1 /*
2  * Copyright (c) 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/etherdevice.h>
34 #include <linux/debugfs.h>
35 #include <linux/mlx5/driver.h>
36 #include <linux/mlx5/mlx5_ifc.h>
37 #include <linux/mlx5/vport.h>
38 #include <linux/mlx5/fs.h>
39 #include <linux/mlx5/mpfs.h>
40 #include "esw/acl/lgcy.h"
41 #include "esw/legacy.h"
42 #include "esw/qos.h"
43 #include "mlx5_core.h"
44 #include "lib/eq.h"
45 #include "lag/lag.h"
46 #include "eswitch.h"
47 #include "fs_core.h"
48 #include "devlink.h"
49 #include "ecpf.h"
50 #include "en/mod_hdr.h"
51 #include "en_accel/ipsec.h"
52 
53 enum {
54 	MLX5_ACTION_NONE = 0,
55 	MLX5_ACTION_ADD  = 1,
56 	MLX5_ACTION_DEL  = 2,
57 };
58 
59 /* Vport UC/MC hash node */
60 struct vport_addr {
61 	struct l2addr_node     node;
62 	u8                     action;
63 	u16                    vport;
64 	struct mlx5_flow_handle *flow_rule;
65 	bool mpfs; /* UC MAC was added to MPFs */
66 	/* A flag indicating that mac was added due to mc promiscuous vport */
67 	bool mc_promisc;
68 };
69 
70 static int mlx5_eswitch_check(const struct mlx5_core_dev *dev)
71 {
72 	if (MLX5_CAP_GEN(dev, port_type) != MLX5_CAP_PORT_TYPE_ETH)
73 		return -EOPNOTSUPP;
74 
75 	if (!MLX5_ESWITCH_MANAGER(dev))
76 		return -EOPNOTSUPP;
77 
78 	return 0;
79 }
80 
81 static struct mlx5_eswitch *__mlx5_devlink_eswitch_get(struct devlink *devlink, bool check)
82 {
83 	struct mlx5_core_dev *dev = devlink_priv(devlink);
84 	int err;
85 
86 	if (check) {
87 		err = mlx5_eswitch_check(dev);
88 		if (err)
89 			return ERR_PTR(err);
90 	}
91 
92 	return dev->priv.eswitch;
93 }
94 
95 struct mlx5_eswitch *__must_check
96 mlx5_devlink_eswitch_get(struct devlink *devlink)
97 {
98 	return __mlx5_devlink_eswitch_get(devlink, true);
99 }
100 
101 struct mlx5_eswitch *mlx5_devlink_eswitch_nocheck_get(struct devlink *devlink)
102 {
103 	return __mlx5_devlink_eswitch_get(devlink, false);
104 }
105 
106 struct mlx5_vport *__must_check
107 mlx5_eswitch_get_vport(struct mlx5_eswitch *esw, u16 vport_num)
108 {
109 	struct mlx5_vport *vport;
110 
111 	if (!esw)
112 		return ERR_PTR(-EPERM);
113 
114 	vport = xa_load(&esw->vports, vport_num);
115 	if (!vport) {
116 		esw_debug(esw->dev, "vport out of range: num(0x%x)\n", vport_num);
117 		return ERR_PTR(-EINVAL);
118 	}
119 	return vport;
120 }
121 
122 static int arm_vport_context_events_cmd(struct mlx5_core_dev *dev, u16 vport,
123 					u32 events_mask)
124 {
125 	u32 in[MLX5_ST_SZ_DW(modify_nic_vport_context_in)] = {};
126 	void *nic_vport_ctx;
127 
128 	MLX5_SET(modify_nic_vport_context_in, in,
129 		 opcode, MLX5_CMD_OP_MODIFY_NIC_VPORT_CONTEXT);
130 	MLX5_SET(modify_nic_vport_context_in, in, field_select.change_event, 1);
131 	MLX5_SET(modify_nic_vport_context_in, in, vport_number, vport);
132 	if (vport || mlx5_core_is_ecpf(dev))
133 		MLX5_SET(modify_nic_vport_context_in, in, other_vport, 1);
134 	nic_vport_ctx = MLX5_ADDR_OF(modify_nic_vport_context_in,
135 				     in, nic_vport_context);
136 
137 	MLX5_SET(nic_vport_context, nic_vport_ctx, arm_change_event, 1);
138 
139 	if (events_mask & MLX5_VPORT_UC_ADDR_CHANGE)
140 		MLX5_SET(nic_vport_context, nic_vport_ctx,
141 			 event_on_uc_address_change, 1);
142 	if (events_mask & MLX5_VPORT_MC_ADDR_CHANGE)
143 		MLX5_SET(nic_vport_context, nic_vport_ctx,
144 			 event_on_mc_address_change, 1);
145 	if (events_mask & MLX5_VPORT_PROMISC_CHANGE)
146 		MLX5_SET(nic_vport_context, nic_vport_ctx,
147 			 event_on_promisc_change, 1);
148 
149 	return mlx5_cmd_exec_in(dev, modify_nic_vport_context, in);
150 }
151 
152 /* E-Switch vport context HW commands */
153 int mlx5_eswitch_modify_esw_vport_context(struct mlx5_core_dev *dev, u16 vport,
154 					  bool other_vport, void *in)
155 {
156 	MLX5_SET(modify_esw_vport_context_in, in, opcode,
157 		 MLX5_CMD_OP_MODIFY_ESW_VPORT_CONTEXT);
158 	MLX5_SET(modify_esw_vport_context_in, in, vport_number, vport);
159 	MLX5_SET(modify_esw_vport_context_in, in, other_vport, other_vport);
160 	return mlx5_cmd_exec_in(dev, modify_esw_vport_context, in);
161 }
162 
163 static int modify_esw_vport_cvlan(struct mlx5_core_dev *dev, u16 vport,
164 				  u16 vlan, u8 qos, u8 set_flags)
165 {
166 	u32 in[MLX5_ST_SZ_DW(modify_esw_vport_context_in)] = {};
167 
168 	if (!MLX5_CAP_ESW(dev, vport_cvlan_strip) ||
169 	    !MLX5_CAP_ESW(dev, vport_cvlan_insert_if_not_exist))
170 		return -EOPNOTSUPP;
171 
172 	esw_debug(dev, "Set Vport[%d] VLAN %d qos %d set=%x\n",
173 		  vport, vlan, qos, set_flags);
174 
175 	if (set_flags & SET_VLAN_STRIP)
176 		MLX5_SET(modify_esw_vport_context_in, in,
177 			 esw_vport_context.vport_cvlan_strip, 1);
178 
179 	if (set_flags & SET_VLAN_INSERT) {
180 		if (MLX5_CAP_ESW(dev, vport_cvlan_insert_always)) {
181 			/* insert either if vlan exist in packet or not */
182 			MLX5_SET(modify_esw_vport_context_in, in,
183 				 esw_vport_context.vport_cvlan_insert,
184 				 MLX5_VPORT_CVLAN_INSERT_ALWAYS);
185 		} else {
186 			/* insert only if no vlan in packet */
187 			MLX5_SET(modify_esw_vport_context_in, in,
188 				 esw_vport_context.vport_cvlan_insert,
189 				 MLX5_VPORT_CVLAN_INSERT_WHEN_NO_CVLAN);
190 		}
191 		MLX5_SET(modify_esw_vport_context_in, in,
192 			 esw_vport_context.cvlan_pcp, qos);
193 		MLX5_SET(modify_esw_vport_context_in, in,
194 			 esw_vport_context.cvlan_id, vlan);
195 	}
196 
197 	MLX5_SET(modify_esw_vport_context_in, in,
198 		 field_select.vport_cvlan_strip, 1);
199 	MLX5_SET(modify_esw_vport_context_in, in,
200 		 field_select.vport_cvlan_insert, 1);
201 
202 	return mlx5_eswitch_modify_esw_vport_context(dev, vport, true, in);
203 }
204 
205 /* E-Switch FDB */
206 static struct mlx5_flow_handle *
207 __esw_fdb_set_vport_rule(struct mlx5_eswitch *esw, u16 vport, bool rx_rule,
208 			 u8 mac_c[ETH_ALEN], u8 mac_v[ETH_ALEN])
209 {
210 	int match_header = (is_zero_ether_addr(mac_c) ? 0 :
211 			    MLX5_MATCH_OUTER_HEADERS);
212 	struct mlx5_flow_handle *flow_rule = NULL;
213 	struct mlx5_flow_act flow_act = {0};
214 	struct mlx5_flow_destination dest = {};
215 	struct mlx5_flow_spec *spec;
216 	void *mv_misc = NULL;
217 	void *mc_misc = NULL;
218 	u8 *dmac_v = NULL;
219 	u8 *dmac_c = NULL;
220 
221 	if (rx_rule)
222 		match_header |= MLX5_MATCH_MISC_PARAMETERS;
223 
224 	spec = kvzalloc(sizeof(*spec), GFP_KERNEL);
225 	if (!spec)
226 		return NULL;
227 
228 	dmac_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
229 			      outer_headers.dmac_47_16);
230 	dmac_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
231 			      outer_headers.dmac_47_16);
232 
233 	if (match_header & MLX5_MATCH_OUTER_HEADERS) {
234 		ether_addr_copy(dmac_v, mac_v);
235 		ether_addr_copy(dmac_c, mac_c);
236 	}
237 
238 	if (match_header & MLX5_MATCH_MISC_PARAMETERS) {
239 		mv_misc  = MLX5_ADDR_OF(fte_match_param, spec->match_value,
240 					misc_parameters);
241 		mc_misc  = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
242 					misc_parameters);
243 		MLX5_SET(fte_match_set_misc, mv_misc, source_port, MLX5_VPORT_UPLINK);
244 		MLX5_SET_TO_ONES(fte_match_set_misc, mc_misc, source_port);
245 	}
246 
247 	dest.type = MLX5_FLOW_DESTINATION_TYPE_VPORT;
248 	dest.vport.num = vport;
249 
250 	esw_debug(esw->dev,
251 		  "\tFDB add rule dmac_v(%pM) dmac_c(%pM) -> vport(%d)\n",
252 		  dmac_v, dmac_c, vport);
253 	spec->match_criteria_enable = match_header;
254 	flow_act.action =  MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
255 	flow_rule =
256 		mlx5_add_flow_rules(esw->fdb_table.legacy.fdb, spec,
257 				    &flow_act, &dest, 1);
258 	if (IS_ERR(flow_rule)) {
259 		esw_warn(esw->dev,
260 			 "FDB: Failed to add flow rule: dmac_v(%pM) dmac_c(%pM) -> vport(%d), err(%ld)\n",
261 			 dmac_v, dmac_c, vport, PTR_ERR(flow_rule));
262 		flow_rule = NULL;
263 	}
264 
265 	kvfree(spec);
266 	return flow_rule;
267 }
268 
269 static struct mlx5_flow_handle *
270 esw_fdb_set_vport_rule(struct mlx5_eswitch *esw, u8 mac[ETH_ALEN], u16 vport)
271 {
272 	u8 mac_c[ETH_ALEN];
273 
274 	eth_broadcast_addr(mac_c);
275 	return __esw_fdb_set_vport_rule(esw, vport, false, mac_c, mac);
276 }
277 
278 static struct mlx5_flow_handle *
279 esw_fdb_set_vport_allmulti_rule(struct mlx5_eswitch *esw, u16 vport)
280 {
281 	u8 mac_c[ETH_ALEN];
282 	u8 mac_v[ETH_ALEN];
283 
284 	eth_zero_addr(mac_c);
285 	eth_zero_addr(mac_v);
286 	mac_c[0] = 0x01;
287 	mac_v[0] = 0x01;
288 	return __esw_fdb_set_vport_rule(esw, vport, false, mac_c, mac_v);
289 }
290 
291 static struct mlx5_flow_handle *
292 esw_fdb_set_vport_promisc_rule(struct mlx5_eswitch *esw, u16 vport)
293 {
294 	u8 mac_c[ETH_ALEN];
295 	u8 mac_v[ETH_ALEN];
296 
297 	eth_zero_addr(mac_c);
