xref: /linux/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c (revision 78445023439506ebd83b86d40b1e428a3b309d4a)
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_obj(*spec);
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(%pe)\n",
261 			 dmac_v, dmac_c, vport, 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] = NULL;
537 	struct l2addr_node *node;
538 	struct vport_addr *addr;
539 	struct hlist_head *hash;
540 	struct hlist_node *tmp;
541 	int size = 0;
542 	int err;
543 	int hi;
544 	int i;
545 
546 	hash = is_uc ? vport->uc_list : vport->mc_list;
547 
548 	for_each_l2hash_node(node, tmp, hash, hi) {
549 		addr = container_of(node, struct vport_addr, node);
550 		addr->action = MLX5_ACTION_DEL;
551 	}
552 
553 	if (!vport->enabled)
554 		goto out;
555 
556 	err = mlx5_query_nic_vport_mac_list(esw->dev, vport->vport, list_type,
557 					    &mac_list, &size);
558 	if (err)
559 		goto out;
560 	esw_debug(esw->dev, "vport[%d] context update %s list size (%d)\n",
561 		  vport->vport, is_uc ? "UC" : "MC", size);
562 
563 	for (i = 0; i < size; i++) {
564 		if (is_uc && !is_valid_ether_addr(mac_list[i]))
565 			continue;
566 
567 		if (!is_uc && !is_multicast_ether_addr(mac_list[i]))
568 			continue;
569 
570 		addr = l2addr_hash_find(hash, mac_list[i], struct vport_addr);
571 		if (addr) {
572 			addr->action = MLX5_ACTION_NONE;
573 			/* If this mac was previously added because of allmulti
574 			 * promiscuous rx mode, its now converted to be original
575 			 * vport mac.
576 			 */
577 			if (addr->mc_promisc) {
578 				struct esw_mc_addr *esw_mc =
579 					l2addr_hash_find(esw->mc_table,
580 							 mac_list[i],
581 							 struct esw_mc_addr);
582 				if (!esw_mc) {
583 					esw_warn(esw->dev,
584 						 "Failed to MAC(%pM) in mcast DB\n",
585 						 mac_list[i]);
586 					continue;
587 				}
588 				esw_mc->refcnt++;
589 				addr->mc_promisc = false;
590 			}
591 			continue;
592 		}
593 
594 		addr = l2addr_hash_add(hash, mac_list[i], struct vport_addr,
595 				       GFP_KERNEL);
596 		if (!addr) {
597 			esw_warn(esw->dev,
598 				 "Failed to add MAC(%pM) to vport[%d] DB\n",
599 				 mac_list[i], vport->vport);
600 			continue;
601 		}
602 		addr->vport = vport->vport;
603 		addr->action = MLX5_ACTION_ADD;
604 	}
605 out:
606 	kfree(mac_list);
607 }
608 
609 /* Sync vport UC/MC list from vport context
610  * Must be called after esw_update_vport_addr_list
611  */
612 static void esw_update_vport_mc_promisc(struct mlx5_eswitch *esw,
613 					struct mlx5_vport *vport)
614 {
615 	struct l2addr_node *node;
616 	struct vport_addr *addr;
617 	struct hlist_head *hash;
618 	struct hlist_node *tmp;
619 	int hi;
620 
621 	hash = vport->mc_list;
622 
623 	for_each_l2hash_node(node, tmp, esw->mc_table, hi) {
624 		u8 *mac = node->addr;
625 
626 		addr = l2addr_hash_find(hash, mac, struct vport_addr);
627 		if (addr) {
628 			if (addr->action == MLX5_ACTION_DEL)
629 				addr->action = MLX5_ACTION_NONE;
630 			continue;
631 		}
632 		addr = l2addr_hash_add(hash, mac, struct vport_addr,
633 				       GFP_KERNEL);
634 		if (!addr) {
635 			esw_warn(esw->dev,
636 				 "Failed to add allmulti MAC(%pM) to vport[%d] DB\n",
637 				 mac, vport->vport);
638 			continue;
639 		}
640 		addr->vport = vport->vport;
641 		addr->action = MLX5_ACTION_ADD;
642 		addr->mc_promisc = true;
643 	}
644 }
645 
646 /* Apply vport rx mode to HW FDB table */
647 static void esw_apply_vport_rx_mode(struct mlx5_eswitch *esw,
648 				    struct mlx5_vport *vport,
649 				    bool promisc, bool mc_promisc)
650 {
651 	struct esw_mc_addr *allmulti_addr = &esw->mc_promisc;
652 
653 	if (IS_ERR_OR_NULL(vport->allmulti_rule) != mc_promisc)
654 		goto promisc;
655 
656 	if (mc_promisc) {
657 		vport->allmulti_rule =
658 			esw_fdb_set_vport_allmulti_rule(esw, vport->vport);
659 		if (!allmulti_addr->uplink_rule)
660 			allmulti_addr->uplink_rule =
661 				esw_fdb_set_vport_allmulti_rule(esw,
662 								MLX5_VPORT_UPLINK);
663 		allmulti_addr->refcnt++;
664 	} else if (vport->allmulti_rule) {
665 		mlx5_del_flow_rules(vport->allmulti_rule);
666 		vport->allmulti_rule = NULL;
667 
668 		if (--allmulti_addr->refcnt > 0)
669 			goto promisc;
670 
671 		if (allmulti_addr->uplink_rule)
672 			mlx5_del_flow_rules(allmulti_addr->uplink_rule);
673 		allmulti_addr->uplink_rule = NULL;
674 	}
675 
676 promisc:
677 	if (IS_ERR_OR_NULL(vport->promisc_rule) != promisc)
678 		return;
679 
680 	if (promisc) {
681 		vport->promisc_rule =
682 			esw_fdb_set_vport_promisc_rule(esw, vport->vport);
683 	} else if (vport->promisc_rule) {
684 		mlx5_del_flow_rules(vport->promisc_rule);
685 		vport->promisc_rule = NULL;
686 	}
687 }
688 
689 /* Sync vport rx mode from vport context */
690 static void esw_update_vport_rx_mode(struct mlx5_eswitch *esw,
691 				     struct mlx5_vport *vport)
692 {
693 	int promisc_all = 0;
694 	int promisc_uc = 0;
695 	int promisc_mc = 0;
696 	int err;
697 
698 	err = mlx5_query_nic_vport_promisc(esw->dev,
699 					   vport->vport,
700 					   &promisc_uc,
701 					   &promisc_mc,
702 					   &promisc_all);
703 	if (err)
704 		return;
705 	esw_debug(esw->dev, "vport[%d] context update rx mode promisc_all=%d, all_multi=%d\n",
706 		  vport->vport, promisc_all, promisc_mc);
707 
708 	if (!vport->info.trusted || !vport->enabled) {
709 		promisc_uc = 0;
710 		promisc_mc = 0;
711 		promisc_all = 0;
712 	}
713 
714 	esw_apply_vport_rx_mode(esw, vport, promisc_all,
715 				(promisc_all || promisc_mc));
716 }
717 
718 void esw_vport_change_handle_locked(struct mlx5_vport *vport)
719 {
720 	struct mlx5_core_dev *dev = vport->dev;
721 	struct mlx5_eswitch *esw = dev->priv.eswitch;
722 	u8 mac[ETH_ALEN];
723 
724 	if (!MLX5_CAP_GEN(dev, log_max_l2_table))
725 		return;
726 
727 	mlx5_query_nic_vport_mac_address(dev, vport->vport, true, mac);
728 	esw_debug(dev, "vport[%d] Context Changed: perm mac: %pM\n",
729 		  vport->vport, mac);
730 
731 	if (vport->enabled_events & MLX5_VPORT_UC_ADDR_CHANGE) {
732 		esw_update_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_UC);
733 		esw_apply_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_UC);
734 	}
735 
736 	if (vport->enabled_events & MLX5_VPORT_MC_ADDR_CHANGE)
737 		esw_update_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_MC);
738 
739 	if (vport->enabled_events & MLX5_VPORT_PROMISC_CHANGE) {
740 		esw_update_vport_rx_mode(esw, vport);
741 		if (!IS_ERR_OR_NULL(vport->allmulti_rule))
742 			esw_update_vport_mc_promisc(esw, vport);
743 	}
744 
745 	if (vport->enabled_events & (MLX5_VPORT_PROMISC_CHANGE | MLX5_VPORT_MC_ADDR_CHANGE))
746 		esw_apply_vport_addr_list(esw, vport, MLX5_NVPRT_LIST_TYPE_MC);
747 
748 	esw_debug(esw->dev, "vport[%d] Context Changed: Done\n", vport->vport);
749 	if (vport->enabled)
750 		arm_vport_context_events_cmd(dev, vport->vport,
751 					     vport->enabled_events);
752 }
753 
754 static void esw_vport_change_handler(struct work_struct *work)
755 {
