {
369 			adev = &priv->adev[i]->adev;
370 
371 			/* Pay attention that this is not PCI driver that
372 			 * mlx5_core_dev is connected, but auxiliary driver.
373 			 *
374 			 * Here we can race of module unload with devlink
375 			 * reload, but we don't need to take extra lock because
376 			 * we are holding global mlx5_intf_mutex.
377 			 */
378 			if (!adev->dev.driver)
379 				continue;
380 			adrv = to_auxiliary_drv(adev->dev.driver);
381 
382 			if (adrv->resume)
383 				ret = adrv->resume(adev);
384 		}
385 		if (ret) {
386 			mlx5_core_warn(dev, "Device[%d] (%s) failed to load\n",
387 				       i, mlx5_adev_devices[i].suffix);
388 
389 			break;
390 		}
391 	}
392 	mutex_unlock(&mlx5_intf_mutex);
393 	return ret;
394 }
395 
396 void mlx5_detach_device(struct mlx5_core_dev *dev)
397 {
398 	struct mlx5_priv *priv = &dev->priv;
399 	struct auxiliary_device *adev;
400 	struct auxiliary_driver *adrv;
401 	pm_message_t pm = {};
402 	int i;
403 
404 	mutex_lock(&mlx5_intf_mutex);
405 	for (i = ARRAY_SIZE(mlx5_adev_devices) - 1; i >= 0; i--) {
406 		if (!priv->adev[i])
407 			continue;
408 
409 		if (mlx5_adev_devices[i].is_enabled) {
410 			bool enabled;
411 
412 			enabled = mlx5_adev_devices[i].is_enabled(dev);
413 			if (!enabled)
414 				goto skip_suspend;
415 		}
416 
417 		adev = &priv->adev[i]->adev;
418 		/* Auxiliary driver was unbind manually through sysfs */
419 		if (!adev->dev.driver)
420 			goto skip_suspend;
421 
422 		adrv = to_auxiliary_drv(adev->dev.driver);
423 
424 		if (adrv->suspend) {
425 			adrv->suspend(adev, pm);
426 			continue;
427 		}
428 
429 skip_suspend:
430 		del_adev(&priv->adev[i]->adev);
431 		priv->adev[i] = NULL;
432 	}
433 	priv->flags |= MLX5_PRIV_FLAGS_DETACH;
434 	mutex_unlock(&mlx5_intf_mutex);
435 }
436 
437 int mlx5_register_device(struct mlx5_core_dev *dev)
438 {
439 	int ret;
440 
441 	mutex_lock(&mlx5_intf_mutex);
442 	dev->priv.flags &= ~MLX5_PRIV_FLAGS_DISABLE_ALL_ADEV;
443 	ret = mlx5_rescan_drivers_locked(dev);
444 	mutex_unlock(&mlx5_intf_mutex);
445 	if (ret)
446 		mlx5_unregister_device(dev);
447 
448 	return ret;
449 }
450 
451 void mlx5_unregister_device(struct mlx5_core_dev *dev)
452 {
453 	mutex_lock(&mlx5_intf_mutex);
454 	dev->priv.flags = MLX5_PRIV_FLAGS_DISABLE_ALL_ADEV;
455 	mlx5_rescan_drivers_locked(dev);
456 	mutex_unlock(&mlx5_intf_mutex);
457 }
458 
459 static int add_drivers(struct mlx5_core_dev *dev)
460 {
461 	struct mlx5_priv *priv = &dev->priv;
462 	int i, ret = 0;
463 
464 	for (i = 0; i < ARRAY_SIZE(mlx5_adev_devices); i++) {
465 		bool is_supported = false;
466 
467 		if (priv->adev[i])
468 			continue;
469 
470 		if (mlx5_adev_devices[i].is_supported)
471 			is_supported = mlx5_adev_devices[i].is_supported(dev);
472 
473 		if (!is_supported)
474 			continue;
475 
476 		priv->adev[i] = add_adev(dev, i);
477 		if (IS_ERR(priv->adev[i])) {
478 			mlx5_core_warn(dev, "Device[%d] (%s) failed to load\n",
479 				       i, mlx5_adev_devices[i].suffix);
480 			/* We continue to rescan drivers and leave to the caller
481 			 * to make decision if to release everything or continue.
482 			 */
483 			ret = PTR_ERR(priv->adev[i]);
484 			priv->adev[i] = NULL;
485 		}
486 	}
487 	return ret;
488 }
489 
490 static void delete_drivers(struct mlx5_core_dev *dev)
491 {
492 	struct mlx5_priv *priv = &dev->priv;
493 	bool delete_all;
494 	int i;
495 
496 	delete_all = priv->flags & MLX5_PRIV_FLAGS_DISABLE_ALL_ADEV;
497 
498 	for (i = ARRAY_SIZE(mlx5_adev_devices) - 1; i >= 0; i--) {
499 		bool is_supported = false;
500 
501 		if (!priv->adev[i])
502 			continue;
503 
504 		if (mlx5_adev_devices[i].is_enabled) {
505 			bool enabled;
506 
507 			enabled = mlx5_adev_devices[i].is_enabled(dev);
508 			if (!enabled)
509 				goto del_adev;
510 		}
511 
512 		if (mlx5_adev_devices[i].is_supported && !delete_all)
513 			is_supported = mlx5_adev_devices[i].is_supported(dev);
514 
515 		if (is_supported)
516 			continue;
517 
518 del_adev:
519 		del_adev(&priv->adev[i]->adev);
520 		priv->adev[i] = NULL;
521 	}
522 }
523 
524 /* This function is used after mlx5_core_dev is reconfigured.