298 	eth_zero_addr(mac_v);
299 	return __esw_fdb_set_vport_rule(esw, vport, true, mac_c, mac_v);
300 }
301 
302 /* E-Switch vport UC/MC lists management */
303 typedef int (*vport_addr_action)(struct mlx5_eswitch *esw,
304 				 struct vport_addr *vaddr);
305 
306 static int esw_add_uc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
307 {
308 	u8 *mac = vaddr->node.addr;
309 	u16 vport = vaddr->vport;
310 	int err;
311 
312 	/* Skip mlx5_mpfs_add_mac for eswitch_managers,
313 	 * it is already done by its netdev in mlx5e_execute_l2_action
314 	 */
315 	if (mlx5_esw_is_manager_vport(esw, vport))
316 		goto fdb_add;
317 
318 	err = mlx5_mpfs_add_mac(esw->dev, mac);
319 	if (err) {
320 		esw_warn(esw->dev,
321 			 "Failed to add L2 table mac(%pM) for vport(0x%x), err(%d)\n",
322 			 mac, vport, err);
323 		return err;
324 	}
325 	vaddr->mpfs = true;
326 
327 fdb_add:
328 	/* SRIOV is enabled: Forward UC MAC to vport */
329 	if (esw->fdb_table.legacy.fdb && esw->mode == MLX5_ESWITCH_LEGACY) {
330 		vaddr->flow_rule = esw_fdb_set_vport_rule(esw, mac, vport);
331 
332 		esw_debug(esw->dev, "\tADDED UC MAC: vport[%d] %pM fr(%p)\n",
333 			  vport, mac, vaddr->flow_rule);
334 	}
335 
336 	return 0;
337 }
338 
339 static int esw_del_uc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
340 {
341 	u8 *mac = vaddr->node.addr;
342 	u16 vport = vaddr->vport;
343 	int err = 0;
344 
345 	/* Skip mlx5_mpfs_del_mac for eswitch managers,
346 	 * it is already done by its netdev in mlx5e_execute_l2_action
347 	 */
348 	if (!vaddr->mpfs || mlx5_esw_is_manager_vport(esw, vport))
349 		goto fdb_del;
350 
351 	err = mlx5_mpfs_del_mac(esw->dev, mac);
352 	if (err)
353 		esw_warn(esw->dev,
354 			 "Failed to del L2 table mac(%pM) for vport(%d), err(%d)\n",
355 			 mac, vport, err);
356 	vaddr->mpfs = false;
357 
358 fdb_del:
359 	if (vaddr->flow_rule)
360 		mlx5_del_flow_rules(vaddr->flow_rule);
361 	vaddr->flow_rule = NULL;
362 
363 	return 0;
364 }
365 
366 static void update_allmulti_vports(struct mlx5_eswitch *esw,
367 				   struct vport_addr *vaddr,
368 				   struct esw_mc_addr *esw_mc)
369 {
370 	u8 *mac = vaddr->node.addr;
371 	struct mlx5_vport *vport;
372 	unsigned long i;
373 	u16 vport_num;
374 
375 	mlx5_esw_for_each_vport(esw, i, vport) {
376 		struct hlist_head *vport_hash = vport->mc_list;
377 		struct vport_addr *iter_vaddr =
378 					l2addr_hash_find(vport_hash,
379 							 mac,
380 							 struct vport_addr);
381 		vport_num = vport->vport;
382 		if (IS_ERR_OR_NULL(vport->allmulti_rule) ||
383 		    vaddr->vport == vport_num)
384 			continue;
385 		switch (vaddr->action) {
386 		case MLX5_ACTION_ADD:
387 			if (iter_vaddr)
388 				continue;
389 			iter_vaddr = l2addr_hash_add(vport_hash, mac,
390 						     struct vport_addr,
391 						     GFP_KERNEL);
392 			if (!iter_vaddr) {
393 				esw_warn(esw->dev,
394 					 "ALL-MULTI: Failed to add MAC(%pM) to vport[%d] DB\n",
395 					 mac, vport_num);
396 				continue;
397 			}
398 			iter_vaddr->vport = vport_num;
399 			iter_vaddr->flow_rule =
400 					esw_fdb_set_vport_rule(esw,
401 							       mac,
402 							       vport_num);
403 			iter_vaddr->mc_promisc = true;
404 			break;
405 		case MLX5_ACTION_DEL:
406 			if (!iter_vaddr)
407 				continue;
408 			mlx5_del_flow_rules(iter_vaddr->flow_rule);
409 			l2addr_hash_del(iter_vaddr);
410 			break;
411 		}
412 	}
413 }
414 
415 static int esw_add_mc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
416 {
417 	struct hlist_head *hash = esw->mc_table;
418 	struct esw_mc_addr *esw_mc;
419 	u8 *mac = vaddr->node.addr;
420 	u16 vport = vaddr->vport;
421 
422 	if (!esw->fdb_table.legacy.fdb)
423 		return 0;
424 
425 	esw_mc = l2addr_hash_find(hash, mac, struct esw_mc_addr);
426 	if (esw_mc)
427 		goto add;
428 
429 	esw_mc = l2addr_hash_add(hash, mac, struct esw_mc_addr, GFP_KERNEL);
430 	if (!esw_mc)
431 		return -ENOMEM;
432 
433 	esw_mc->uplink_rule = /* Forward MC MAC to Uplink */
434 		esw_fdb_set_vport_rule(esw, mac, MLX5_VPORT_UPLINK);
435 
436 	/* Add this multicast mac to all the mc promiscuous vports */
437 	update_allmulti_vports(esw, vaddr, esw_mc);
438 
439 add:
440 	/* If the multicast mac is added as a result of mc promiscuous vport,
441 	 * don't increment the multicast ref count
442 	 */
443 	if (!vaddr->mc_promisc)
444 		esw_mc->refcnt++;
445 
446 	/* Forward MC MAC to vport */
447 	vaddr->flow_rule = esw_fdb_set_vport_rule(esw, mac, vport);
448 	esw_debug(esw->dev,
449 		  "\tADDED MC MAC: vport[%d] %pM fr(%p) refcnt(%d) uplinkfr(%p)\n",
450 		  vport, mac, vaddr->flow_rule,
451 		  esw_mc->refcnt, esw_mc->uplink_rule);
452 	return 0;
453 }
454 
455 static int esw_del_mc_addr(struct mlx5_eswitch *esw, struct vport_addr *vaddr)
456 {
457 	struct hlist_head *hash = esw->mc_table;
458 	struct esw_mc_addr *esw_mc;
459 	u8 *mac = vaddr->node.addr;
460 	u16 vport = vaddr->vport;
461 
462 	if (!esw->fdb_table.legacy.fdb)
463 		return 0;
464 
465 	esw_mc = l2addr_hash_find(hash, mac, struct esw_mc_addr);
466 	if (!esw_mc) {
467 		esw_warn(esw->dev,
468 			 "Failed to find eswitch MC addr for MAC(%pM) vport(%d)",
469 			 mac, vport);
470 		return -EINVAL;
471 	}
472 	esw_debug(esw->dev,
473 		  "\tDELETE MC MAC: vport[%d] %pM fr(%p) refcnt(%d) uplinkfr(%p)\n",
474 		  vport, mac, vaddr->flow_rule, esw_mc->refcnt,
475 		  esw_mc->uplink_rule);
476 
477 	if (vaddr->flow_rule)
478 		mlx5_del_flow_rules(vaddr->flow_rule);
479 	vaddr->flow_rule = NULL;
480 
481 	/* If the multicast mac is added as a result of mc promiscuous vport,
482 	 * don't decrement the multicast ref count.
483 	 */
484 	if (vaddr->mc_promisc || (--esw_mc->refcnt > 0))
485 		return 0;
486 
487 	/* Remove this multicast mac from all the mc promiscuous vports */
488 	update_allmulti_vports(esw, vaddr, esw_mc);
489 
490 	if (esw_mc->uplink_rule)
491 		mlx5_del_flow_rules(esw_mc->uplink_rule);
492 
493 	l2addr_hash_del(esw_mc);
494 	return 0;
495 }
496 
497 /* Apply vport UC/MC list to HW l2 table and FDB table */
498 static void esw_apply_vport_addr_list(struct mlx5_eswitch *esw,
499 				      struct mlx5_vport *vport, int list_type)
500 {
501 	bool is_uc = list_type == MLX5_NVPRT_LIST_TYPE_UC;
502 	vport_addr_action vport_addr_add;
503 	vport_addr_action vport_addr_del;
504 	struct vport_addr *addr;
505 	struct l2addr_node *node;
506 	struct hlist_head *hash;
507 	struct hlist_node *tmp;
508 	int hi;
509 
510 	vport_addr_add = is_uc ? esw_add_uc_addr :
511 				 esw_add_mc_addr;
512 	vport_addr_del = is_uc ? esw_del_uc_addr :
513 				 esw_del_mc_addr;
514 
515 	hash = is_uc ? vport->uc_list : vport->mc_list;
516 	for_each_l2hash_node(node, tmp, hash, hi) {
517 		addr = container_of(node, struct vport_addr, node);
518 		switch (addr->action) {
519 		case MLX5_ACTION_ADD:
520 			vport_addr_add(esw, addr);
521 			addr->action = MLX5_ACTION_NONE;
522 			break;
523 		case MLX5_ACTION_DEL:
524 			vport_addr_del(esw, addr);
525 			l2addr_hash_del(addr);
526 			break;
527 		}
528 	}
529 }
530 
531 /* Sync vport UC/MC list from vport context */
532 static void esw_update_vport_addr_list(struct mlx5_eswitch *esw,
533 				       struct mlx5_vport *vport, int list_type)
534 {
535 	bool is_uc = list_type == MLX5_NVPRT_LIST_TYPE_UC;
536 	u8 (*mac_list)[ETH_ALEN];
537 	struct l2addr_node *node;
538 	struct vport_addr *addr;
539 	struct hlist_head *hash;
540 	struct hlist_node *tmp;
541 	int size;
542 	int err;
543 	int hi;
544 	int i;
545 
546 	size = is_uc ? MLX5_MAX_UC_PER_VPORT(esw->dev) :
547 		       MLX5_MAX_MC_PER_VPORT(esw->dev);
548 
549 	mac_list = kcalloc(size, ETH_ALEN, GFP_KERNEL);
550 	if (!mac_list)
551 		return;
552 
553 	hash = is_uc ? vport->uc_list : vport->mc_list;
554 
555 	for_each_l2hash_node(node, tmp, hash, hi) {
556 		addr = container_of(node, struct vport_addr, node);
557 		addr->action = MLX5_ACTION_DEL;
558 	}
559 
560 	if (!vport->enabled)
561 		goto out;
562 
563 	err = mlx5_query_nic_vport_mac_list(esw->dev, vport->vport, list_type,
564 					    mac_list, &size);
565 	if (err)
566 		goto out;
567 	esw_debug(esw->dev, "vport[%d] context update %s list size (%d)\n",
568 		  vport->vport, is_uc ? "UC" : "MC", size);
569 
570 	for (i = 0; i < size; i++) {
571 		if (is_uc && !is_valid_ether_addr(mac_list[i]))
572 			continue;
573 
574 		if (!is_uc && !is_multicast_ether_addr(mac_list[i]))
575 			continue;
576 
577 		addr = l2addr_hash_find(hash, mac_list[i], struct vport_addr);
578 		if (addr) {
579 			addr->action = MLX5_ACTION_NONE;
580 			/* If this mac was previously added because of allmulti
581 			 * promiscuous rx mode, its now converted to be original
582 			 * vport mac.