756 	struct mlx5_vport *vport =
757 		container_of(work, struct mlx5_vport, vport_change_handler);
758 	struct mlx5_eswitch *esw = vport->dev->priv.eswitch;
759 
760 	mutex_lock(&esw->state_lock);
761 	esw_vport_change_handle_locked(vport);
762 	mutex_unlock(&esw->state_lock);
763 }
764 
765 static void node_guid_gen_from_mac(u64 *node_guid, const u8 *mac)
766 {
767 	((u8 *)node_guid)[7] = mac[0];
768 	((u8 *)node_guid)[6] = mac[1];
769 	((u8 *)node_guid)[5] = mac[2];
770 	((u8 *)node_guid)[4] = 0xff;
771 	((u8 *)node_guid)[3] = 0xfe;
772 	((u8 *)node_guid)[2] = mac[3];
773 	((u8 *)node_guid)[1] = mac[4];
774 	((u8 *)node_guid)[0] = mac[5];
775 }
776 
777 static int esw_vport_setup_acl(struct mlx5_eswitch *esw,
778 			       struct mlx5_vport *vport)
779 {
780 	if (esw->mode == MLX5_ESWITCH_LEGACY)
781 		return esw_legacy_vport_acl_setup(esw, vport);
782 	else
783 		return esw_vport_create_offloads_acl_tables(esw, vport);
784 }
785 
786 static void esw_vport_cleanup_acl(struct mlx5_eswitch *esw,
787 				  struct mlx5_vport *vport)
788 {
789 	if (esw->mode == MLX5_ESWITCH_LEGACY)
790 		esw_legacy_vport_acl_cleanup(esw, vport);
791 	else
792 		esw_vport_destroy_offloads_acl_tables(esw, vport);
793 }
794 
795 static int mlx5_esw_vport_caps_get(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
796 {
797 	int query_out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
798 	void *query_ctx;
799 	void *hca_caps;
800 	int err;
801 
802 	if (!MLX5_CAP_GEN(esw->dev, vport_group_manager))
803 		return 0;
804 
805 	query_ctx = kzalloc(query_out_sz, GFP_KERNEL);
806 	if (!query_ctx)
807 		return -ENOMEM;
808 
809 	err = mlx5_vport_get_other_func_cap(esw->dev, vport->vport, query_ctx,
810 					    MLX5_CAP_GENERAL);
811 	if (err)
812 		goto out_free;
813 
814 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
815 	vport->info.roce_enabled = MLX5_GET(cmd_hca_cap, hca_caps, roce);
816 	vport->vhca_id = MLX5_GET(cmd_hca_cap, hca_caps, vhca_id);
817 
818 	if (!MLX5_CAP_GEN_MAX(esw->dev, hca_cap_2))
819 		goto out_free;
820 
821 	memset(query_ctx, 0, query_out_sz);
822 	err = mlx5_vport_get_other_func_cap(esw->dev, vport->vport, query_ctx,
823 					    MLX5_CAP_GENERAL_2);
824 	if (err)
825 		goto out_free;
826 
827 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
828 	vport->info.mig_enabled = MLX5_GET(cmd_hca_cap_2, hca_caps, migratable);
829 
830 	err = mlx5_esw_ipsec_vf_offload_get(esw->dev, vport);
831 out_free:
832 	kfree(query_ctx);
833 	return err;
834 }
835 
836 bool mlx5_esw_vport_vhca_id(struct mlx5_eswitch *esw, u16 vportn, u16 *vhca_id)
837 {
838 	struct mlx5_vport *vport;
839 
840 	vport = mlx5_eswitch_get_vport(esw, vportn);
841 	if (IS_ERR(vport) || MLX5_VPORT_INVAL_VHCA_ID(vport))
842 		return false;
843 
844 	*vhca_id = vport->vhca_id;
845 	return true;
846 }
847 
848 static enum mlx5_func_type
849 esw_vport_to_func_type(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
850 {
851 	u16 vport_num = vport->vport;
852 
853 	if (vport_num == MLX5_VPORT_HOST_PF)
854 		return MLX5_HOST_PF;
855 	if (xa_get_mark(&esw->vports, vport_num, MLX5_ESW_VPT_SF))
856 		return MLX5_SF;
857 	if (xa_get_mark(&esw->vports, vport_num, MLX5_ESW_VPT_VF))
858 		return MLX5_VF;
859 	if (mlx5_esw_is_spf_vport(esw, vport_num))
860 		return MLX5_SPF;
861 	return MLX5_EC_VF;
862 }
863 
864 u16 mlx5_esw_vhca_id_to_func_type(struct mlx5_core_dev *dev, u16 vhca_id)
865 {
866 	struct mlx5_eswitch *esw = dev->priv.eswitch;
867 	void *entry;
868 
869 	if (vhca_id == MLX5_CAP_GEN(dev, vhca_id))
870 		return MLX5_SELF;
871 
872 	if (!esw)
873 		return MLX5_FUNC_TYPE_NONE;
874 
875 	entry = xa_load(&esw->vhca_type_map, vhca_id);
876 	if (entry)
877 		return xa_to_value(entry);
878 
879 	return MLX5_FUNC_TYPE_NONE;
880 }
881 
882 static int esw_vport_setup(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
883 {
884 	bool vst_mode_steering = esw_vst_mode_is_steering(esw);
885 	u16 vport_num = vport->vport;
886 	int flags;
887 	int err;
888 
889 	err = esw_vport_setup_acl(esw, vport);
890 	if (err)
891 		return err;
892 
893 	if (mlx5_esw_is_manager_vport(esw, vport_num))
894 		return 0;
895 
896 	err = mlx5_esw_vport_caps_get(esw, vport);
897 	if (err)
898 		goto err_caps;
899 
900 	mlx5_modify_vport_admin_state(esw->dev,
901 				      MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
902 				      vport_num, 1,
903 				      vport->info.link_state);
904 
905 	mlx5_query_nic_vport_mac_address(esw->dev, vport_num, true,
906 					 vport->info.mac);
907 	mlx5_query_nic_vport_node_guid(esw->dev, vport_num, true,
908 				       &vport->info.node_guid);
909 
910 	flags = (vport->info.vlan || vport->info.qos) ?
911 		SET_VLAN_STRIP | SET_VLAN_INSERT : 0;
912 	if (esw->mode == MLX5_ESWITCH_OFFLOADS || !vst_mode_steering)
913 		modify_esw_vport_cvlan(esw->dev, vport_num, vport->info.vlan,
914 				       vport->info.qos, flags);
915 
916 	return 0;
917 
918 err_caps:
919 	esw_vport_cleanup_acl(esw, vport);
920 	return err;
921 }
922 
923 /* Don't cleanup vport->info, it's needed to restore vport configuration */
924 static void esw_vport_cleanup(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
925 {
926 	u16 vport_num = vport->vport;
927 
928 	if (!mlx5_esw_is_manager_vport(esw, vport_num))
929 		mlx5_modify_vport_admin_state(esw->dev,
930 					      MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
931 					      vport_num, 1,
932 					      MLX5_VPORT_ADMIN_STATE_DOWN);
933 
934 	mlx5_esw_qos_vport_disable(vport);
935 	esw_vport_cleanup_acl(esw, vport);
936 }
937 
938 static void mlx5_esw_vport_set_max_tx_speed(struct mlx5_eswitch *esw,
939 					    struct mlx5_vport *vport)
940 {
941 	int ret;
942 
943 	if (!MLX5_CAP_ESW(esw->dev, esw_vport_state_max_tx_speed))
944 		return;
945 
946 	ret = mlx5_modify_vport_max_tx_speed(esw->dev,
947 					     MLX5_VPORT_STATE_OP_MOD_ESW_VPORT,
948 					     vport->vport, true,
949 					     vport->agg_max_tx_speed);
950 	if (ret)
951 		mlx5_core_dbg(esw->dev,
952 			      "Failed to set vport %d speed %d, err=%d\n",
953 			      vport->vport, vport->agg_max_tx_speed, ret);
954 }
955 
956 int mlx5_esw_vport_enable(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
957 			  enum mlx5_eswitch_vport_event enabled_events)
958 {
959 	u16 vport_num = vport->vport;
960 	int ret;
961 
962 	mutex_lock(&esw->state_lock);
963 	WARN_ON(vport->enabled);
964 
965 	esw_debug(esw->dev, "Enabling VPORT(%d)\n", vport_num);
966 
967 	ret = esw_vport_setup(esw, vport);
968 	if (ret)
969 		goto done;
970 
971 	/* Sync with current vport context */
972 	vport->enabled_events = enabled_events;
973 	vport->enabled = true;
974 	if (mlx5_eswitch_is_vf_vport(esw, vport_num) &&
975 	    (vport->info.ipsec_crypto_enabled || vport->info.ipsec_packet_enabled))
976 		esw->enabled_ipsec_vf_count++;
977 
978 	/* Esw manager is trusted by default. Host PF (vport 0) is trusted as well
979 	 * in smartNIC as it's a vport group manager.