525  */
526 int mlx5_rescan_drivers_locked(struct mlx5_core_dev *dev)
527 {
528 	struct mlx5_priv *priv = &dev->priv;
529 
530 	lockdep_assert_held(&mlx5_intf_mutex);
531 	if (priv->flags & MLX5_PRIV_FLAGS_DETACH)
532 		return 0;
533 
534 	delete_drivers(dev);
535 	if (priv->flags & MLX5_PRIV_FLAGS_DISABLE_ALL_ADEV)
536 		return 0;
537 
538 	return add_drivers(dev);
539 }
540 
541 bool mlx5_same_hw_devs(struct mlx5_core_dev *dev, struct mlx5_core_dev *peer_dev)
542 {
543 	u64 fsystem_guid, psystem_guid;
544 
545 	fsystem_guid = mlx5_query_nic_system_image_guid(dev);
546 	psystem_guid = mlx5_query_nic_system_image_guid(peer_dev);
547 
548 	return (fsystem_guid && psystem_guid && fsystem_guid == psystem_guid);
549 }
550 
551 static u32 mlx5_gen_pci_id(const struct mlx5_core_dev *dev)
552 {
553 	return (u32)((pci_domain_nr(dev->pdev->bus) << 16) |
554 		     (dev->pdev->bus->number << 8) |
555 		     PCI_SLOT(dev->pdev->devfn));
556 }
557 
558 static int _next_phys_dev(struct mlx5_core_dev *mdev,
559 			  const struct mlx5_core_dev *curr)
560 {
561 	if (!mlx5_core_is_pf(mdev))
562 		return 0;
563 
564 	if (mdev == curr)
565 		return 0;
566 
567 	if (!mlx5_same_hw_devs(mdev, (struct mlx5_core_dev *)curr) &&
568 	    mlx5_gen_pci_id(mdev) != mlx5_gen_pci_id(curr))
569 		return 0;
570 
571 	return 1;
572 }
573 
574 static int next_phys_dev(struct device *dev, const void *data)
575 {
576 	struct mlx5_adev *madev = container_of(dev, struct mlx5_adev, adev.dev);
577 	struct mlx5_core_dev *mdev = madev->mdev;
578 
579 	return _next_phys_dev(mdev, data);
580 }
581 
582 static int next_phys_dev_lag(struct device *dev, const void *data)
583 {
584 	struct mlx5_adev *madev = container_of(dev, struct mlx5_adev, adev.dev);
585 	struct mlx5_core_dev *mdev = madev->mdev;
586 
587 	if (!MLX5_CAP_GEN(mdev, vport_group_manager) ||
588 	    !MLX5_CAP_GEN(mdev, lag_master) ||
589 	    MLX5_CAP_GEN(mdev, num_lag_ports) != MLX5_MAX_PORTS)
590 		return 0;
591 
592 	return _next_phys_dev(mdev, data);
593 }
594 
595 static struct mlx5_core_dev *mlx5_get_next_dev(struct mlx5_core_dev *dev,
596 					       int (*match)(struct device *dev, const void *data))
597 {
598 	struct auxiliary_device *adev;
599 	struct mlx5_adev *madev;
600 
601 	if (!mlx5_core_is_pf(dev))
602 		return NULL;
603 
604 	adev = auxiliary_find_device(NULL, dev, match);
605 	if (!adev)
606 		return NULL;
607 
608 	madev = container_of(adev, struct mlx5_adev, adev);
609 	put_device(&adev->dev);
610 	return madev->mdev;
611 }
612 
613 /* Must be called with intf_mutex held */
614 struct mlx5_core_dev *mlx5_get_next_phys_dev(struct mlx5_core_dev *dev)
615 {
616 	lockdep_assert_held(&mlx5_intf_mutex);
617 	return mlx5_get_next_dev(dev, &next_phys_dev);
618 }
619 
620 /* Must be called with intf_mutex held */
621 struct mlx5_core_dev *mlx5_get_next_phys_dev_lag(struct mlx5_core_dev *dev)
622 {
623 	lockdep_assert_held(&mlx5_intf_mutex);
624 	return mlx5_get_next_dev(dev, &next_phys_dev_lag);
625 }
626 
627 void mlx5_dev_list_lock(void)
628 {
629 	mutex_lock(&mlx5_intf_mutex);
630 }
631 void mlx5_dev_list_unlock(void)
632 {
633 	mutex_unlock(&mlx5_intf_mutex);
634 }
635 
636 int mlx5_dev_list_trylock(void)
637 {
638 	return mutex_trylock(&mlx5_intf_mutex);
639 }
640