583 			 */
584 			if (addr->mc_promisc) {
585 				struct esw_mc_addr *esw_mc =
586 					l2addr_hash_find(esw->mc_table,
587 							 mac_list[i],
588 							 struct esw_mc_addr);
589 				if (!esw_mc) {
590 					esw_warn(esw->dev,
591 						 "Failed to MAC(%pM) in mcast DB\n",
592 						 mac_list[i]);
593 					continue;
594 				}
595 				esw_mc->refcnt++;
596 				addr->mc_promisc = false;
597 			}
598 			continue;
599 		}
600 
601 		addr = l2addr_hash_add(hash, mac_list[i], struct vport_addr,
602 				       GFP_KERNEL);
603 		if (!addr) {
604 			esw_warn(esw->dev,
605 				 "Failed to add MAC(%pM) to vport[%d] DB\n",
606 				 mac_list[i], vport->vport);
607 			continue;
608 		}
609 		addr->vport = vport->vport;
610 		addr->action = MLX5_ACTION_ADD;
611 	}
612 out:
613 	kfree(mac_list);
614 }
615 
616 /* Sync vport UC/MC list from vport context
617  * Must be called after esw_update_vport_addr_list
618  */
619 static void esw_update_vport_mc_promisc(struct mlx5_eswitch *esw,
620 					struct mlx5_vport *vport)
621 {
622 	struct l2addr_node *node;
623 	struct vport_addr *addr;
624 	struct hlist_head *hash;
625 	struct hlist_node *tmp;
626 	int hi;
627 
628 	hash = vport->mc_list;
629 
630 	for_each_l2hash_node(node, tmp, esw->mc_table, hi) {
631 		u8 *mac = node->addr;
632 
633 		addr = l2addr_hash_find(hash, mac, struct vport_addr);
634 		if (addr) {
635 			if (addr->action == MLX5_ACTION_DEL)
636 				addr->action = MLX5_ACTION_NONE;
637 			continue;
638 		}
639 		addr = l2addr_hash_add(hash, mac, struct vport_addr,
640 				       GFP_KERNEL);
641 		if (!addr) {
642 			esw_warn(esw->dev,
643 				 "Failed to add allmulti MAC(%pM) to vport[%d] DB\n",
644 				 mac, vport->vport);
645 			continue;
646 		}
647 		addr->vport = vport->vport;
648 		addr->action = MLX5_ACTION_ADD;
649 		addr->mc_promisc = true;
650 	}
651 }
652 
653 /* Apply vport rx mode to HW FDB table */
654 static void esw_apply_vport_rx_mode(struct mlx5_eswitch *esw,
655 				    struct mlx5_vport *vport,
656 				    bool promisc, bool mc_promisc)
657 {
658 	struct esw_mc_addr *allmulti_addr = &esw->mc_promisc;
659 
660 	if (IS_ERR_OR_NULL(vport->allmulti_rule) != mc_promisc)
661 		goto promisc;
662 
663 	if (mc_promisc) {
664 		vport->allmulti_rule =
665 			esw_fdb_set_vport_allmulti_rule(esw, vport->vport);
666 		if (!allmulti_addr->uplink_rule)
667 			allmulti_addr->uplink_rule =
668 				esw_fdb_set_vport_allmulti_rule(esw,
669 								MLX5_VPORT_UPLINK);
670 		allmulti_addr->refcnt++;
671 	} else if (vport->allmulti_rule) {
672 		mlx5_del_flow_rules(vport->allmulti_rule);
673 		vport->allmulti_rule = NULL;
674 
675 		if (--allmulti_addr->refcnt > 0)
676 			goto promisc;
677 
678 		if (allmulti_addr->uplink_rule)
679 			mlx5_del_flow_rules(allmulti_addr->uplink_rule);
680 		allmulti_addr->uplink_rule = NULL;
681 	}
682 
683 promisc:
684 	if (IS_ERR_OR_NULL(vport->promisc_rule) != promisc)
685 		return;
686 
687 	if (promisc) {
688 		vport->promisc_rule =
689 			esw_fdb_set_vport_promisc_rule(esw, vport->vport);
690 	} else if (vport->promisc_rule) {
691 		mlx5_del_flow_rules(vport->promisc_rule);
692 		vport->promisc_rule = NULL;
693 	}
694 }
695 
696 /* Sync vport rx mode from vport context */
697 static void esw_update_vport_rx_mode(struct mlx5_eswitch *esw,
698 				     struct mlx5_vport *vport)
699 {
700 	int promisc_all = 0;
701 	int promisc_uc = 0;
702 	int promisc_mc = 0;
703 	int err;
704 
705 	err = mlx5_query_nic_vport_promisc(esw->dev,
706 					   vport->vport,
707 					   &promisc_uc,
708 					   &promisc_mc,
709 					   &promisc_all);
710 	if (err)
711 		return;
712 	esw_debug(esw->dev, "vport[%d] context update rx mode promisc_all=%d, all_multi=%d\n",
713 		  vport->vport, promisc_all, promisc_mc);
714 
715 	if (!vport->info.trusted || !vport->enabled) {
716 		promisc_uc = 0;
717 		promisc_mc = 0;
718 		promisc_all = 0;
719 	}
720 
721 	esw_apply_vport_rx_mode(esw, vport, promisc_all,
722 				(promisc_all || promisc_mc));
723 }
724 
725 void esw_vport_change_handle_locked(struct mlx5_vport *vport)
726 {
727 	struct mlx5_core_dev *dev = vport->dev;
728 	struct mlx5_eswitch *esw = dev->priv.eswitch;
729 	u8 mac[ETH_ALEN];
730 
731 	if (!MLX5_CAP_GEN(dev, log_max_l2_table))
732 		return;
733 
734 	mlx5_query_nic_vport_mac_address(dev, vport->vport, true, mac);
735 	esw_debug(dev, "vport[%d] Context Changed: perm mac: %pM\n",
736 		  vport->vport, mac);
737 
738 	if (vport->enabled_events & MLX5_VPORT_UC_ADDR_CHANGE) {
739 		esw_update_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_UC);
740 		esw_apply_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_UC);
741 	}
742 
743 	if (vport->enabled_events & MLX5_VPORT_MC_ADDR_CHANGE)
744 		esw_update_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_MC);
745 
746 	if (vport->enabled_events & MLX5_VPORT_PROMISC_CHANGE) {
747 		esw_update_vport_rx_mode(esw, vport);
748 		if (!IS_ERR_OR_NULL(vport->allmulti_rule))
749 			esw_update_vport_mc_promisc(esw, vport);
750 	}
751 
752 	if (vport->enabled_events & (MLX5_VPORT_PROMISC_CHANGE | MLX5_VPORT_MC_ADDR_CHANGE))
753 		esw_apply_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_MC);
754 
755 	esw_debug(esw->dev, "vport[%d] Context Changed: Done\n", vport->vport);
756 	if (vport->enabled)
757 		arm_vport_context_events_cmd(dev, vport->vport,
758 					     vport->enabled_events);
759 }
760 
761 static void esw_vport_change_handler(struct work_struct *work)
762 {
763 	struct mlx5_vport *vport =
764 		container_of(work, struct mlx5_vport, vport_change_handler);
765 	struct mlx5_eswitch *esw = vport->dev->priv.eswitch;
766 
767 	mutex_lock(&esw->state_lock);
768 	esw_vport_change_handle_locked(vport);
769 	mutex_unlock(&esw->state_lock);
770 }
771 
772 static void node_guid_gen_from_mac(u64 *node_guid, const u8 *mac)
773 {
774 	((u8 *)node_guid)[7] = mac[0];
775 	((u8 *)node_guid)[6] = mac[1];
776 	((u8 *)node_guid)[5] = mac[2];
777 	((u8 *)node_guid)[4] = 0xff;
778 	((u8 *)node_guid)[3] = 0xfe;
779 	((u8 *)node_guid)[2] = mac[3];
780 	((u8 *)node_guid)[1] = mac[4];
781 	((u8 *)node_guid)[0] = mac[5];
782 }
783 
784 static int esw_vport_setup_acl(struct mlx5_eswitch *esw,
785 			       struct mlx5_vport *vport)
786 {
787 	if (esw->mode == MLX5_ESWITCH_LEGACY)
788 		return esw_legacy_vport_acl_setup(esw, vport);
789 	else
790 		return esw_vport_create_offloads_acl_tables(esw, vport);
791 }
792 
793 static void esw_vport_cleanup_acl(struct mlx5_eswitch *esw,
794 				  struct mlx5_vport *vport)
795 {
796 	if (esw->mode == MLX5_ESWITCH_LEGACY)
797 		esw_legacy_vport_acl_cleanup(esw, vport);
798 	else
799 		esw_vport_destroy_offloads_acl_tables(esw, vport);
800 }
801 
802 static int mlx5_esw_vport_caps_get(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
803 {
804 	int query_out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
805 	void *query_ctx;
806 	void *hca_caps;
807 	int err;
808 
809 	if (!MLX5_CAP_GEN(esw->dev, vhca_resource_manager))
810 		return 0;
811 
812 	query_ctx = kzalloc(query_out_sz, GFP_KERNEL);
813 	if (!query_ctx)
814 		return -ENOMEM;
815 
816 	err = mlx5_vport_get_other_func_cap(esw->dev, vport->vport, query_ctx,
817 					    MLX5_CAP_GENERAL);
818 	if (err)
819 		goto out_free;
820 
821 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
822 	vport->info.roce_enabled = MLX5_GET(cmd_hca_cap, hca_caps, roce);
823 
824 	if (!MLX5_CAP_GEN_MAX(esw->dev, hca_cap_2))
825 		goto out_free;
826 
827 	memset(query_ctx, 0, query_out_sz);
828 	err = mlx5_vport_get_other_func_cap(esw->dev, vport->vport, query_ctx,
829 					    MLX5_CAP_GENERAL_2);
830 	if (err)
831 		goto out_free;
832 
833 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
834 	vport->info.mig_enabled = MLX5_GET(cmd_hca_cap_2, hca_caps, migratable);
835 
836 	err = mlx5_esw_ipsec_vf_offload_get(esw->dev, vport);
837 out_free:
838 	kfree(query_ctx);
839 	return err;
840 }
841 
842 static int esw_vport_setup(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
843 {
844 	bool vst_mode_steering = esw_vst_mode_is_steering(esw);
845 	u16 vport_num = vport->vport;
846 	int flags;
847 	int err;
848 
849 	err = esw_vport_setup_acl(esw, vport);
850 	if (err)
851 		return err;
852 
853 	if (mlx5_esw_is_manager_vport(esw, vport_num))
854 		return 0;
855 
856 	err = mlx5_esw_vport_caps_get(esw, vport);
857 	if (err)
858 		goto err_caps;
859 
860 	mlx5_modify_vport_admin_state(esw->dev,
861 				      MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
862 				      vport_num, 1,
863 				      vport->info.link_state);
864 
865 	/* Host PF has its own mac/guid. */
866 	if (vport_num) {
867 		mlx5_modify_nic_vport_mac_address(esw->dev, vport_num,
868 						  vport->info.mac);
869 		mlx5_modify_nic_vport_node_guid(esw->dev, vport_num,
870 						vport->info.node_guid);
871 	}
872 
873 	flags = (vport->info.vlan || vport->info.qos) ?
874 		SET_VLAN_STRIP | SET_VLAN_INSERT : 0;
875 	if (esw->mode == MLX5_ESWITCH_OFFLOADS || !vst_mode_steering)
876 		modify_esw_vport_cvlan(esw->dev, vport_num, vport->info.vlan,
877 				       vport->info.qos, flags);
878 
879 	return 0;
880 
881 err_caps:
882 	esw_vport_cleanup_acl(esw, vport);
883 	return err;
884 }
885 
886 /* Don't cleanup vport->info, it's needed to restore vport configuration */
887 static void esw_vport_cleanup(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
888 {
889 	u16 vport_num = vport->vport;
890 
891 	if (!mlx5_esw_is_manager_vport(esw, vport_num))
892 		mlx5_modify_vport_admin_state(esw->dev,
893 					      MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
894 					      vport_num, 1,
895 					      MLX5_VPORT_ADMIN_STATE_DOWN);
896 
897 	mlx5_esw_qos_vport_disable(esw, vport);
898 	esw_vport_cleanup_acl(esw, vport);
899 }
900 
901 int mlx5_esw_vport_enable(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
902 			  enum mlx5_eswitch_vport_event enabled_events)
903 {
904 	u16 vport_num = vport->vport;
905 	int ret;
906 
907 	mutex_lock(&esw->state_lock);
908 	WARN_ON(vport->enabled);
909 
910 	esw_debug(esw->dev, "Enabling VPORT(%d)\n", vport_num);
911 
912 	ret = esw_vport_setup(esw, vport);
913 	if (ret)
914 		goto done;
915 
916 	/* Sync with current vport context */
917 	vport->enabled_events = enabled_events;
918 	vport->enabled = true;
919 	if (vport->vport != MLX5_VPORT_PF &&
920 	    (vport->info.ipsec_crypto_enabled || vport->info.ipsec_packet_enabled))
921 		esw->enabled_ipsec_vf_count++;
922 
923 	/* Esw manager is trusted by default. Host PF (vport 0) is trusted as well
924 	 * in smartNIC as it's a vport group manager.