980 	 */
981 	if (mlx5_esw_is_manager_vport(esw, vport_num) ||
982 	    (!vport_num && mlx5_core_is_ecpf(esw->dev)))
983 		vport->info.trusted = true;
984 
985 	if (!mlx5_esw_is_manager_vport(esw, vport_num) &&
986 	    MLX5_CAP_GEN(esw->dev, vport_group_manager)) {
987 		ret = mlx5_esw_vport_vhca_id_map(esw, vport);
988 		if (ret)
989 			goto err_vhca_mapping;
990 		ret = xa_insert(&esw->vhca_type_map, vport->vhca_id,
991 				xa_mk_value(esw_vport_to_func_type(esw, vport)),
992 				GFP_KERNEL);
993 		if (ret)
994 			goto err_type_map;
995 	}
996 
997 	esw_vport_change_handle_locked(vport);
998 
999 	esw->enabled_vports++;
1000 	esw_debug(esw->dev, "Enabled VPORT(%d)\n", vport_num);
1001 
1002 	if (vport->agg_max_tx_speed)
1003 		mlx5_esw_vport_set_max_tx_speed(esw, vport);
1004 done:
1005 	mutex_unlock(&esw->state_lock);
1006 	return ret;
1007 
1008 err_type_map:
1009 	mlx5_esw_vport_vhca_id_unmap(esw, vport);
1010 err_vhca_mapping:
1011 	esw_vport_cleanup(esw, vport);
1012 	mutex_unlock(&esw->state_lock);
1013 	return ret;
1014 }
1015 
1016 void mlx5_esw_vport_disable(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1017 {
1018 	u16 vport_num = vport->vport;
1019 
1020 	mutex_lock(&esw->state_lock);
1021 
1022 	if (!vport->enabled)
1023 		goto done;
1024 
1025 	esw_debug(esw->dev, "Disabling vport(%d)\n", vport_num);
1026 	/* Mark this vport as disabled to discard new events */
1027 	vport->enabled = false;
1028 
1029 	/* Disable events from this vport */
1030 	if (MLX5_CAP_GEN(esw->dev, log_max_l2_table))
1031 		arm_vport_context_events_cmd(esw->dev, vport_num, 0);
1032 
1033 	if (!mlx5_esw_is_manager_vport(esw, vport_num) &&
1034 	    MLX5_CAP_GEN(esw->dev, vport_group_manager)) {
1035 		xa_erase(&esw->vhca_type_map, vport->vhca_id);
1036 		mlx5_esw_vport_vhca_id_unmap(esw, vport);
1037 	}
1038 
1039 	if (mlx5_eswitch_is_vf_vport(esw, vport_num) &&
1040 	    (vport->info.ipsec_crypto_enabled || vport->info.ipsec_packet_enabled))
1041 		esw->enabled_ipsec_vf_count--;
1042 
1043 	/* Clear rx-mode before esw_vport_change_handle_locked(): on
1044 	 * MLX5_VPORT_PROMISC_CHANGE it calls esw_update_vport_mc_promisc()
1045 	 * when vport->allmulti_rule is set, repopulating mc_list with FDB
1046 	 * rules that dangle once the FDB is destroyed. NULL allmulti_rule
1047 	 * here skips that path.
1048 	 */
1049 	esw_apply_vport_rx_mode(esw, vport, false, false);
1050 	/* We don't assume VFs will cleanup after themselves.
1051 	 * Calling vport change handler while vport is disabled will cleanup
1052 	 * the vport resources.
1053 	 */
1054 	esw_vport_change_handle_locked(vport);
1055 	vport->enabled_events = 0;
1056 	esw_vport_cleanup(esw, vport);
1057 	esw->enabled_vports--;
1058 
1059 done:
1060 	mutex_unlock(&esw->state_lock);
1061 }
1062 
1063 static int eswitch_vport_event(struct notifier_block *nb,
1064 			       unsigned long type, void *data)
1065 {
1066 	struct mlx5_eswitch *esw = mlx5_nb_cof(nb, struct mlx5_eswitch, nb);
1067 	struct mlx5_eqe *eqe = data;
1068 	struct mlx5_vport *vport;
1069 	u16 vport_num;
1070 
1071 	vport_num = be16_to_cpu(eqe->data.vport_change.vport_num);
1072 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1073 	if (!IS_ERR(vport))
1074 		queue_work(esw->work_queue, &vport->vport_change_handler);
1075 	return NOTIFY_OK;
1076 }
1077 
1078 /**
1079  * mlx5_esw_query_functions - Returns raw output about functions state
1080  * @dev:	Pointer to device to query
1081  *
1082  * mlx5_esw_query_functions() allocates and returns functions changed
1083  * raw output memory pointer from device on success. Otherwise returns ERR_PTR.
1084  * Caller must free the memory using kvfree() when valid pointer is returned.
1085  */
1086 const u32 *mlx5_esw_query_functions(struct mlx5_core_dev *dev)
1087 {
1088 	bool net_func_v1 = MLX5_CAP_GEN(dev, query_host_net_function_v1);
1089 	u32 in[MLX5_ST_SZ_DW(query_esw_functions_in)] = {};
1090 	int alloc_entries;
1091 	int outlen;
1092 	u32 *out;
1093 	int err;
1094 
1095 	if (net_func_v1) {
1096 		alloc_entries = MLX5_CAP_GEN(dev,
1097 					     query_host_net_function_num_max);
1098 		alloc_entries = max(alloc_entries, 1);
1099 		MLX5_SET(query_esw_functions_in, in, op_mod,
1100 			 MLX5_QUERY_ESW_FUNC_OP_MOD_LAYOUT_V1);
1101 		outlen = MLX5_BYTE_OFF(query_esw_functions_out,
1102 				       net_function_params) +
1103 			 alloc_entries * MLX5_UN_SZ_BYTES(net_function_params);
1104 		outlen = max_t(int, outlen,
1105 			       MLX5_ST_SZ_BYTES(query_esw_functions_out));
1106 	} else {
1107 		outlen = MLX5_ST_SZ_BYTES(query_esw_functions_out);
1108 	}
1109 
1110 	out = kvzalloc(outlen, GFP_KERNEL);
1111 	if (!out)
1112 		return ERR_PTR(-ENOMEM);
1113 
1114 	MLX5_SET(query_esw_functions_in, in, opcode,
1115 		 MLX5_CMD_OP_QUERY_ESW_FUNCTIONS);
1116 
1117 	err = mlx5_cmd_exec(dev, in, sizeof(in), out, outlen);
1118 	if (err)
1119 		goto free;
1120 
1121 	if (net_func_v1) {
1122 		int num_entries;
1123 
1124 		num_entries = MLX5_GET(query_esw_functions_out, out,
1125 				       net_function_num);
1126 		if (num_entries > alloc_entries) {
1127 			mlx5_core_warn(dev, "Got %d entries, max expected %d\n",
1128 				       num_entries, alloc_entries);
1129 			err = -EINVAL;
1130 			goto free;
1131 		}
1132 	}
1133 
1134 	return out;
1135 
1136 free:
1137 	kvfree(out);
1138 	return ERR_PTR(err);
1139 }
1140 
1141 static struct mlx5_esw_pf_info
1142 mlx5_esw_host_pf_from_host_params(const void *entry)
1143 {
1144 	return (struct mlx5_esw_pf_info) {
1145 		.pf_not_exist = MLX5_GET(host_params_context, entry,
1146 					 host_pf_not_exist),
1147 		.pf_disabled = MLX5_GET(host_params_context, entry,
1148 					host_pf_disabled),
1149 		.num_of_vfs = MLX5_GET(host_params_context, entry,
1150 				       host_num_of_vfs),
1151 		.total_vfs = MLX5_GET(host_params_context, entry,
1152 				      host_total_vfs),
1153 		.host_number = MLX5_GET(host_params_context, entry,
1154 					host_number),
1155 	};
1156 }
1157 
1158 static struct mlx5_esw_pf_info
1159 mlx5_esw_host_pf_from_net_func_params(const u8 *entry, int num_entries)
1160 {
1161 	int i;
1162 
1163 	for (i = 0; i < num_entries; i++) {
1164 		int pf_type, state;
1165 
1166 		pf_type = MLX5_GET(network_function_params, entry, pci_pf_type);
1167 		if (pf_type != MLX5_PCI_PF_TYPE_EXTERNAL_HOST_PF) {
1168 			entry += MLX5_UN_SZ_BYTES(net_function_params);
1169 			continue;
1170 		}
1171 
1172 		state = MLX5_GET(network_function_params, entry, vhca_state);
1173 
1174 		return (struct mlx5_esw_pf_info) {
1175 			.pf_disabled = state != MLX5_VHCA_STATE_IN_USE,
1176 			.num_of_vfs = MLX5_GET(network_function_params,
1177 					       entry, pci_num_vfs),
1178 			.total_vfs = MLX5_GET(network_function_params,
1179 					      entry, pci_total_vfs),
1180 			.host_number = MLX5_GET(network_function_params,
1181 						entry, host_number),
1182 			.pf_num = MLX5_GET(network_function_params, entry,
1183 					   pci_device_function),
1184 		};
1185 	}
1186 
1187 	/* No external host PF entry found */
1188 	return (struct mlx5_esw_pf_info) {
1189 		.pf_not_exist = true,
1190 		.pf_disabled = true,
1191 	};
1192 }
1193 
1194 struct mlx5_esw_pf_info
1195 mlx5_esw_get_host_pf_info(struct mlx5_core_dev *dev, const u32 *out)
1196 {
1197 	const void *entry;
1198 
1199 	entry = MLX5_ADDR_OF(query_esw_functions_out, out, net_function_params);
1200 
1201 	if (MLX5_CAP_GEN(dev, query_host_net_function_v1)) {
1202 		int num_entries = MLX5_GET(query_esw_functions_out, out,
1203 					   net_function_num);
1204 
1205 		return mlx5_esw_host_pf_from_net_func_params(entry,
1206 							     num_entries);
1207 	}
1208 
1209 	return mlx5_esw_host_pf_from_host_params(entry);
1210 }
1211 
1212 bool mlx5_esw_get_spf_disabled(struct mlx5_core_dev *dev, const u32 *out,
1213 			       u16 vhca_id)
1214 {
1215 	int num_entries;
1216 	const u8 *entry;
1217 	int i;
1218 
1219 	num_entries = MLX5_GET(query_esw_functions_out, out, net_function_num);
1220 	entry = MLX5_ADDR_OF(query_esw_functions_out, out, net_function_params);
1221 
1222 	for (i = 0; i < num_entries; i++) {
1223 		u16 entry_vhca_id = MLX5_GET(network_function_params,
1224 					     entry, vhca_id);
1225 
1226 		if (entry_vhca_id == vhca_id) {
1227 			int state;
1228 
1229 			state = MLX5_GET(network_function_params, entry,
1230 					 vhca_state);
1231 			return state != MLX5_VHCA_STATE_IN_USE;
1232 		}
1233 		entry += MLX5_UN_SZ_BYTES(net_function_params);