925 	 */
926 	if (mlx5_esw_is_manager_vport(esw, vport_num) ||
927 	    (!vport_num && mlx5_core_is_ecpf(esw->dev)))
928 		vport->info.trusted = true;
929 
930 	if (!mlx5_esw_is_manager_vport(esw, vport_num) &&
931 	    MLX5_CAP_GEN(esw->dev, vhca_resource_manager)) {
932 		ret = mlx5_esw_vport_vhca_id_set(esw, vport_num);
933 		if (ret)
934 			goto err_vhca_mapping;
935 	}
936 
937 	/* External controller host PF has factory programmed MAC.
938 	 * Read it from the device.
939 	 */
940 	if (mlx5_core_is_ecpf(esw->dev) && vport_num == MLX5_VPORT_PF)
941 		mlx5_query_nic_vport_mac_address(esw->dev, vport_num, true, vport->info.mac);
942 
943 	esw_vport_change_handle_locked(vport);
944 
945 	esw->enabled_vports++;
946 	esw_debug(esw->dev, "Enabled VPORT(%d)\n", vport_num);
947 done:
948 	mutex_unlock(&esw->state_lock);
949 	return ret;
950 
951 err_vhca_mapping:
952 	esw_vport_cleanup(esw, vport);
953 	mutex_unlock(&esw->state_lock);
954 	return ret;
955 }
956 
957 void mlx5_esw_vport_disable(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
958 {
959 	u16 vport_num = vport->vport;
960 
961 	mutex_lock(&esw->state_lock);
962 
963 	if (!vport->enabled)
964 		goto done;
965 
966 	esw_debug(esw->dev, "Disabling vport(%d)\n", vport_num);
967 	/* Mark this vport as disabled to discard new events */
968 	vport->enabled = false;
969 
970 	/* Disable events from this vport */
971 	if (MLX5_CAP_GEN(esw->dev, log_max_l2_table))
972 		arm_vport_context_events_cmd(esw->dev, vport_num, 0);
973 
974 	if (!mlx5_esw_is_manager_vport(esw, vport_num) &&
975 	    MLX5_CAP_GEN(esw->dev, vhca_resource_manager))
976 		mlx5_esw_vport_vhca_id_clear(esw, vport_num);
977 
978 	if (vport->vport != MLX5_VPORT_PF &&
979 	    (vport->info.ipsec_crypto_enabled || vport->info.ipsec_packet_enabled))
980 		esw->enabled_ipsec_vf_count--;
981 
982 	/* We don't assume VFs will cleanup after themselves.
983 	 * Calling vport change handler while vport is disabled will cleanup
984 	 * the vport resources.
985 	 */
986 	esw_vport_change_handle_locked(vport);
987 	vport->enabled_events = 0;
988 	esw_apply_vport_rx_mode(esw, vport, false, false);
989 	esw_vport_cleanup(esw, vport);
990 	esw->enabled_vports--;
991 
992 done:
993 	mutex_unlock(&esw->state_lock);
994 }
995 
996 static int eswitch_vport_event(struct notifier_block *nb,
997 			       unsigned long type, void *data)
998 {
999 	struct mlx5_eswitch *esw = mlx5_nb_cof(nb, struct mlx5_eswitch, nb);
1000 	struct mlx5_eqe *eqe = data;
1001 	struct mlx5_vport *vport;
1002 	u16 vport_num;
1003 
1004 	vport_num = be16_to_cpu(eqe->data.vport_change.vport_num);
1005 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1006 	if (!IS_ERR(vport))
1007 		queue_work(esw->work_queue, &vport->vport_change_handler);
1008 	return NOTIFY_OK;
1009 }
1010 
1011 /**
1012  * mlx5_esw_query_functions - Returns raw output about functions state
1013  * @dev:	Pointer to device to query
1014  *
1015  * mlx5_esw_query_functions() allocates and returns functions changed
1016  * raw output memory pointer from device on success. Otherwise returns ERR_PTR.
1017  * Caller must free the memory using kvfree() when valid pointer is returned.
1018  */
1019 const u32 *mlx5_esw_query_functions(struct mlx5_core_dev *dev)
1020 {
1021 	int outlen = MLX5_ST_SZ_BYTES(query_esw_functions_out);
1022 	u32 in[MLX5_ST_SZ_DW(query_esw_functions_in)] = {};
1023 	u32 *out;
1024 	int err;
1025 
1026 	out = kvzalloc(outlen, GFP_KERNEL);
1027 	if (!out)
1028 		return ERR_PTR(-ENOMEM);
1029 
1030 	MLX5_SET(query_esw_functions_in, in, opcode,
1031 		 MLX5_CMD_OP_QUERY_ESW_FUNCTIONS);
1032 
1033 	err = mlx5_cmd_exec(dev, in, sizeof(in), out, outlen);
1034 	if (!err)
1035 		return out;
1036 
1037 	kvfree(out);
1038 	return ERR_PTR(err);
1039 }
1040 
1041 static void mlx5_eswitch_event_handler_register(struct mlx5_eswitch *esw)
1042 {
1043 	if (esw->mode == MLX5_ESWITCH_OFFLOADS && mlx5_eswitch_is_funcs_handler(esw->dev)) {
1044 		MLX5_NB_INIT(&esw->esw_funcs.nb, mlx5_esw_funcs_changed_handler,
1045 			     ESW_FUNCTIONS_CHANGED);
1046 		mlx5_eq_notifier_register(esw->dev, &esw->esw_funcs.nb);
1047 	}
1048 }
1049 
1050 static void mlx5_eswitch_event_handler_unregister(struct mlx5_eswitch *esw)
1051 {
1052 	if (esw->mode == MLX5_ESWITCH_OFFLOADS && mlx5_eswitch_is_funcs_handler(esw->dev))
1053 		mlx5_eq_notifier_unregister(esw->dev, &esw->esw_funcs.nb);
1054 
1055 	flush_workqueue(esw->work_queue);
1056 }
1057 
1058 static void mlx5_eswitch_clear_vf_vports_info(struct mlx5_eswitch *esw)
1059 {
1060 	struct mlx5_vport *vport;
1061 	unsigned long i;
1062 
1063 	mlx5_esw_for_each_vf_vport(esw, i, vport, esw->esw_funcs.num_vfs) {
1064 		memset(&vport->qos, 0, sizeof(vport->qos));
1065 		memset(&vport->info, 0, sizeof(vport->info));
1066 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1067 	}
1068 }
1069 
1070 static void mlx5_eswitch_clear_ec_vf_vports_info(struct mlx5_eswitch *esw)
1071 {
1072 	struct mlx5_vport *vport;
1073 	unsigned long i;
1074 
1075 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, esw->esw_funcs.num_ec_vfs) {
1076 		memset(&vport->qos, 0, sizeof(vport->qos));
1077 		memset(&vport->info, 0, sizeof(vport->info));
1078 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1079 	}
1080 }
1081 
1082 static int mlx5_eswitch_load_vport(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
1083 				   enum mlx5_eswitch_vport_event enabled_events)
1084 {
1085 	int err;
1086 
1087 	err = mlx5_esw_vport_enable(esw, vport, enabled_events);
1088 	if (err)
1089 		return err;
1090 
1091 	err = mlx5_esw_offloads_load_rep(esw, vport);
1092 	if (err)
1093 		goto err_rep;
1094 
1095 	return err;
1096 
1097 err_rep:
1098 	mlx5_esw_vport_disable(esw, vport);
1099 	return err;
1100 }
1101 
1102 static void mlx5_eswitch_unload_vport(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1103 {
1104 	mlx5_esw_offloads_unload_rep(esw, vport);
1105 	mlx5_esw_vport_disable(esw, vport);
1106 }
1107 
1108 static int mlx5_eswitch_load_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num,
1109 					 enum mlx5_eswitch_vport_event enabled_events)
1110 {
1111 	struct mlx5_vport *vport;
1112 	int err;
1113 
1114 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1115 	if (IS_ERR(vport))
1116 		return PTR_ERR(vport);
1117 
1118 	err = mlx5_esw_offloads_init_pf_vf_rep(esw, vport);
1119 	if (err)
1120 		return err;
1121 
1122 	err = mlx5_eswitch_load_vport(esw, vport, enabled_events);
1123 	if (err)
1124 		goto err_load;
1125 	return 0;
1126 
1127 err_load:
1128 	mlx5_esw_offloads_cleanup_pf_vf_rep(esw, vport);
1129 	return err;
1130 }
1131 
1132 static void mlx5_eswitch_unload_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
1133 {
1134 	struct mlx5_vport *vport;
1135 
1136 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1137 	if (IS_ERR(vport))
1138 		return;
1139 
1140 	mlx5_eswitch_unload_vport(esw, vport);
1141 	mlx5_esw_offloads_cleanup_pf_vf_rep(esw, vport);
1142 }
1143 
1144 int mlx5_eswitch_load_sf_vport(struct mlx5_eswitch *esw, u16 vport_num,
1145 			       enum mlx5_eswitch_vport_event enabled_events,
1146 			       struct mlx5_devlink_port *dl_port, u32 controller, u32 sfnum)
1147 {
1148 	struct mlx5_vport *vport;
1149 	int err;
1150 
1151 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1152 	if (IS_ERR(vport))
1153 		return PTR_ERR(vport);
1154 
1155 	err = mlx5_esw_offloads_init_sf_rep(esw, vport, dl_port, controller, sfnum);
1156 	if (err)
1157 		return err;
1158 
1159 	err = mlx5_eswitch_load_vport(esw, vport, enabled_events);
1160 	if (err)
1161 		goto err_load;
1162 
1163 	return 0;
1164 
1165 err_load:
1166 	mlx5_esw_offloads_cleanup_sf_rep(esw, vport);
1167 	return err;
1168 }
1169 
1170 void mlx5_eswitch_unload_sf_vport(struct mlx5_eswitch *esw, u16 vport_num)
1171 {
1172 	struct mlx5_vport *vport;
1173 
1174 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1175 	if (IS_ERR(vport))
1176 		return;
1177 
1178 	mlx5_eswitch_unload_vport(esw, vport);
1179 	mlx5_esw_offloads_cleanup_sf_rep(esw, vport);
1180 }
1181 
1182 void mlx5_eswitch_unload_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs)
1183 {
1184 	struct mlx5_vport *vport;
1185 	unsigned long i;
1186 
1187 	mlx5_esw_for_each_vf_vport(esw, i, vport, num_vfs) {
1188 		if (!vport->enabled)
1189 			continue;
1190 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1191 	}
1192 }
1193 
1194 static void mlx5_eswitch_unload_ec_vf_vports(struct mlx5_eswitch *esw,
1195 					     u16 num_ec_vfs)
1196 {
1197 	struct mlx5_vport *vport;
1198 	unsigned long i;
1199 
1200 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, num_ec_vfs) {
1201 		if (!vport->enabled)
1202 			continue;
1203 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1204 	}
1205 }
1206 
1207 int mlx5_eswitch_load_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs,
1208 				enum mlx5_eswitch_vport_event enabled_events)
1209 {
1210 	struct mlx5_vport *vport;
1211 	unsigned long i;
1212 	int err;
1213 
1214 	mlx5_esw_for_each_vf_vport(esw, i, vport, num_vfs) {
1215 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport, enabled_events);
1216 		if (err)
1217 			goto vf_err;
1218 	}
1219 
1220 	return 0;
1221 
1222 vf_err:
1223 	mlx5_eswitch_unload_vf_vports(esw, num_vfs);
1224 	return err;
1225 }
1226 
1227 static int mlx5_eswitch_load_ec_vf_vports(struct mlx5_eswitch *esw, u16 num_ec_vfs,
1228 					  enum mlx5_eswitch_vport_event enabled_events)
1229 {
1230 	struct mlx5_vport *vport;
1231 	unsigned long i;
1232 	int err;
1233 
1234 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, num_ec_vfs) {
1235 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport, enabled_events);
1236 		if (err)
1237 			goto vf_err;
1238 	}
1239 
1240 	return 0;
1241 
1242 vf_err:
1243 	mlx5_eswitch_unload_ec_vf_vports(esw, num_ec_vfs);
1244 	return err;
1245 }
1246 
1247 static int host_pf_enable_hca(struct mlx5_core_dev *dev)
1248 {
1249 	if (!mlx5_core_is_ecpf(dev))
1250 		return 0;
1251 
1252 	/* Once vport and representor are ready, take out the external host PF
1253 	 * out of initializing state. Enabling HCA clears the iser->initializing
1254 	 * bit and host PF driver loading can progress.
1255 	 */
1256 	return mlx5_cmd_host_pf_enable_hca(dev);
1257 }
1258 
1259 static void host_pf_disable_hca(struct mlx5_core_dev *dev)
1260 {
1261 	if (!mlx5_core_is_ecpf(dev))
1262 		return;
1263 
1264 	mlx5_cmd_host_pf_disable_hca(dev);
1265 }
1266 
1267 /* mlx5_eswitch_enable_pf_vf_vports() enables vports of PF, ECPF and VFs
1268  * whichever are present on the eswitch.
1269  */
1270 int
1271 mlx5_eswitch_enable_pf_vf_vports(struct mlx5_eswitch *esw,
1272 				 enum mlx5_eswitch_vport_event enabled_events)
1273 {
1274 	bool pf_needed;
1275 	int ret;
1276 
1277 	pf_needed = mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1278 		    esw->mode == MLX5_ESWITCH_LEGACY;
1279 
1280 	/* Enable PF vport */
1281 	if (pf_needed) {
1282 		ret = mlx5_eswitch_load_pf_vf_vport(esw, MLX5_VPORT_PF,
1283 						    enabled_events);
1284 		if (ret)
1285 			return ret;
1286 	}
1287 
1288 	/* Enable external host PF HCA */
1289 	ret = host_pf_enable_hca(esw->dev);
1290 	if (ret)
1291 		goto pf_hca_err;
1292 
1293 	/* Enable ECPF vport */
1294 	if (mlx5_ecpf_vport_exists(esw->dev)) {
1295 		ret = mlx5_eswitch_load_pf_vf_vport(esw, MLX5_VPORT_ECPF, enabled_events);
1296 		if (ret)
1297 			goto ecpf_err;
1298 		if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1299 			ret = mlx5_eswitch_load_ec_vf_vports(esw, esw->esw_funcs.num_ec_vfs,
1300 							     enabled_events);
1301 			if (ret)
1302 				goto ec_vf_err;
1303 		}
1304 	}
1305 
1306 	/* Enable VF vports */
1307 	ret = mlx5_eswitch_load_vf_vports(esw, esw->esw_funcs.num_vfs,
1308 					  enabled_events);
1309 	if (ret)
1310 		goto vf_err;
1311 	return 0;
1312 
1313 vf_err:
1314 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1315 		mlx5_eswitch_unload_ec_vf_vports(esw, esw->esw_funcs.num_ec_vfs);
1316 ec_vf_err:
1317 	if (mlx5_ecpf_vport_exists(esw->dev))
1318 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_ECPF);
1319 ecpf_err:
1320 	host_pf_disable_hca(esw->dev);
1321 pf_hca_err:
1322 	if (pf_needed)
1323 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_PF);
1324 	return ret;
1325 }
1326 
1327 /* mlx5_eswitch_disable_pf_vf_vports() disables vports of PF, ECPF and VFs
1328  * whichever are previously enabled on the eswitch.
1329  */
1330 void mlx5_eswitch_disable_pf_vf_vports(struct mlx5_eswitch *esw)
1331 {
1332 	mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
1333 
1334 	if (mlx5_ecpf_vport_exists(esw->dev)) {
1335 		if (mlx5_core_ec_sriov_enabled(esw->dev))
1336 			mlx5_eswitch_unload_ec_vf_vports(esw, esw->esw_funcs.num_vfs);
1337 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_ECPF);
1338 	}
1339 
1340 	host_pf_disable_hca(esw->dev);
1341 
1342 	if (mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1343 	    esw->mode == MLX5_ESWITCH_LEGACY)
1344 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_PF);
1345 }
1346 
1347 static void mlx5_eswitch_get_devlink_param(struct mlx5_eswitch *esw)
1348 {
1349 	struct devlink *devlink = priv_to_devlink(esw->dev);
1350 	union devlink_param_value val;
1351 	int err;
1352 
1353 	err = devl_param_driverinit_value_get(devlink,
1354 					      MLX5_DEVLINK_PARAM_ID_ESW_LARGE_GROUP_NUM,
1355 					      &val);
1356 	if (!err) {
1357 		esw->params.large_group_num = val.vu32;
1358 	} else {
1359 		esw_warn(esw->dev,
1360 			 "Devlink can't get param fdb_large_groups, uses default (%d).\n",
1361 			 ESW_OFFLOADS_DEFAULT_NUM_GROUPS);
1362 		esw->params.large_group_num = ESW_OFFLOADS_DEFAULT_NUM_GROUPS;
1363 	}
1364 }
1365 
1366 static void
1367 mlx5_eswitch_update_num_of_vfs(struct mlx5_eswitch *esw, int num_vfs)
1368 {
1369 	const u32 *out;
1370 
1371 	if (num_vfs < 0)
1372 		return;
1373 
1374 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1375 		esw->esw_funcs.num_vfs = num_vfs;
1376 		return;
1377 	}
1378 
1379 	out = mlx5_esw_query_functions(esw->dev);
1380 	if (IS_ERR(out))
1381 		return;
1382 
1383 	esw->esw_funcs.num_vfs = MLX5_GET(query_esw_functions_out, out,
1384 					  host_params_context.host_num_of_vfs);
1385 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1386 		esw->esw_funcs.num_ec_vfs = num_vfs;
1387 
1388 	kvfree(out);
1389 }
1390 
1391 static void mlx5_esw_mode_change_notify(struct mlx5_eswitch *esw, u16 mode)
1392 {
1393 	struct mlx5_esw_event_info info = {};
1394 
1395 	info.new_mode = mode;
1396 
1397 	blocking_notifier_call_chain(&esw->n_head, 0, &info);
1398 }
1399 
1400 static int mlx5_esw_acls_ns_init(struct mlx5_eswitch *esw)
1401 {
1402 	struct mlx5_core_dev *dev = esw->dev;
1403 	int total_vports;
1404 	int err;
1405 
1406 	if (esw->flags & MLX5_ESWITCH_VPORT_ACL_NS_CREATED)
1407 		return 0;
1408 
1409 	total_vports = mlx5_eswitch_get_total_vports(dev);
1410 
1411 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
1412 		err = mlx5_fs_egress_acls_init(dev, total_vports);
1413 		if (err)
1414 			return err;
1415 	} else {
1416 		esw_warn(dev, "egress ACL is not supported by FW\n");
1417 	}
1418 
1419 	if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
1420 		err = mlx5_fs_ingress_acls_init(dev, total_vports);
1421 		if (err)
1422 			goto err;
1423 	} else {
1424 		esw_warn(dev, "ingress ACL is not supported by FW\n");
1425 	}
1426 	esw->flags |= MLX5_ESWITCH_VPORT_ACL_NS_CREATED;
1427 	return 0;
1428 
1429 err:
1430 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
1431 		mlx5_fs_egress_acls_cleanup(dev);
1432 	return err;
1433 }
1434 
1435 static void mlx5_esw_acls_ns_cleanup(struct mlx5_eswitch *esw)
1436 {
1437 	struct mlx5_core_dev *dev = esw->dev;
1438 
1439 	esw->flags &= ~MLX5_ESWITCH_VPORT_ACL_NS_CREATED;
1440 	if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support))
1441 		mlx5_fs_ingress_acls_cleanup(dev);
1442 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
1443 		mlx5_fs_egress_acls_cleanup(dev);
1444 }
1445 
1446 /**
1447  * mlx5_eswitch_enable_locked - Enable eswitch
1448  * @esw:	Pointer to eswitch
1449  * @num_vfs:	Enable eswitch for given number of VFs. This is optional.
1450  *		Valid value are 0, > 0 and MLX5_ESWITCH_IGNORE_NUM_VFS.
1451  *		Caller should pass num_vfs > 0 when enabling eswitch for
1452  *		vf vports. Caller should pass num_vfs = 0, when eswitch
1453  *		is enabled without sriov VFs or when caller
1454  *		is unaware of the sriov state of the host PF on ECPF based
1455  *		eswitch. Caller should pass < 0 when num_vfs should be
1456  *		completely ignored. This is typically the case when eswitch
1457  *		is enabled without sriov regardless of PF/ECPF system.
1458  * mlx5_eswitch_enable_locked() Enables eswitch in either legacy or offloads
1459  * mode. If num_vfs >=0 is provided, it setup VF related eswitch vports.
1460  * It returns 0 on success or error code on failure.
1461  */
1462 int mlx5_eswitch_enable_locked(struct mlx5_eswitch *esw, int num_vfs)
1463 {
1464 	int err;
1465 
1466 	devl_assert_locked(priv_to_devlink(esw->dev));
1467 
1468 	if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, ft_support)) {
1469 		esw_warn(esw->dev, "FDB is not supported, aborting ...\n");
1470 		return -EOPNOTSUPP;
1471 	}
1472 
1473 	mlx5_eswitch_get_devlink_param(esw);
1474 
1475 	err = mlx5_esw_acls_ns_init(esw);
1476 	if (err)
1477 		return err;
1478 
1479 	mlx5_eswitch_update_num_of_vfs(esw, num_vfs);
1480 
1481 	MLX5_NB_INIT(&esw->nb, eswitch_vport_event, NIC_VPORT_CHANGE);
1482 	mlx5_eq_notifier_register(esw->dev, &esw->nb);
1483 
1484 	if (esw->mode == MLX5_ESWITCH_LEGACY) {
1485 		err = esw_legacy_enable(esw);
1486 	} else {
1487 		mlx5_rescan_drivers(esw->dev);
1488 		err = esw_offloads_enable(esw);
1489 	}
1490 
1491 	if (err)
1492 		goto abort;
1493 
1494 	esw->fdb_table.flags |= MLX5_ESW_FDB_CREATED;
1495 
1496 	mlx5_eswitch_event_handler_register(esw);
1497 
1498 	esw_info(esw->dev, "Enable: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1499 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1500 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1501 
1502 	mlx5_esw_mode_change_notify(esw, esw->mode);
1503 
1504 	return 0;
1505 
1506 abort:
1507 	mlx5_esw_acls_ns_cleanup(esw);
1508 	return err;
1509 }
1510 
1511 /**
1512  * mlx5_eswitch_enable - Enable eswitch
1513  * @esw:	Pointer to eswitch
1514  * @num_vfs:	Enable eswitch switch for given number of VFs.
1515  *		Caller must pass num_vfs > 0 when enabling eswitch for
1516  *		vf vports.
1517  * mlx5_eswitch_enable() returns 0 on success or error code on failure.