1234 	}
1235 
1236 	return true;
1237 }
1238 
1239 static int mlx5_esw_host_functions_enabled_query(struct mlx5_eswitch *esw)
1240 {
1241 	struct mlx5_esw_pf_info host_pf_info;
1242 	const u32 *query_host_out;
1243 
1244 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev))
1245 		return 0;
1246 
1247 	query_host_out = mlx5_esw_query_functions(esw->dev);
1248 	if (IS_ERR(query_host_out))
1249 		return PTR_ERR(query_host_out);
1250 
1251 	host_pf_info = mlx5_esw_get_host_pf_info(esw->dev, query_host_out);
1252 	esw->esw_funcs.host_funcs_disabled = host_pf_info.pf_not_exist;
1253 
1254 	kvfree(query_host_out);
1255 	return 0;
1256 }
1257 
1258 static void mlx5_eswitch_event_handler_register(struct mlx5_eswitch *esw)
1259 {
1260 	if (esw->mode == MLX5_ESWITCH_OFFLOADS && mlx5_eswitch_is_funcs_handler(esw->dev)) {
1261 		MLX5_NB_INIT(&esw->esw_funcs.nb, mlx5_esw_funcs_changed_handler,
1262 			     ESW_FUNCTIONS_CHANGED);
1263 		mlx5_eq_notifier_register(esw->dev, &esw->esw_funcs.nb);
1264 	}
1265 }
1266 
1267 static void mlx5_eswitch_invalidate_wq(struct mlx5_eswitch *esw)
1268 {
1269 	atomic_inc(&esw->generation);
1270 	wake_up_all(&esw->work_queue_wait);
1271 }
1272 
1273 static void mlx5_eswitch_event_handler_unregister(struct mlx5_eswitch *esw)
1274 {
1275 	if (esw->mode == MLX5_ESWITCH_OFFLOADS &&
1276 	    mlx5_eswitch_is_funcs_handler(esw->dev))
1277 		mlx5_eq_notifier_unregister(esw->dev, &esw->esw_funcs.nb);
1278 }
1279 
1280 static void mlx5_eswitch_clear_vf_vports_info(struct mlx5_eswitch *esw)
1281 {
1282 	struct mlx5_vport *vport;
1283 	unsigned long i;
1284 
1285 	mlx5_esw_for_each_vf_vport(esw, i, vport, esw->esw_funcs.num_vfs) {
1286 		mlx5_esw_qos_vport_qos_free(vport);
1287 		memset(&vport->info, 0, sizeof(vport->info));
1288 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1289 	}
1290 }
1291 
1292 static void mlx5_eswitch_clear_ec_vf_vports_info(struct mlx5_eswitch *esw)
1293 {
1294 	struct mlx5_vport *vport;
1295 	unsigned long i;
1296 
1297 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, esw->esw_funcs.num_ec_vfs) {
1298 		mlx5_esw_qos_vport_qos_free(vport);
1299 		memset(&vport->info, 0, sizeof(vport->info));
1300 		vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
1301 	}
1302 }
1303 
1304 static int mlx5_eswitch_load_vport(struct mlx5_eswitch *esw, struct mlx5_vport *vport,
1305 				   enum mlx5_eswitch_vport_event enabled_events)
1306 {
1307 	int err;
1308 
1309 	err = mlx5_esw_vport_enable(esw, vport, enabled_events);
1310 	if (err)
1311 		return err;
1312 
1313 	err = mlx5_esw_offloads_load_rep(esw, vport);
1314 	if (err)
1315 		goto err_rep;
1316 
1317 	return err;
1318 
1319 err_rep:
1320 	mlx5_esw_vport_disable(esw, vport);
1321 	return err;
1322 }
1323 
1324 static void mlx5_eswitch_unload_vport(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
1325 {
1326 	mlx5_esw_offloads_unload_rep(esw, vport);
1327 	mlx5_esw_vport_disable(esw, vport);
1328 }
1329 
1330 static int mlx5_eswitch_load_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num,
1331 					 enum mlx5_eswitch_vport_event enabled_events)
1332 {
1333 	struct mlx5_vport *vport;
1334 	int err;
1335 
1336 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1337 	if (IS_ERR(vport))
1338 		return PTR_ERR(vport);
1339 
1340 	err = mlx5_esw_offloads_init_pf_vf_rep(esw, vport);
1341 	if (err)
1342 		return err;
1343 
1344 	err = mlx5_eswitch_load_vport(esw, vport, enabled_events);
1345 	if (err)
1346 		goto err_load;
1347 	return 0;
1348 
1349 err_load:
1350 	mlx5_esw_offloads_cleanup_pf_vf_rep(esw, vport);
1351 	return err;
1352 }
1353 
1354 static void mlx5_eswitch_unload_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
1355 {
1356 	struct mlx5_vport *vport;
1357 
1358 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1359 	if (IS_ERR(vport))
1360 		return;
1361 
1362 	mlx5_eswitch_unload_vport(esw, vport);
1363 	mlx5_esw_offloads_cleanup_pf_vf_rep(esw, vport);
1364 }
1365 
1366 int mlx5_eswitch_load_sf_vport(struct mlx5_eswitch *esw, u16 vport_num,
1367 			       enum mlx5_eswitch_vport_event enabled_events,
1368 			       struct mlx5_devlink_port *dl_port, u32 controller, u32 sfnum)
1369 {
1370 	struct mlx5_vport *vport;
1371 	int err;
1372 
1373 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1374 	if (IS_ERR(vport))
1375 		return PTR_ERR(vport);
1376 
1377 	err = mlx5_esw_offloads_init_sf_rep(esw, vport, dl_port, controller, sfnum);
1378 	if (err)
1379 		return err;
1380 
1381 	err = mlx5_eswitch_load_vport(esw, vport, enabled_events);
1382 	if (err)
1383 		goto err_load;
1384 
1385 	return 0;
1386 
1387 err_load:
1388 	mlx5_esw_offloads_cleanup_sf_rep(esw, vport);
1389 	return err;
1390 }
1391 
1392 void mlx5_eswitch_unload_sf_vport(struct mlx5_eswitch *esw, u16 vport_num)
1393 {
1394 	struct mlx5_vport *vport;
1395 
1396 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1397 	if (IS_ERR(vport))
1398 		return;
1399 
1400 	mlx5_eswitch_unload_vport(esw, vport);
1401 	mlx5_esw_offloads_cleanup_sf_rep(esw, vport);
1402 }
1403 
1404 void mlx5_eswitch_unload_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs)
1405 {
1406 	struct mlx5_vport *vport;
1407 	unsigned long i;
1408 
1409 	mlx5_esw_for_each_vf_vport(esw, i, vport, num_vfs) {
1410 		/* Adjacent VFs are unloaded separately */
1411 		if (!vport->enabled || vport->adjacent)
1412 			continue;
1413 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1414 	}
1415 }
1416 
1417 static void mlx5_eswitch_unload_ec_vf_vports(struct mlx5_eswitch *esw,
1418 					     u16 num_ec_vfs)
1419 {
1420 	struct mlx5_vport *vport;
1421 	unsigned long i;
1422 
1423 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, num_ec_vfs) {
1424 		if (!vport->enabled)
1425 			continue;
1426 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1427 	}
1428 }
1429 
1430 static void mlx5_eswitch_unload_adj_vf_vports(struct mlx5_eswitch *esw)
1431 {
1432 	struct mlx5_vport *vport;
1433 	unsigned long i;
1434 
1435 	mlx5_esw_for_each_vf_vport(esw, i, vport, U16_MAX) {
1436 		if (!vport->enabled || !vport->adjacent)
1437 			continue;
1438 		mlx5_eswitch_unload_pf_vf_vport(esw, vport->vport);
1439 	}
1440 }
1441 
1442 static int
1443 mlx5_eswitch_load_adj_vf_vports(struct mlx5_eswitch *esw,
1444 				enum mlx5_eswitch_vport_event enabled_events)
1445 {
1446 	struct mlx5_vport *vport;
1447 	unsigned long i;
1448 	int err;
1449 
1450 	mlx5_esw_for_each_vf_vport(esw, i, vport, U16_MAX) {
1451 		if (!vport->adjacent)
1452 			continue;
1453 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport,
1454 						    enabled_events);
1455 		if (err)
1456 			goto unload_adj_vf_vport;
1457 	}
1458 
1459 	return 0;
1460 
1461 unload_adj_vf_vport:
1462 	mlx5_eswitch_unload_adj_vf_vports(esw);
1463 	return err;
1464 }
1465 
1466 int mlx5_eswitch_load_vf_vports(struct mlx5_eswitch *esw, u16 num_vfs,
1467 				enum mlx5_eswitch_vport_event enabled_events)
1468 {
1469 	struct mlx5_vport *vport;
1470 	unsigned long i;
1471 	int err;
1472 
1473 	mlx5_esw_for_each_vf_vport(esw, i, vport, num_vfs) {
1474 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport, enabled_events);
1475 		if (err)
1476 			goto vf_err;
1477 	}
1478 
1479 	return 0;
1480 
1481 vf_err:
1482 	mlx5_eswitch_unload_vf_vports(esw, num_vfs);
1483 	return err;
1484 }
1485 
1486 static int mlx5_eswitch_load_ec_vf_vports(struct mlx5_eswitch *esw, u16 num_ec_vfs,
1487 					  enum mlx5_eswitch_vport_event enabled_events)
1488 {
1489 	struct mlx5_vport *vport;
1490 	unsigned long i;
1491 	int err;
1492 
1493 	mlx5_esw_for_each_ec_vf_vport(esw, i, vport, num_ec_vfs) {
1494 		err = mlx5_eswitch_load_pf_vf_vport(esw, vport->vport, enabled_events);
1495 		if (err)
1496 			goto vf_err;
1497 	}
1498 
1499 	return 0;
1500 
1501 vf_err:
1502 	mlx5_eswitch_unload_ec_vf_vports(esw, num_ec_vfs);
1503 	return err;
1504 }
1505 
1506 int mlx5_esw_pf_enable_hca(struct mlx5_core_dev *dev, u16 vport_num)
1507 {
1508 	struct mlx5_eswitch *esw = dev->priv.eswitch;
1509 	struct mlx5_vport *vport;
1510 	int err;
1511 
1512 	if (!mlx5_core_is_ecpf(dev) || !mlx5_esw_allowed(esw))
1513 		return 0;
1514 
1515 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1516 	if (IS_ERR(vport))
1517 		return PTR_ERR(vport);
1518 
1519 	/* Once vport and representor are ready, take the PF out of
1520 	 * initializing state. Enabling HCA clears the iser->initializing
1521 	 * bit and PF driver loading can progress.
1522 	 */
1523 	err = mlx5_cmd_pf_enable_hca(dev, vport_num);
1524 	if (err)
1525 		return err;
1526 
1527 	vport->pf_activated = true;
1528 
1529 	return 0;
1530 }
1531 
1532 int mlx5_esw_pf_disable_hca(struct mlx5_core_dev *dev, u16 vport_num)
1533 {
1534 	struct mlx5_eswitch *esw = dev->priv.eswitch;
1535 	struct mlx5_vport *vport;
1536 	int err;
1537 
1538 	if (!mlx5_core_is_ecpf(dev) || !mlx5_esw_allowed(esw))
1539 		return 0;
1540 
1541 	vport = mlx5_eswitch_get_vport(esw, vport_num);
1542 	if (IS_ERR(vport))
1543 		return PTR_ERR(vport);
1544 
1545 	err = mlx5_cmd_pf_disable_hca(dev, vport_num);
1546 	if (err)
1547 		return err;
1548 
1549 	vport->pf_activated = false;
1550 
1551 	return 0;
1552 }
1553 
1554 int mlx5_esw_host_pf_enable_hca(struct mlx5_core_dev *dev)
1555 {
1556 	return mlx5_esw_pf_enable_hca(dev, MLX5_VPORT_HOST_PF);
1557 }
1558 
1559 int mlx5_esw_host_pf_disable_hca(struct mlx5_core_dev *dev)
1560 {
1561 	return mlx5_esw_pf_disable_hca(dev, MLX5_VPORT_HOST_PF);
1562 }
1563 
1564 /* mlx5_eswitch_enable_pf_vf_vports() enables vports of PF, ECPF and VFs
1565  * whichever are present on the eswitch.
1566  */
1567 int
1568 mlx5_eswitch_enable_pf_vf_vports(struct mlx5_eswitch *esw,
1569 				 enum mlx5_eswitch_vport_event enabled_events)
1570 {
1571 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
1572 	bool pf_needed;
1573 	u16 vport_num;
1574 	int ret;
1575 	int i;
1576 
1577 	pf_needed = mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1578 		    esw->mode == MLX5_ESWITCH_LEGACY;
1579 
1580 	/* Enable PF vport */
1581 	if (pf_needed && mlx5_esw_host_functions_enabled(esw->dev)) {
1582 		ret = mlx5_eswitch_load_pf_vf_vport(esw, MLX5_VPORT_HOST_PF,
1583 						    enabled_events);
1584 		if (ret)
1585 			return ret;
1586 	}
1587 
1588 	if (mlx5_esw_host_functions_enabled(esw->dev)) {
1589 		/* Enable external host PF HCA */
1590 		ret = mlx5_esw_host_pf_enable_hca(esw->dev);
1591 		if (ret)
1592 			goto pf_hca_err;
1593 	}
1594 
1595 	/* Enable ECPF vport */
1596 	if (mlx5_ecpf_vport_exists(esw->dev)) {
1597 		ret = mlx5_eswitch_load_pf_vf_vport(esw, MLX5_VPORT_ECPF, enabled_events);
1598 		if (ret)
1599 			goto ecpf_err;
1600 	}
1601 
1602 	/* Enable ECVF vports */
1603 	if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1604 		ret = mlx5_eswitch_load_ec_vf_vports(esw,
1605 						     esw_funcs->num_ec_vfs,
1606 						     enabled_events);
1607 		if (ret)
1608 			goto ec_vf_err;
1609 	}
1610 
1611 	/* Enable VF vports */
1612 	ret = mlx5_eswitch_load_vf_vports(esw, esw_funcs->num_vfs,
1613 					  enabled_events);
1614 	if (ret)
1615 		goto vf_err;
1616 
1617 	/* Enable adjacent VF vports */
1618 	ret = mlx5_eswitch_load_adj_vf_vports(esw, enabled_events);
1619 	if (ret)
1620 		goto unload_vf_vports;
1621 
1622 	/* Enable satellite PF vports */
1623 	for (i = 0; i < esw_funcs->num_spfs; i++) {
1624 		vport_num = esw_funcs->spfs[i].vport_num;
1625 
1626 		ret = mlx5_eswitch_load_pf_vf_vport(esw, vport_num,
1627 						    enabled_events);
1628 		if (ret)
1629 			goto spf_err;
1630 
1631 		ret = mlx5_esw_pf_enable_hca(esw->dev, vport_num);
1632 		if (ret) {
1633 			mlx5_eswitch_unload_pf_vf_vport(esw, vport_num);
1634 			goto spf_err;
1635 		}
1636 	}
1637 
1638 	return 0;
1639 
1640 spf_err:
1641 	while (i-- > 0) {
1642 		vport_num = esw_funcs->spfs[i].vport_num;
1643 		mlx5_esw_pf_disable_hca(esw->dev, vport_num);
1644 		mlx5_eswitch_unload_pf_vf_vport(esw, vport_num);
1645 	}
1646 	mlx5_eswitch_unload_adj_vf_vports(esw);
1647 unload_vf_vports:
1648 	mlx5_eswitch_unload_vf_vports(esw, esw_funcs->num_vfs);
1649 vf_err:
1650 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1651 		mlx5_eswitch_unload_ec_vf_vports(esw, esw_funcs->num_ec_vfs);
1652 ec_vf_err:
1653 	if (mlx5_ecpf_vport_exists(esw->dev))
1654 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_ECPF);
1655 ecpf_err:
1656 	if (mlx5_esw_host_functions_enabled(esw->dev))
1657 		mlx5_esw_host_pf_disable_hca(esw->dev);
1658 pf_hca_err:
1659 	if (pf_needed && mlx5_esw_host_functions_enabled(esw->dev))
1660 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_HOST_PF);
1661 	return ret;
1662 }
1663 
1664 /* mlx5_eswitch_disable_pf_vf_vports() disables vports of PF, ECPF and VFs
1665  * whichever are previously enabled on the eswitch.