1518  */
1519 int mlx5_eswitch_enable(struct mlx5_eswitch *esw, int num_vfs)
1520 {
1521 	bool toggle_lag;
1522 	int ret = 0;
1523 
1524 	if (!mlx5_esw_allowed(esw))
1525 		return 0;
1526 
1527 	devl_assert_locked(priv_to_devlink(esw->dev));
1528 
1529 	toggle_lag = !mlx5_esw_is_fdb_created(esw);
1530 
1531 	if (toggle_lag)
1532 		mlx5_lag_disable_change(esw->dev);
1533 
1534 	if (!mlx5_esw_is_fdb_created(esw)) {
1535 		ret = mlx5_eswitch_enable_locked(esw, num_vfs);
1536 	} else {
1537 		enum mlx5_eswitch_vport_event vport_events;
1538 
1539 		vport_events = (esw->mode == MLX5_ESWITCH_LEGACY) ?
1540 					MLX5_LEGACY_SRIOV_VPORT_EVENTS : MLX5_VPORT_UC_ADDR_CHANGE;
1541 		/* If this is the ECPF the number of host VFs is managed via the
1542 		 * eswitch function change event handler, and any num_vfs provided
1543 		 * here are intended to be EC VFs.
1544 		 */
1545 		if (!mlx5_core_is_ecpf(esw->dev)) {
1546 			ret = mlx5_eswitch_load_vf_vports(esw, num_vfs, vport_events);
1547 			if (!ret)
1548 				esw->esw_funcs.num_vfs = num_vfs;
1549 		} else if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1550 			ret = mlx5_eswitch_load_ec_vf_vports(esw, num_vfs, vport_events);
1551 			if (!ret)
1552 				esw->esw_funcs.num_ec_vfs = num_vfs;
1553 		}
1554 	}
1555 
1556 	if (toggle_lag)
1557 		mlx5_lag_enable_change(esw->dev);
1558 
1559 	return ret;
1560 }
1561 
1562 /* When disabling sriov, free driver level resources. */
1563 void mlx5_eswitch_disable_sriov(struct mlx5_eswitch *esw, bool clear_vf)
1564 {
1565 	if (!mlx5_esw_allowed(esw))
1566 		return;
1567 
1568 	devl_assert_locked(priv_to_devlink(esw->dev));
1569 	/* If driver is unloaded, this function is called twice by remove_one()
1570 	 * and mlx5_unload(). Prevent the second call.
1571 	 */
1572 	if (!esw->esw_funcs.num_vfs && !esw->esw_funcs.num_ec_vfs && !clear_vf)
1573 		return;
1574 
1575 	esw_info(esw->dev, "Unload vfs: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1576 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1577 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1578 
1579 	if (!mlx5_core_is_ecpf(esw->dev)) {
1580 		mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
1581 		if (clear_vf)
1582 			mlx5_eswitch_clear_vf_vports_info(esw);
1583 	} else if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1584 		mlx5_eswitch_unload_ec_vf_vports(esw, esw->esw_funcs.num_ec_vfs);
1585 		if (clear_vf)
1586 			mlx5_eswitch_clear_ec_vf_vports_info(esw);
1587 	}
1588 
1589 	if (esw->mode == MLX5_ESWITCH_OFFLOADS) {
1590 		struct devlink *devlink = priv_to_devlink(esw->dev);
1591 
1592 		devl_rate_nodes_destroy(devlink);
1593 	}
1594 	/* Destroy legacy fdb when disabling sriov in legacy mode. */
1595 	if (esw->mode == MLX5_ESWITCH_LEGACY)
1596 		mlx5_eswitch_disable_locked(esw);
1597 
1598 	if (!mlx5_core_is_ecpf(esw->dev))
1599 		esw->esw_funcs.num_vfs = 0;
1600 	else
1601 		esw->esw_funcs.num_ec_vfs = 0;
1602 }
1603 
1604 /* Free resources for corresponding eswitch mode. It is called by devlink
1605  * when changing eswitch mode or modprobe when unloading driver.
1606  */
1607 void mlx5_eswitch_disable_locked(struct mlx5_eswitch *esw)
1608 {
1609 	struct devlink *devlink = priv_to_devlink(esw->dev);
1610 
1611 	/* Notify eswitch users that it is exiting from current mode.
1612 	 * So that it can do necessary cleanup before the eswitch is disabled.
1613 	 */
1614 	mlx5_esw_mode_change_notify(esw, MLX5_ESWITCH_LEGACY);
1615 
1616 	mlx5_eq_notifier_unregister(esw->dev, &esw->nb);
1617 	mlx5_eswitch_event_handler_unregister(esw);
1618 
1619 	esw_info(esw->dev, "Disable: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1620 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1621 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1622 
1623 	if (esw->fdb_table.flags & MLX5_ESW_FDB_CREATED) {
1624 		esw->fdb_table.flags &= ~MLX5_ESW_FDB_CREATED;
1625 		if (esw->mode == MLX5_ESWITCH_OFFLOADS)
1626 			esw_offloads_disable(esw);
1627 		else if (esw->mode == MLX5_ESWITCH_LEGACY)
1628 			esw_legacy_disable(esw);
1629 		mlx5_esw_acls_ns_cleanup(esw);
1630 	}
1631 
1632 	if (esw->mode == MLX5_ESWITCH_OFFLOADS)
1633 		devl_rate_nodes_destroy(devlink);
1634 }
1635 
1636 void mlx5_eswitch_disable(struct mlx5_eswitch *esw)
1637 {
1638 	if (!mlx5_esw_allowed(esw))
1639 		return;
1640 
1641 	devl_assert_locked(priv_to_devlink(esw->dev));
1642 	mlx5_lag_disable_change(esw->dev);
1643 	mlx5_eswitch_disable_locked(esw);
1644 	esw->mode = MLX5_ESWITCH_LEGACY;
1645 	mlx5_lag_enable_change(esw->dev);
1646 }
1647 
1648 static int mlx5_query_hca_cap_host_pf(struct mlx5_core_dev *dev, void *out)
1649 {
1650 	u16 opmod = (MLX5_CAP_GENERAL << 1) | (HCA_CAP_OPMOD_GET_MAX & 0x01);
1651 	u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)] = {};
1652 
1653 	MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP);
1654 	MLX5_SET(query_hca_cap_in, in, op_mod, opmod);
1655 	MLX5_SET(query_hca_cap_in, in, function_id, MLX5_VPORT_PF);
1656 	MLX5_SET(query_hca_cap_in, in, other_function, true);
1657 	return mlx5_cmd_exec_inout(dev, query_hca_cap, in, out);
1658 }
1659 
1660 int mlx5_esw_sf_max_hpf_functions(struct mlx5_core_dev *dev, u16 *max_sfs, u16 *sf_base_id)
1661 
1662 {
1663 	int query_out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
1664 	void *query_ctx;
1665 	void *hca_caps;
1666 	int err;
1667 
1668 	if (!mlx5_core_is_ecpf(dev)) {
1669 		*max_sfs = 0;
1670 		return 0;
1671 	}
1672 
1673 	query_ctx = kzalloc(query_out_sz, GFP_KERNEL);
1674 	if (!query_ctx)
1675 		return -ENOMEM;
1676 
1677 	err = mlx5_query_hca_cap_host_pf(dev, query_ctx);
1678 	if (err)
1679 		goto out_free;
1680 
1681 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
1682 	*max_sfs = MLX5_GET(cmd_hca_cap, hca_caps, max_num_sf);
1683 	*sf_base_id = MLX5_GET(cmd_hca_cap, hca_caps, sf_base_id);
1684 
1685 out_free:
1686 	kfree(query_ctx);
1687 	return err;
1688 }
1689 
1690 static int mlx5_esw_vport_alloc(struct mlx5_eswitch *esw,
1691 				int index, u16 vport_num)
1692 {
1693 	struct mlx5_vport *vport;
1694 	int err;
1695 
1696 	vport = kzalloc(sizeof(*vport), GFP_KERNEL);
1697 	if (!vport)
1698 		return -ENOMEM;
1699 
1700 	vport->dev = esw->dev;
1701 	vport->vport = vport_num;
1702 	vport->index = index;
1703 	vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1704 	INIT_WORK(&vport->vport_change_handler, esw_vport_change_handler);
1705 	err = xa_insert(&esw->vports, vport_num, vport, GFP_KERNEL);
1706 	if (err)
1707 		goto insert_err;
1708 
1709 	esw->total_vports++;
1710 	return 0;
1711 
1712 insert_err:
1713 	kfree(vport);
1714 	return err;
1715 }
1716 
1717 static void mlx5_esw_vport_free(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1718 {
1719 	xa_erase(&esw->vports, vport->vport);
1720 	kfree(vport);
1721 }
1722 
1723 static void mlx5_esw_vports_cleanup(struct mlx5_eswitch *esw)
1724 {
1725 	struct mlx5_vport *vport;
1726 	unsigned long i;
1727 
1728 	mlx5_esw_for_each_vport(esw, i, vport)
1729 		mlx5_esw_vport_free(esw, vport);
1730 	xa_destroy(&esw->vports);
1731 }
1732 
1733 static int mlx5_esw_vports_init(struct mlx5_eswitch *esw)
1734 {
1735 	struct mlx5_core_dev *dev = esw->dev;
1736 	u16 max_host_pf_sfs;
1737 	u16 base_sf_num;
1738 	int idx = 0;
1739 	int err;
1740 	int i;
1741 
1742 	xa_init(&esw->vports);
1743 
1744 	err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_PF);
1745 	if (err)
1746 		goto err;
1747 	if (esw->first_host_vport == MLX5_VPORT_PF)
1748 		xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_HOST_FN);
1749 	idx++;
1750 
1751 	for (i = 0; i < mlx5_core_max_vfs(dev); i++) {
1752 		err = mlx5_esw_vport_alloc(esw, idx, idx);
1753 		if (err)
1754 			goto err;
1755 		xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_VF);
1756 		xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_HOST_FN);
1757 		idx++;
1758 	}
1759 	base_sf_num = mlx5_sf_start_function_id(dev);
1760 	for (i = 0; i < mlx5_sf_max_functions(dev); i++) {
1761 		err = mlx5_esw_vport_alloc(esw, idx, base_sf_num + i);
1762 		if (err)
1763 			goto err;
1764 		xa_set_mark(&esw->vports, base_sf_num + i, MLX5_ESW_VPT_SF);
1765 		idx++;
1766 	}
1767 
1768 	err = mlx5_esw_sf_max_hpf_functions(dev, &max_host_pf_sfs, &base_sf_num);
1769 	if (err)
1770 		goto err;
1771 	for (i = 0; i < max_host_pf_sfs; i++) {
1772 		err = mlx5_esw_vport_alloc(esw, idx, base_sf_num + i);
1773 		if (err)
1774 			goto err;
1775 		xa_set_mark(&esw->vports, base_sf_num + i, MLX5_ESW_VPT_SF);
1776 		idx++;
1777 	}
1778 
1779 	if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1780 		int ec_vf_base_num = mlx5_core_ec_vf_vport_base(dev);
1781 
1782 		for (i = 0; i < mlx5_core_max_ec_vfs(esw->dev); i++) {
1783 			err = mlx5_esw_vport_alloc(esw, idx, ec_vf_base_num + i);
1784 			if (err)
1785 				goto err;
1786 			idx++;
1787 		}
1788 	}
1789 
1790 	if (mlx5_ecpf_vport_exists(dev) ||
1791 	    mlx5_core_is_ecpf_esw_manager(dev)) {
1792 		err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_ECPF);
1793 		if (err)
1794 			goto err;
1795 		idx++;
1796 	}
1797 	err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_UPLINK);
1798 	if (err)
1799 		goto err;
1800 	return 0;
1801 
1802 err:
1803 	mlx5_esw_vports_cleanup(esw);
1804 	return err;
1805 }
1806 