1666  */
1667 void mlx5_eswitch_disable_pf_vf_vports(struct mlx5_eswitch *esw)
1668 {
1669 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
1670 	u16 vport_num;
1671 	int i;
1672 
1673 	for (i = 0; i < esw_funcs->num_spfs; i++) {
1674 		vport_num = esw_funcs->spfs[i].vport_num;
1675 		mlx5_esw_pf_disable_hca(esw->dev, vport_num);
1676 		mlx5_eswitch_unload_pf_vf_vport(esw, vport_num);
1677 	}
1678 
1679 	mlx5_eswitch_unload_adj_vf_vports(esw);
1680 
1681 	mlx5_eswitch_unload_vf_vports(esw, esw_funcs->num_vfs);
1682 
1683 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1684 		mlx5_eswitch_unload_ec_vf_vports(esw, esw_funcs->num_ec_vfs);
1685 
1686 	if (mlx5_ecpf_vport_exists(esw->dev)) {
1687 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_ECPF);
1688 	}
1689 
1690 	if (mlx5_esw_host_functions_enabled(esw->dev))
1691 		mlx5_esw_host_pf_disable_hca(esw->dev);
1692 
1693 	if ((mlx5_core_is_ecpf_esw_manager(esw->dev) ||
1694 	     esw->mode == MLX5_ESWITCH_LEGACY) &&
1695 	    mlx5_esw_host_functions_enabled(esw->dev))
1696 		mlx5_eswitch_unload_pf_vf_vport(esw, MLX5_VPORT_HOST_PF);
1697 }
1698 
1699 static void mlx5_eswitch_get_devlink_param(struct mlx5_eswitch *esw)
1700 {
1701 	struct devlink *devlink = priv_to_devlink(esw->dev);
1702 	union devlink_param_value val;
1703 	int err;
1704 
1705 	err = devl_param_driverinit_value_get(devlink,
1706 					      MLX5_DEVLINK_PARAM_ID_ESW_LARGE_GROUP_NUM,
1707 					      &val);
1708 	if (!err) {
1709 		esw->params.large_group_num = val.vu32;
1710 	} else {
1711 		esw_warn(esw->dev,
1712 			 "Devlink can't get param fdb_large_groups, uses default (%d).\n",
1713 			 ESW_OFFLOADS_DEFAULT_NUM_GROUPS);
1714 		esw->params.large_group_num = ESW_OFFLOADS_DEFAULT_NUM_GROUPS;
1715 	}
1716 }
1717 
1718 static void
1719 mlx5_eswitch_update_num_of_vfs(struct mlx5_eswitch *esw, int num_vfs)
1720 {
1721 	struct mlx5_esw_pf_info host_pf_info;
1722 	const u32 *out;
1723 
1724 	if (num_vfs < 0)
1725 		return;
1726 
1727 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev)) {
1728 		esw->esw_funcs.num_vfs = num_vfs;
1729 		return;
1730 	}
1731 
1732 	out = mlx5_esw_query_functions(esw->dev);
1733 	if (IS_ERR(out))
1734 		return;
1735 
1736 	host_pf_info = mlx5_esw_get_host_pf_info(esw->dev, out);
1737 	esw->esw_funcs.num_vfs = host_pf_info.num_of_vfs;
1738 	if (mlx5_core_ec_sriov_enabled(esw->dev))
1739 		esw->esw_funcs.num_ec_vfs = num_vfs;
1740 
1741 	kvfree(out);
1742 }
1743 
1744 static void mlx5_esw_mode_change_notify(struct mlx5_eswitch *esw, u16 mode)
1745 {
1746 	struct mlx5_esw_event_info info = {};
1747 
1748 	info.new_mode = mode;
1749 
1750 	blocking_notifier_call_chain(&esw->dev->priv.esw_n_head, 0, &info);
1751 }
1752 
1753 static int mlx5_esw_egress_acls_init(struct mlx5_core_dev *dev)
1754 {
1755 	struct mlx5_flow_steering *steering = dev->priv.steering;
1756 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1757 	int err;
1758 	int i;
1759 
1760 	for (i = 0; i < total_vports; i++) {
1761 		err = mlx5_fs_vport_egress_acl_ns_add(steering, i);
1762 		if (err)
1763 			goto acl_ns_remove;
1764 	}
1765 	return 0;
1766 
1767 acl_ns_remove:
1768 	while (i--)
1769 		mlx5_fs_vport_egress_acl_ns_remove(steering, i);
1770 	return err;
1771 }
1772 
1773 static void mlx5_esw_egress_acls_cleanup(struct mlx5_core_dev *dev)
1774 {
1775 	struct mlx5_flow_steering *steering = dev->priv.steering;
1776 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1777 	int i;
1778 
1779 	for (i = total_vports - 1; i >= 0; i--)
1780 		mlx5_fs_vport_egress_acl_ns_remove(steering, i);
1781 }
1782 
1783 static int mlx5_esw_ingress_acls_init(struct mlx5_core_dev *dev)
1784 {
1785 	struct mlx5_flow_steering *steering = dev->priv.steering;
1786 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1787 	int err;
1788 	int i;
1789 
1790 	for (i = 0; i < total_vports; i++) {
1791 		err = mlx5_fs_vport_ingress_acl_ns_add(steering, i);
1792 		if (err)
1793 			goto acl_ns_remove;
1794 	}
1795 	return 0;
1796 
1797 acl_ns_remove:
1798 	while (i--)
1799 		mlx5_fs_vport_ingress_acl_ns_remove(steering, i);
1800 	return err;
1801 }
1802 
1803 static void mlx5_esw_ingress_acls_cleanup(struct mlx5_core_dev *dev)
1804 {
1805 	struct mlx5_flow_steering *steering = dev->priv.steering;
1806 	int total_vports = mlx5_eswitch_get_total_vports(dev);
1807 	int i;
1808 
1809 	for (i = total_vports - 1; i >= 0; i--)
1810 		mlx5_fs_vport_ingress_acl_ns_remove(steering, i);
1811 }
1812 
1813 static int mlx5_esw_acls_ns_init(struct mlx5_eswitch *esw)
1814 {
1815 	struct mlx5_core_dev *dev = esw->dev;
1816 	int err;
1817 
1818 	if (esw->flags & MLX5_ESWITCH_VPORT_ACL_NS_CREATED)
1819 		return 0;
1820 
1821 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) {
1822 		err = mlx5_esw_egress_acls_init(dev);
1823 		if (err)
1824 			return err;
1825 	} else {
1826 		esw_warn(dev, "egress ACL is not supported by FW\n");
1827 	}
1828 
1829 	if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) {
1830 		err = mlx5_esw_ingress_acls_init(dev);
1831 		if (err)
1832 			goto err;
1833 	} else {
1834 		esw_warn(dev, "ingress ACL is not supported by FW\n");
1835 	}
1836 	esw->flags |= MLX5_ESWITCH_VPORT_ACL_NS_CREATED;
1837 	return 0;
1838 
1839 err:
1840 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
1841 		mlx5_esw_egress_acls_cleanup(dev);
1842 	return err;
1843 }
1844 
1845 static void mlx5_esw_acls_ns_cleanup(struct mlx5_eswitch *esw)
1846 {
1847 	struct mlx5_core_dev *dev = esw->dev;
1848 
1849 	esw->flags &= ~MLX5_ESWITCH_VPORT_ACL_NS_CREATED;
1850 	if (MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support))
1851 		mlx5_esw_ingress_acls_cleanup(dev);
1852 	if (MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support))
1853 		mlx5_esw_egress_acls_cleanup(dev);
1854 }
1855 
1856 /**
1857  * mlx5_eswitch_enable_locked - Enable eswitch
1858  * @esw:	Pointer to eswitch
1859  * @num_vfs:	Enable eswitch for given number of VFs. This is optional.
1860  *		Valid value are 0, > 0 and MLX5_ESWITCH_IGNORE_NUM_VFS.
1861  *		Caller should pass num_vfs > 0 when enabling eswitch for
1862  *		vf vports. Caller should pass num_vfs = 0, when eswitch
1863  *		is enabled without sriov VFs or when caller
1864  *		is unaware of the sriov state of the host PF on ECPF based
1865  *		eswitch. Caller should pass < 0 when num_vfs should be
1866  *		completely ignored. This is typically the case when eswitch
1867  *		is enabled without sriov regardless of PF/ECPF system.
1868  * mlx5_eswitch_enable_locked() Enables eswitch in either legacy or offloads
1869  * mode. If num_vfs >=0 is provided, it setup VF related eswitch vports.
1870  * It returns 0 on success or error code on failure.
1871  */
1872 int mlx5_eswitch_enable_locked(struct mlx5_eswitch *esw, int num_vfs)
1873 {
1874 	int err;
1875 
1876 	devl_assert_locked(priv_to_devlink(esw->dev));
1877 
1878 	if (!MLX5_CAP_ESW_FLOWTABLE_FDB(esw->dev, ft_support)) {
1879 		esw_warn(esw->dev, "FDB is not supported, aborting ...\n");
1880 		return -EOPNOTSUPP;
1881 	}
1882 
1883 	mlx5_eswitch_get_devlink_param(esw);
1884 
1885 	err = mlx5_esw_acls_ns_init(esw);
1886 	if (err)
1887 		return err;
1888 
1889 	mlx5_eswitch_update_num_of_vfs(esw, num_vfs);
1890 
1891 	MLX5_NB_INIT(&esw->nb, eswitch_vport_event, NIC_VPORT_CHANGE);
1892 	mlx5_eq_notifier_register(esw->dev, &esw->nb);
1893 
1894 	if (esw->mode == MLX5_ESWITCH_LEGACY) {
1895 		err = esw_legacy_enable(esw);
1896 	} else {
1897 		err = esw_offloads_enable(esw);
1898 	}
1899 
1900 	if (err)
1901 		goto err_esw_init;
1902 
1903 	esw->fdb_table.flags |= MLX5_ESW_FDB_CREATED;
1904 
1905 	mlx5_eswitch_event_handler_register(esw);
1906 
1907 	esw_info(esw->dev, "Enable: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1908 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1909 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1910 
1911 	mlx5_esw_mode_change_notify(esw, esw->mode);
1912 
1913 	return 0;
1914 
1915 err_esw_init:
1916 	mlx5_eq_notifier_unregister(esw->dev, &esw->nb);
1917 	mlx5_esw_acls_ns_cleanup(esw);
1918 	return err;
1919 }
1920 
1921 /**
1922  * mlx5_eswitch_enable - Enable eswitch
1923  * @esw:	Pointer to eswitch
1924  * @num_vfs:	Enable eswitch switch for given number of VFs.
1925  *		Caller must pass num_vfs > 0 when enabling eswitch for
1926  *		vf vports.
1927  * mlx5_eswitch_enable() returns 0 on success or error code on failure.
1928  */
1929 int mlx5_eswitch_enable(struct mlx5_eswitch *esw, int num_vfs)
1930 {
1931 	bool toggle_lag;
1932 	int ret = 0;
1933 
1934 	if (!mlx5_esw_allowed(esw))
1935 		return 0;
1936 
1937 	devl_assert_locked(priv_to_devlink(esw->dev));
1938 
1939 	toggle_lag = !mlx5_esw_is_fdb_created(esw);
1940 
1941 	if (toggle_lag)
1942 		mlx5_lag_disable_change(esw->dev);
1943 
1944 	mlx5_eswitch_invalidate_wq(esw);
1945 	mlx5_esw_reps_block(esw);
1946 
1947 	if (!mlx5_esw_is_fdb_created(esw)) {
1948 		ret = mlx5_eswitch_enable_locked(esw, num_vfs);
1949 	} else {
1950 		enum mlx5_eswitch_vport_event vport_events;
1951 
1952 		vport_events = (esw->mode == MLX5_ESWITCH_LEGACY) ?
1953 					MLX5_LEGACY_SRIOV_VPORT_EVENTS : MLX5_VPORT_UC_ADDR_CHANGE;
1954 		/* If this is the ECPF the number of host VFs is managed via the
1955 		 * eswitch function change event handler, and any num_vfs provided
1956 		 * here are intended to be EC VFs.
1957 		 */
1958 		if (!mlx5_core_is_ecpf(esw->dev)) {
1959 			ret = mlx5_eswitch_load_vf_vports(esw, num_vfs, vport_events);
1960 			if (!ret)
1961 				esw->esw_funcs.num_vfs = num_vfs;
1962 		} else if (mlx5_core_ec_sriov_enabled(esw->dev)) {
1963 			ret = mlx5_eswitch_load_ec_vf_vports(esw, num_vfs, vport_events);
1964 			if (!ret)
1965 				esw->esw_funcs.num_ec_vfs = num_vfs;
1966 		}
1967 	}
1968 
1969 	mlx5_esw_reps_unblock(esw);
1970 
1971 	if (toggle_lag)
1972 		mlx5_lag_enable_change(esw->dev);
1973 
1974 	return ret;
1975 }
1976 
1977 /* When disabling sriov, free driver level resources. */
1978 void mlx5_eswitch_disable_sriov(struct mlx5_eswitch *esw, bool clear_vf)
1979 {
1980 	if (!mlx5_esw_allowed(esw))
1981 		return;
1982 
1983 	devl_assert_locked(priv_to_devlink(esw->dev));
1984 	/* If driver is unloaded, this function is called twice by remove_one()
1985 	 * and mlx5_unload(). Prevent the second call.
1986 	 */
1987 	if (!esw->esw_funcs.num_vfs && !esw->esw_funcs.num_ec_vfs && !clear_vf)
1988 		return;
1989 
1990 	esw_info(esw->dev, "Unload vfs: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
1991 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
1992 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
1993 
1994 	mlx5_eswitch_invalidate_wq(esw);
1995 
1996 	if (esw->mode == MLX5_ESWITCH_OFFLOADS) {
1997 		struct devlink *devlink = priv_to_devlink(esw->dev);
1998 
1999 		devl_rate_nodes_destroy(devlink);
2000 	}
2001 
2002 	mlx5_esw_reps_block(esw);
2003 
2004 	if (!mlx5_core_is_ecpf(esw->dev)) {
2005 		mlx5_eswitch_unload_vf_vports(esw, esw->esw_funcs.num_vfs);
2006 		if (clear_vf)
2007 			mlx5_eswitch_clear_vf_vports_info(esw);
2008 	} else if (mlx5_core_ec_sriov_enabled(esw->dev)) {
2009 		mlx5_eswitch_unload_ec_vf_vports(esw, esw->esw_funcs.num_ec_vfs);
2010 		if (clear_vf)
2011 			mlx5_eswitch_clear_ec_vf_vports_info(esw);
2012 	}
2013 
2014 	mlx5_esw_reps_unblock(esw);
2015 	/* Destroy legacy fdb when disabling sriov in legacy mode. */
2016 	if (esw->mode == MLX5_ESWITCH_LEGACY)
2017 		mlx5_eswitch_disable_locked(esw);
2018 
2019 	if (!mlx5_core_is_ecpf(esw->dev))
2020 		esw->esw_funcs.num_vfs = 0;
2021 	else
2022 		esw->esw_funcs.num_ec_vfs = 0;
2023 }
2024 
2025 /* Free resources for corresponding eswitch mode. It is called by devlink
2026  * when changing eswitch mode or modprobe when unloading driver.
2027  */
2028 void mlx5_eswitch_disable_locked(struct mlx5_eswitch *esw)
2029 {
2030 	struct devlink *devlink = priv_to_devlink(esw->dev);
2031 
2032 	/* Notify eswitch users that it is exiting from current mode.
2033 	 * So that it can do necessary cleanup before the eswitch is disabled.
2034 	 */
2035 	mlx5_esw_mode_change_notify(esw, MLX5_ESWITCH_LEGACY);
2036 
2037 	mlx5_eq_notifier_unregister(esw->dev, &esw->nb);
2038 	mlx5_eswitch_event_handler_unregister(esw);
2039 	mlx5_eswitch_invalidate_wq(esw);
2040 
2041 	esw_info(esw->dev, "Disable: mode(%s), nvfs(%d), necvfs(%d), active vports(%d)\n",
2042 		 esw->mode == MLX5_ESWITCH_LEGACY ? "LEGACY" : "OFFLOADS",
2043 		 esw->esw_funcs.num_vfs, esw->esw_funcs.num_ec_vfs, esw->enabled_vports);
2044 
2045 	if (esw->mode == MLX5_ESWITCH_OFFLOADS)
2046 		devl_rate_nodes_destroy(devlink);
2047 
2048 	if (esw->fdb_table.flags & MLX5_ESW_FDB_CREATED) {
2049 		esw->fdb_table.flags &= ~MLX5_ESW_FDB_CREATED;
2050 		if (esw->mode == MLX5_ESWITCH_OFFLOADS)
2051 			esw_offloads_disable(esw);
2052 		else if (esw->mode == MLX5_ESWITCH_LEGACY)
2053 			esw_legacy_disable(esw);
2054 		mlx5_esw_acls_ns_cleanup(esw);
2055 	}
2056 }
2057 
2058 void mlx5_eswitch_disable(struct mlx5_eswitch *esw)
2059 {
2060 	if (!mlx5_esw_allowed(esw))
2061 		return;
2062 
2063 	devl_assert_locked(priv_to_devlink(esw->dev));
2064 	mlx5_lag_disable_change(esw->dev);
2065 
2066 	mlx5_esw_reps_block(esw);
2067 	mlx5_eswitch_disable_locked(esw);
2068 	mlx5_esw_reps_unblock(esw);
2069 
2070 	esw->mode = MLX5_ESWITCH_LEGACY;
2071 	mlx5_sd_eswitch_mode_set(esw->dev, MLX5_ESWITCH_LEGACY);
2072 	mlx5_lag_enable_change(esw->dev);
2073 }
2074 
2075 static int mlx5_esw_sf_max_pf_functions(struct mlx5_core_dev *dev,
2076 					u16 vport_num, u16 *max_sfs,
2077 					u16 *sf_base_id)
2078 {
2079 	int query_out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out);
2080 	void *query_ctx;
2081 	void *hca_caps;
2082 	int err;
2083 
2084 	query_ctx = kzalloc(query_out_sz, GFP_KERNEL);
2085 	if (!query_ctx)
2086 		return -ENOMEM;
2087 
2088 	err = mlx5_vport_get_other_func_general_cap(dev, vport_num, query_ctx);
2089 	if (err)
2090 		goto out_free;
2091 
2092 	hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability);
2093 	*max_sfs = MLX5_GET(cmd_hca_cap, hca_caps, max_num_sf);
2094 	*sf_base_id = MLX5_GET(cmd_hca_cap, hca_caps, sf_base_id);
2095 
2096 out_free:
2097 	kfree(query_ctx);
2098 	return err;
2099 }
2100 
2101 int mlx5_esw_sf_max_hpf_functions(struct mlx5_core_dev *dev, u16 *max_sfs,
2102 				  u16 *sf_base_id)
2103 {
2104 	if (!mlx5_core_is_ecpf(dev) ||
2105 	    !mlx5_esw_host_functions_enabled(dev)) {
2106 		*max_sfs = 0;
2107 		return 0;
2108 	}
2109 
2110 	return mlx5_esw_sf_max_pf_functions(dev, MLX5_VPORT_HOST_PF, max_sfs,
2111 					    sf_base_id);
2112 }
2113 
2114 int mlx5_esw_sf_max_spf_functions(struct mlx5_core_dev *dev, int spf_idx,
2115 				  u16 *max_sfs, u16 *sf_base_id)
2116 {
2117 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2118 	u16 vport_num;
2119 
2120 	if (!mlx5_esw_allowed(esw)) {
2121 		*max_sfs = 0;
2122 		return 0;
2123 	}
2124 
2125 	if (spf_idx >= esw->esw_funcs.num_spfs)
2126 		return -EINVAL;
2127 
2128 	vport_num = esw->esw_funcs.spfs[spf_idx].vport_num;
2129 	return mlx5_esw_sf_max_pf_functions(dev, vport_num, max_sfs,
2130 					    sf_base_id);
2131 }
2132 
2133 int mlx5_esw_get_num_spfs(struct mlx5_core_dev *dev)
2134 {
2135 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2136 
2137 	if (!mlx5_esw_allowed(esw))
2138 		return 0;
2139 
2140 	return esw->esw_funcs.num_spfs;
2141 }
2142 
2143 int mlx5_esw_spf_get_host_number(struct mlx5_core_dev *dev, int spf_idx,
2144 				 u16 *host_number)
2145 {
2146 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2147 
2148 	if (!mlx5_esw_allowed(esw))
2149 		return -EPERM;
2150 
2151 	if (spf_idx >= esw->esw_funcs.num_spfs)
2152 		return -EINVAL;
2153 
2154 	*host_number = esw->esw_funcs.spfs[spf_idx].host_number;
2155 	return 0;
2156 }
2157 
2158 u16 mlx5_esw_get_hpf_host_number(struct mlx5_core_dev *dev)
2159 {
2160 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2161 
2162 	if (!mlx5_esw_allowed(esw))
2163 		return 0;
2164 
2165 	return esw->esw_funcs.hpf_host_number;
2166 }
2167 
2168 u16 mlx5_esw_get_hpf_pf_num(struct mlx5_core_dev *dev)
2169 {
2170 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2171 
2172 	if (mlx5_core_is_ecpf_esw_manager(dev) &&
2173 	    MLX5_CAP_GEN(dev, query_host_net_function_v1))
2174 		return esw->esw_funcs.hpf_pf_num;
2175 