1807 static int mlx5_devlink_esw_multiport_set(struct devlink *devlink, u32 id,
1808 					  struct devlink_param_gset_ctx *ctx)
1809 {
1810 	struct mlx5_core_dev *dev = devlink_priv(devlink);
1811 
1812 	if (!MLX5_ESWITCH_MANAGER(dev))
1813 		return -EOPNOTSUPP;
1814 
1815 	if (ctx->val.vbool)
1816 		return mlx5_lag_mpesw_enable(dev);
1817 
1818 	mlx5_lag_mpesw_disable(dev);
1819 	return 0;
1820 }
1821 
1822 static int mlx5_devlink_esw_multiport_get(struct devlink *devlink, u32 id,
1823 					  struct devlink_param_gset_ctx *ctx)
1824 {
1825 	struct mlx5_core_dev *dev = devlink_priv(devlink);
1826 
1827 	ctx->val.vbool = mlx5_lag_is_mpesw(dev);
1828 	return 0;
1829 }
1830 
1831 static const struct devlink_param mlx5_eswitch_params[] = {
1832 	DEVLINK_PARAM_DRIVER(MLX5_DEVLINK_PARAM_ID_ESW_MULTIPORT,
1833 			     "esw_multiport", DEVLINK_PARAM_TYPE_BOOL,
1834 			     BIT(DEVLINK_PARAM_CMODE_RUNTIME),
1835 			     mlx5_devlink_esw_multiport_get,
1836 			     mlx5_devlink_esw_multiport_set, NULL),
1837 };
1838 
1839 int mlx5_eswitch_init(struct mlx5_core_dev *dev)
1840 {
1841 	struct mlx5_eswitch *esw;
1842 	int err;
1843 
1844 	if (!MLX5_VPORT_MANAGER(dev) && !MLX5_ESWITCH_MANAGER(dev))
1845 		return 0;
1846 
1847 	esw = kzalloc(sizeof(*esw), GFP_KERNEL);
1848 	if (!esw)
1849 		return -ENOMEM;
1850 
1851 	err = devl_params_register(priv_to_devlink(dev), mlx5_eswitch_params,
1852 				   ARRAY_SIZE(mlx5_eswitch_params));
1853 	if (err)
1854 		goto free_esw;
1855 
1856 	esw->dev = dev;
1857 	esw->manager_vport = mlx5_eswitch_manager_vport(dev);
1858 	esw->first_host_vport = mlx5_eswitch_first_host_vport_num(dev);
1859 
1860 	esw->debugfs_root = debugfs_create_dir("esw", mlx5_debugfs_get_dev_root(dev));
1861 	esw->work_queue = create_singlethread_workqueue("mlx5_esw_wq");
1862 	if (!esw->work_queue) {
1863 		err = -ENOMEM;
1864 		goto abort;
1865 	}
1866 
1867 	err = mlx5_esw_vports_init(esw);
1868 	if (err)
1869 		goto abort;
1870 
1871 	dev->priv.eswitch = esw;
1872 	err = esw_offloads_init(esw);
1873 	if (err)
1874 		goto reps_err;
1875 
1876 	mutex_init(&esw->offloads.encap_tbl_lock);
1877 	hash_init(esw->offloads.encap_tbl);
1878 	mutex_init(&esw->offloads.decap_tbl_lock);
1879 	hash_init(esw->offloads.decap_tbl);
1880 	mlx5e_mod_hdr_tbl_init(&esw->offloads.mod_hdr);
1881 	atomic64_set(&esw->offloads.num_flows, 0);
1882 	ida_init(&esw->offloads.vport_metadata_ida);
1883 	xa_init_flags(&esw->offloads.vhca_map, XA_FLAGS_ALLOC);
1884 	mutex_init(&esw->state_lock);
1885 	init_rwsem(&esw->mode_lock);
1886 	refcount_set(&esw->qos.refcnt, 0);
1887 
1888 	esw->enabled_vports = 0;
1889 	esw->mode = MLX5_ESWITCH_LEGACY;
1890 	esw->offloads.inline_mode = MLX5_INLINE_MODE_NONE;
1891 	if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, reformat) &&
1892 	    MLX5_CAP_ESW_FLOWTABLE_FDB(dev, decap))
1893 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_BASIC;
1894 	else
1895 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
1896 	BLOCKING_INIT_NOTIFIER_HEAD(&esw->n_head);
1897 
1898 	esw_info(dev,
1899 		 "Total vports %d, per vport: max uc(%d) max mc(%d)\n",
1900 		 esw->total_vports,
1901 		 MLX5_MAX_UC_PER_VPORT(dev),
1902 		 MLX5_MAX_MC_PER_VPORT(dev));
1903 	return 0;
1904 
1905 reps_err:
1906 	mlx5_esw_vports_cleanup(esw);
1907 	dev->priv.eswitch = NULL;
1908 abort:
1909 	if (esw->work_queue)
1910 		destroy_workqueue(esw->work_queue);
1911 	debugfs_remove_recursive(esw->debugfs_root);
1912 	devl_params_unregister(priv_to_devlink(dev), mlx5_eswitch_params,
1913 			       ARRAY_SIZE(mlx5_eswitch_params));
1914 free_esw:
1915 	kfree(esw);
1916 	return err;
1917 }
1918 
1919 void mlx5_eswitch_cleanup(struct mlx5_eswitch *esw)
1920 {
1921 	if (!esw)
1922 		return;
1923 
1924 	esw_info(esw->dev, "cleanup\n");
1925 
1926 	destroy_workqueue(esw->work_queue);
1927 	WARN_ON(refcount_read(&esw->qos.refcnt));
1928 	mutex_destroy(&esw->state_lock);
1929 	WARN_ON(!xa_empty(&esw->offloads.vhca_map));
1930 	xa_destroy(&esw->offloads.vhca_map);
1931 	ida_destroy(&esw->offloads.vport_metadata_ida);
1932 	mlx5e_mod_hdr_tbl_destroy(&esw->offloads.mod_hdr);
1933 	mutex_destroy(&esw->offloads.encap_tbl_lock);
1934 	mutex_destroy(&esw->offloads.decap_tbl_lock);
1935 	esw_offloads_cleanup(esw);
1936 	esw->dev->priv.eswitch = NULL;
1937 	mlx5_esw_vports_cleanup(esw);
1938 	debugfs_remove_recursive(esw->debugfs_root);
1939 	devl_params_unregister(priv_to_devlink(esw->dev), mlx5_eswitch_params,
1940 			       ARRAY_SIZE(mlx5_eswitch_params));
1941 	kfree(esw);
1942 }
1943 
1944 /* Vport Administration */
1945 static int
1946 mlx5_esw_set_vport_mac_locked(struct mlx5_eswitch *esw,
1947 			      struct mlx5_vport *evport, const u8 *mac)
1948 {
1949 	u16 vport_num = evport->vport;
1950 	u64 node_guid;
1951 	int err = 0;
1952 
1953 	if (is_multicast_ether_addr(mac))
1954 		return -EINVAL;
1955 
1956 	if (evport->info.spoofchk && !is_valid_ether_addr(mac))
1957 		mlx5_core_warn(esw->dev,
1958 			       "Set invalid MAC while spoofchk is on, vport(%d)\n",
1959 			       vport_num);
1960 
1961 	err = mlx5_modify_nic_vport_mac_address(esw->dev, vport_num, mac);
1962 	if (err) {
1963 		mlx5_core_warn(esw->dev,
1964 			       "Failed to mlx5_modify_nic_vport_mac vport(%d) err=(%d)\n",
1965 			       vport_num, err);
1966 		return err;
1967 	}
1968 
1969 	node_guid_gen_from_mac(&node_guid, mac);
1970 	err = mlx5_modify_nic_vport_node_guid(esw->dev, vport_num, node_guid);
1971 	if (err)
1972 		mlx5_core_warn(esw->dev,
1973 			       "Failed to set vport %d node guid, err = %d. RDMA_CM will not function properly for this VF.\n",
1974 			       vport_num, err);
1975 
1976 	ether_addr_copy(evport->info.mac, mac);
1977 	evport->info.node_guid = node_guid;
1978 	if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY)
1979 		err = esw_acl_ingress_lgcy_setup(esw, evport);
1980 
1981 	return err;
1982 }
1983 
1984 int mlx5_eswitch_set_vport_mac(struct mlx5_eswitch *esw,
1985 			       u16 vport, const u8 *mac)
1986 {
1987 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
1988 	int err = 0;
1989 
1990 	if (IS_ERR(evport))
1991 		return PTR_ERR(evport);
1992 
1993 	mutex_lock(&esw->state_lock);
1994 	err = mlx5_esw_set_vport_mac_locked(esw, evport, mac);
1995 	mutex_unlock(&esw->state_lock);
1996 	return err;
1997 }
1998 
1999 static bool mlx5_esw_check_port_type(struct mlx5_eswitch *esw, u16 vport_num, xa_mark_t mark)
2000 {
2001 	return xa_get_mark(&esw->vports, vport_num, mark);
2002 }
2003 
2004 bool mlx5_eswitch_is_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2005 {
2006 	return mlx5_esw_check_port_type(esw, vport_num, MLX5_ESW_VPT_VF);
2007 }
2008 
2009 bool mlx5_eswitch_is_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2010 {
2011 	return vport_num == MLX5_VPORT_PF ||
2012 		mlx5_eswitch_is_vf_vport(esw, vport_num);
2013 }
2014 
2015 bool mlx5_esw_is_sf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2016 {
2017 	return mlx5_esw_check_port_type(esw, vport_num, MLX5_ESW_VPT_SF);
2018 }
2019 
2020 int mlx5_eswitch_set_vport_state(struct mlx5_eswitch *esw,
2021 				 u16 vport, int link_state)
2022 {
2023 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2024 	int opmod = MLX5_VPORT_STATE_OP_MOD_ESW_VPORT;
2025 	int other_vport = 1;
2026 	int err = 0;
2027 
2028 	if (!mlx5_esw_allowed(esw))
2029 		return -EPERM;
2030 	if (IS_ERR(evport))
2031 		return PTR_ERR(evport);
2032 
2033 	if (vport == MLX5_VPORT_UPLINK) {
2034 		opmod = MLX5_VPORT_STATE_OP_MOD_UPLINK;
2035 		other_vport = 0;
2036 		vport = 0;
2037 	}
2038 	mutex_lock(&esw->state_lock);
2039 	if (esw->mode != MLX5_ESWITCH_LEGACY) {
2040 		err = -EOPNOTSUPP;
2041 		goto unlock;
2042 	}
2043 
2044 	err = mlx5_modify_vport_admin_state(esw->dev, opmod, vport, other_vport, link_state);
2045 	if (err) {
2046 		mlx5_core_warn(esw->dev, "Failed to set vport %d link state, opmod = %d, err = %d",
2047 			       vport, opmod, err);
2048 		goto unlock;
2049 	}
2050 
2051 	evport->info.link_state = link_state;
2052 
2053 unlock:
2054 	mutex_unlock(&esw->state_lock);
2055 	return err;
2056 }
2057 
2058 int mlx5_eswitch_get_vport_config(struct mlx5_eswitch *esw,
2059 				  u16 vport, struct ifla_vf_info *ivi)
2060 {
2061 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2062 
2063 	if (IS_ERR(evport))
2064 		return PTR_ERR(evport);
2065 
2066 	memset(ivi, 0, sizeof(*ivi));
2067 	ivi->vf = vport - 1;
2068 
2069 	mutex_lock(&esw->state_lock);
2070 	ether_addr_copy(ivi->mac, evport->info.mac);
2071 	ivi->linkstate = evport->info.link_state;
2072 	ivi->vlan = evport->info.vlan;
2073 	ivi->qos = evport->info.qos;
2074 	ivi->spoofchk = evport->info.spoofchk;
2075 	ivi->trusted = evport->info.trusted;
2076 	if (evport->qos.enabled) {
2077 		ivi->min_tx_rate = evport->qos.min_rate;
2078 		ivi->max_tx_rate = evport->qos.max_rate;
2079 	}
2080 	mutex_unlock(&esw->state_lock);
2081 
2082 	return 0;
2083 }
2084 
2085 int __mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch *esw,
2086 				  u16 vport, u16 vlan, u8 qos, u8 set_flags)
2087 {
2088 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2089 	bool vst_mode_steering = esw_vst_mode_is_steering(esw);
2090 	int err = 0;
2091 
2092 	if (IS_ERR(evport))
2093 		return PTR_ERR(evport);
2094 	if (vlan > 4095 || qos > 7)
2095 		return -EINVAL;
2096 
2097 	if (esw->mode == MLX5_ESWITCH_OFFLOADS || !vst_mode_steering) {
2098 		err = modify_esw_vport_cvlan(esw->dev, vport, vlan, qos, set_flags);
2099 		if (err)
2100 			return err;
2101 	}
2102 
2103 	evport->info.vlan = vlan;
2104 	evport->info.qos = qos;
2105 	if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY) {
2106 		err = esw_acl_ingress_lgcy_setup(esw, evport);
2107 		if (err)
2108 			return err;
2109 		err = esw_acl_egress_lgcy_setup(esw, evport);
2110 	}
2111 
2112 	return err;
2113 }
2114 
2115 int mlx5_eswitch_get_vport_stats(struct mlx5_eswitch *esw,
2116 				 u16 vport_num,
2117 				 struct ifla_vf_stats *vf_stats)
2118 {
2119 	struct mlx5_vport *vport = mlx5_eswitch_get_vport(esw, vport_num);
2120 	int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
2121 	u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)] = {};
2122 	struct mlx5_vport_drop_stats stats = {};
2123 	int err = 0;
2124 	u32 *out;
2125 
2126 	if (IS_ERR(vport))
2127 		return PTR_ERR(vport);
2128 
2129 	out = kvzalloc(outlen, GFP_KERNEL);
2130 	if (!out)
2131 		return -ENOMEM;
2132 
2133 	MLX5_SET(query_vport_counter_in, in, opcode,
2134 		 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
2135 	MLX5_SET(query_vport_counter_in, in, op_mod, 0);
2136 	MLX5_SET(query_vport_counter_in, in, vport_number, vport->vport);
2137 	MLX5_SET(query_vport_counter_in, in, other_vport, 1);
2138 
2139 	err = mlx5_cmd_exec_inout(esw->dev, query_vport_counter, in, out);
2140 	if (err)
2141 		goto free_out;
2142 
2143 	#define MLX5_GET_CTR(p, x) \
2144 		MLX5_GET64(query_vport_counter_out, p, x)
2145 
2146 	memset(vf_stats, 0, sizeof(*vf_stats));
2147 	vf_stats->rx_packets =
2148 		MLX5_GET_CTR(out, received_eth_unicast.packets) +
2149 		MLX5_GET_CTR(out, received_ib_unicast.packets) +
2150 		MLX5_GET_CTR(out, received_eth_multicast.packets) +
2151 		MLX5_GET_CTR(out, received_ib_multicast.packets) +
2152 		MLX5_GET_CTR(out, received_eth_broadcast.packets);
2153 
2154 	vf_stats->rx_bytes =
2155 		MLX5_GET_CTR(out, received_eth_unicast.octets) +
2156 		MLX5_GET_CTR(out, received_ib_unicast.octets) +
2157 		MLX5_GET_CTR(out, received_eth_multicast.octets) +
2158 		MLX5_GET_CTR(out, received_ib_multicast.octets) +
2159 		MLX5_GET_CTR(out, received_eth_broadcast.octets);
2160 
2161 	vf_stats->tx_packets =
2162 		MLX5_GET_CTR(out, transmitted_eth_unicast.packets) +
2163 		MLX5_GET_CTR(out, transmitted_ib_unicast.packets) +
2164 		MLX5_GET_CTR(out, transmitted_eth_multicast.packets) +
2165 		MLX5_GET_CTR(out, transmitted_ib_multicast.packets) +
2166 		MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
2167 
2168 	vf_stats->tx_bytes =
2169 		MLX5_GET_CTR(out, transmitted_eth_unicast.octets) +
2170 		MLX5_GET_CTR(out, transmitted_ib_unicast.octets) +
2171 		MLX5_GET_CTR(out, transmitted_eth_multicast.octets) +
2172 		MLX5_GET_CTR(out, transmitted_ib_multicast.octets) +
2173 		MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
2174 
2175 	vf_stats->multicast =
2176 		MLX5_GET_CTR(out, received_eth_multicast.packets) +
2177 		MLX5_GET_CTR(out, received_ib_multicast.packets);
2178 
2179 	vf_stats->broadcast =
2180 		MLX5_GET_CTR(out, received_eth_broadcast.packets);
2181 
2182 	err = mlx5_esw_query_vport_drop_stats(esw->dev, vport, &stats);
2183 	if (err)
2184 		goto free_out;
2185 	vf_stats->rx_dropped = stats.rx_dropped;
2186 	vf_stats->tx_dropped = stats.tx_dropped;
2187 
2188 free_out:
2189 	kvfree(out);
2190 	return err;
2191 }
2192 
2193 u8 mlx5_eswitch_mode(const struct mlx5_core_dev *dev)
2194 {
2195 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2196 
2197 	return mlx5_esw_allowed(esw) ? esw->mode : MLX5_ESWITCH_LEGACY;
2198 }
2199 EXPORT_SYMBOL_GPL(mlx5_eswitch_mode);
2200 
2201 enum devlink_eswitch_encap_mode
2202 mlx5_eswitch_get_encap_mode(const struct mlx5_core_dev *dev)
2203 {
2204 	struct mlx5_eswitch *esw;
2205 
2206 	esw = dev->priv.eswitch;
2207 	return (mlx5_eswitch_mode(dev) == MLX5_ESWITCH_OFFLOADS)  ? esw->offloads.encap :
2208 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2209 }
2210 EXPORT_SYMBOL(mlx5_eswitch_get_encap_mode);
2211 
2212 bool mlx5_esw_multipath_prereq(struct mlx5_core_dev *dev0,
2213 			       struct mlx5_core_dev *dev1)
2214 {
2215 	return (dev0->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS &&
2216 		dev1->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS);
2217 }
2218 
2219 int mlx5_esw_event_notifier_register(struct mlx5_eswitch *esw, struct notifier_block *nb)
2220 {
2221 	return blocking_notifier_chain_register(&esw->n_head, nb);
2222 }
2223 
2224 void mlx5_esw_event_notifier_unregister(struct mlx5_eswitch *esw, struct notifier_block *nb)
2225 {
2226 	blocking_notifier_chain_unregister(&esw->n_head, nb);
2227 }
2228 
2229 /**
2230  * mlx5_esw_hold() - Try to take a read lock on esw mode lock.
2231  * @mdev: mlx5 core device.
2232  *
2233  * Should be called by esw resources callers.
2234  *
2235  * Return: true on success or false.
2236  */
2237 bool mlx5_esw_hold(struct mlx5_core_dev *mdev)
2238 {
2239 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2240 
2241 	/* e.g. VF doesn't have eswitch so nothing to do */
2242 	if (!mlx5_esw_allowed(esw))
2243 		return true;
2244 
2245 	if (down_read_trylock(&esw->mode_lock) != 0) {
2246 		if (esw->eswitch_operation_in_progress) {
2247 			up_read(&esw->mode_lock);
2248 			return false;
2249 		}
2250 		return true;
2251 	}
2252 
2253 	return false;
2254 }
2255 
2256 /**
2257  * mlx5_esw_release() - Release a read lock on esw mode lock.
2258  * @mdev: mlx5 core device.
2259  */
2260 void mlx5_esw_release(struct mlx5_core_dev *mdev)
2261 {
2262 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2263 
2264 	if (mlx5_esw_allowed(esw))
2265 		up_read(&esw->mode_lock);
2266 }
2267 
2268 /**
2269  * mlx5_esw_get() - Increase esw user count.
2270  * @mdev: mlx5 core device.
2271  */
2272 void mlx5_esw_get(struct mlx5_core_dev *mdev)
2273 {
2274 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2275 
2276 	if (mlx5_esw_allowed(esw))
2277 		atomic64_inc(&esw->user_count);
2278 }
2279 
2280 /**
2281  * mlx5_esw_put() - Decrease esw user count.
2282  * @mdev: mlx5 core device.
2283  */
2284 void mlx5_esw_put(struct mlx5_core_dev *mdev)
2285 {
2286 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2287 
2288 	if (mlx5_esw_allowed(esw))
2289 		atomic64_dec_if_positive(&esw->user_count);
2290 }
2291 
2292 /**
2293  * mlx5_esw_try_lock() - Take a write lock on esw mode lock.
2294  * @esw: eswitch device.
2295  *
2296  * Should be called by esw mode change routine.
2297  *
2298  * Return:
2299  * * 0       - esw mode if successfully locked and refcount is 0.
2300  * * -EBUSY  - refcount is not 0.
2301  * * -EINVAL - In the middle of switching mode or lock is already held.
2302  */
2303 int mlx5_esw_try_lock(struct mlx5_eswitch *esw)
2304 {
2305 	if (down_write_trylock(&esw->mode_lock) == 0)
2306 		return -EINVAL;
2307 
2308 	if (esw->eswitch_operation_in_progress ||
2309 	    atomic64_read(&esw->user_count) > 0) {
2310 		up_write(&esw->mode_lock);
2311 		return -EBUSY;
2312 	}
2313 
2314 	return esw->mode;
2315 }
2316 
2317 int mlx5_esw_lock(struct mlx5_eswitch *esw)
2318 {
2319 	down_write(&esw->mode_lock);
2320 
2321 	if (esw->eswitch_operation_in_progress) {
2322 		up_write(&esw->mode_lock);
2323 		return -EBUSY;
2324 	}
2325 
2326 	return 0;
2327 }
2328 
2329 /**
2330  * mlx5_esw_unlock() - Release write lock on esw mode lock
2331  * @esw: eswitch device.
2332  */
2333 void mlx5_esw_unlock(struct mlx5_eswitch *esw)
2334 {
2335 	up_write(&esw->mode_lock);
2336 }
2337 
2338 /**
2339  * mlx5_eswitch_get_total_vports - Get total vports of the eswitch
2340  *
2341  * @dev: Pointer to core device
2342  *
2343  * mlx5_eswitch_get_total_vports returns total number of eswitch vports.
2344  */
2345 u16 mlx5_eswitch_get_total_vports(const struct mlx5_core_dev *dev)
2346 {
2347 	struct mlx5_eswitch *esw;
2348 
2349 	esw = dev->priv.eswitch;
2350 	return mlx5_esw_allowed(esw) ? esw->total_vports : 0;
2351 }
2352 EXPORT_SYMBOL_GPL(mlx5_eswitch_get_total_vports);
2353 
2354 /**
2355  * mlx5_eswitch_get_core_dev - Get the mdev device
2356  * @esw : eswitch device.
2357  *
2358  * Return the mellanox core device which manages the eswitch.
2359  */
2360 struct mlx5_core_dev *mlx5_eswitch_get_core_dev(struct mlx5_eswitch *esw)
2361 {
2362 	return mlx5_esw_allowed(esw) ? esw->dev : NULL;
2363 }
2364 EXPORT_SYMBOL(mlx5_eswitch_get_core_dev);
2365 
2366 bool mlx5_eswitch_block_ipsec(struct mlx5_core_dev *dev)
2367 {
2368 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2369 
2370 	if (!mlx5_esw_allowed(esw))
2371 		return true;
2372 
2373 	mutex_lock(&esw->state_lock);
2374 	if (esw->enabled_ipsec_vf_count) {
2375 		mutex_unlock(&esw->state_lock);
2376 		return false;
2377 	}
2378 
2379 	dev->num_ipsec_offloads++;
2380 	mutex_unlock(&esw->state_lock);
2381 	return true;
2382 }
2383 
2384 void mlx5_eswitch_unblock_ipsec(struct mlx5_core_dev *dev)
2385 {
2386 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2387 
2388 	if (!mlx5_esw_allowed(esw))
2389 		/* Failure means no eswitch => core dev is not a PF */
2390 		return;
2391 
2392 	mutex_lock(&esw->state_lock);
2393 	dev->num_ipsec_offloads--;
2394 	mutex_unlock(&esw->state_lock);
2395 }
2396