2176 	return PCI_FUNC(dev->pdev->devfn);
2177 }
2178 
2179 u16 mlx5_esw_sf_controller_to_pfnum(struct mlx5_core_dev *dev, u32 controller)
2180 {
2181 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2182 	struct mlx5_esw_functions *esw_funcs;
2183 	int i;
2184 
2185 	if (!controller)
2186 		return PCI_FUNC(dev->pdev->devfn);
2187 
2188 	esw_funcs = &esw->esw_funcs;
2189 	for (i = 0; i < esw_funcs->num_spfs; i++)
2190 		if (controller == esw_funcs->spfs[i].host_number + 1)
2191 			return esw_funcs->spfs[i].pf_num;
2192 
2193 	return mlx5_esw_get_hpf_pf_num(dev);
2194 }
2195 
2196 bool mlx5_esw_has_spf_sfs(struct mlx5_core_dev *dev)
2197 {
2198 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2199 
2200 	if (!mlx5_esw_allowed(esw))
2201 		return false;
2202 
2203 	return esw->esw_funcs.has_spf_sfs;
2204 }
2205 
2206 static int mlx5_esw_hpf_info_init(struct mlx5_eswitch *esw)
2207 {
2208 	struct mlx5_esw_pf_info host_pf_info;
2209 	const u32 *query_host_out;
2210 
2211 	if (!mlx5_core_is_ecpf_esw_manager(esw->dev))
2212 		return 0;
2213 
2214 	query_host_out = mlx5_esw_query_functions(esw->dev);
2215 	if (IS_ERR(query_host_out))
2216 		return PTR_ERR(query_host_out);
2217 
2218 	/* Mark non local controller with non zero controller number. */
2219 	host_pf_info = mlx5_esw_get_host_pf_info(esw->dev, query_host_out);
2220 	esw->esw_funcs.hpf_host_number = host_pf_info.host_number;
2221 	esw->esw_funcs.hpf_pf_num = host_pf_info.pf_num;
2222 	kvfree(query_host_out);
2223 	return 0;
2224 }
2225 
2226 int mlx5_esw_vport_alloc(struct mlx5_eswitch *esw, int index, u16 vport_num)
2227 {
2228 	struct mlx5_vport *vport;
2229 	int err;
2230 
2231 	vport = kzalloc_obj(*vport);
2232 	if (!vport)
2233 		return -ENOMEM;
2234 
2235 	vport->dev = esw->dev;
2236 	vport->vport = vport_num;
2237 	vport->index = index;
2238 	vport->info.link_state = MLX5_VPORT_ADMIN_STATE_AUTO;
2239 	vport->vhca_id = MLX5_VHCA_ID_INVALID;
2240 	INIT_WORK(&vport->vport_change_handler, esw_vport_change_handler);
2241 	err = xa_insert(&esw->vports, vport_num, vport, GFP_KERNEL);
2242 	if (err)
2243 		goto insert_err;
2244 
2245 	esw->total_vports++;
2246 	return 0;
2247 
2248 insert_err:
2249 	kfree(vport);
2250 	return err;
2251 }
2252 
2253 void mlx5_esw_vport_free(struct mlx5_eswitch *esw, struct mlx5_vport *vport)
2254 {
2255 	esw->total_vports--;
2256 	xa_erase(&esw->vports, vport->vport);
2257 	kfree(vport);
2258 }
2259 
2260 static void mlx5_esw_spfs_cleanup(struct mlx5_eswitch *esw)
2261 {
2262 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
2263 	int i;
2264 
2265 	for (i = 0; i < esw_funcs->num_spfs; i++)
2266 		mlx5_esw_destroy_esw_vport(esw->dev,
2267 					   esw_funcs->spfs[i].vport_num);
2268 
2269 	kfree(esw_funcs->spfs);
2270 	esw_funcs->spfs = NULL;
2271 	esw_funcs->num_spfs = 0;
2272 }
2273 
2274 static int mlx5_esw_spfs_init(struct mlx5_eswitch *esw)
2275 {
2276 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
2277 	struct mlx5_core_dev *dev = esw->dev;
2278 	int num_entries;
2279 	const u8 *entry;
2280 	const u32 *out;
2281 	int err = 0;
2282 	int pf_type;
2283 	u16 vhca_id;
2284 	int i;
2285 
2286 	if (!MLX5_CAP_GEN(dev, query_host_net_function_v1))
2287 		return 0;
2288 
2289 	out = mlx5_esw_query_functions(dev);
2290 	if (IS_ERR(out))
2291 		return PTR_ERR(out);
2292 
2293 	num_entries = MLX5_GET(query_esw_functions_out, out, net_function_num);
2294 	if (!num_entries)
2295 		goto out_free;
2296 
2297 	esw_funcs->spfs = kzalloc_objs(*esw_funcs->spfs, num_entries);
2298 	if (!esw_funcs->spfs) {
2299 		err = -ENOMEM;
2300 		goto out_free;
2301 	}
2302 
2303 	entry = MLX5_ADDR_OF(query_esw_functions_out, out, net_function_params);
2304 
2305 	for (i = 0; i < num_entries; i++) {
2306 		u16 vport_num;
2307 
2308 		pf_type = MLX5_GET(network_function_params, entry, pci_pf_type);
2309 		if (pf_type != MLX5_PCI_PF_TYPE_SATELLITE_PF) {
2310 			entry += MLX5_UN_SZ_BYTES(net_function_params);
2311 			continue;
2312 		}
2313 
2314 		if (!MLX5_GET(network_function_params, entry,
2315 			      esw_vport_manual)) {
2316 			esw_warn(dev, "Satellite PF without esw_vport_manual is not supported\n");
2317 			entry += MLX5_UN_SZ_BYTES(net_function_params);
2318 			continue;
2319 		}
2320 
2321 		vhca_id = MLX5_GET(network_function_params, entry, vhca_id);
2322 
2323 		err = mlx5_esw_create_esw_vport(dev, vhca_id, &vport_num);
2324 		if (err) {
2325 			esw_warn(dev, "Failed to create satellite PF vport for vhca_id 0x%x, err %d\n",
2326 				 vhca_id, err);
2327 			goto spfs_cleanup;
2328 		}
2329 
2330 		esw_funcs->spfs[esw_funcs->num_spfs].vport_num = vport_num;
2331 		esw_funcs->spfs[esw_funcs->num_spfs].vhca_id = vhca_id;
2332 		esw_funcs->spfs[esw_funcs->num_spfs].host_number =
2333 			MLX5_GET(network_function_params, entry, host_number);
2334 		esw_funcs->spfs[esw_funcs->num_spfs].pf_num =
2335 			MLX5_GET(network_function_params, entry,
2336 				 pci_device_function);
2337 		esw_funcs->num_spfs++;
2338 
2339 		entry += MLX5_UN_SZ_BYTES(net_function_params);
2340 	}
2341 
2342 	if (!esw_funcs->num_spfs) {
2343 		kfree(esw_funcs->spfs);
2344 		esw_funcs->spfs = NULL;
2345 	}
2346 
2347 	kvfree(out);
2348 	return 0;
2349 
2350 spfs_cleanup:
2351 	mlx5_esw_spfs_cleanup(esw);
2352 out_free:
2353 	kvfree(out);
2354 	return err;
2355 }
2356 
2357 static void mlx5_esw_vports_cleanup(struct mlx5_eswitch *esw)
2358 {
2359 	struct mlx5_vport *vport;
2360 	unsigned long i;
2361 
2362 	mlx5_esw_spfs_cleanup(esw);
2363 	esw->esw_funcs.has_spf_sfs = false;
2364 	mlx5_esw_for_each_vport(esw, i, vport)
2365 		mlx5_esw_vport_free(esw, vport);
2366 	xa_destroy(&esw->vports);
2367 }
2368 
2369 static int mlx5_esw_vports_init(struct mlx5_eswitch *esw)
2370 {
2371 	struct mlx5_core_dev *dev = esw->dev;
2372 	u16 max_sfs, base_sf_num;
2373 	int idx = 0;
2374 	int err;
2375 	int i;
2376 
2377 	xa_init(&esw->vports);
2378 
2379 	err = mlx5_esw_hpf_info_init(esw);
2380 	if (err)
2381 		goto err;
2382 
2383 	if (mlx5_esw_host_functions_enabled(dev)) {
2384 		err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_HOST_PF);
2385 		if (err)
2386 			goto err;
2387 		if (esw->first_host_vport == MLX5_VPORT_HOST_PF)
2388 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_HOST_FN);
2389 		idx++;
2390 		for (i = 0; i < mlx5_core_max_vfs(dev); i++) {
2391 			err = mlx5_esw_vport_alloc(esw, idx, idx);
2392 			if (err)
2393 				goto err;
2394 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_VF);
2395 			xa_set_mark(&esw->vports, idx, MLX5_ESW_VPT_HOST_FN);
2396 			idx++;
2397 		}
2398 	}
2399 
2400 	base_sf_num = mlx5_sf_start_function_id(dev);
2401 	for (i = 0; i < mlx5_sf_max_functions(dev); i++) {
2402 		err = mlx5_esw_vport_alloc(esw, idx, base_sf_num + i);
2403 		if (err)
2404 			goto err;
2405 		xa_set_mark(&esw->vports, base_sf_num + i, MLX5_ESW_VPT_SF);
2406 		idx++;
2407 	}
2408 
2409 	err = mlx5_esw_sf_max_hpf_functions(dev, &max_sfs, &base_sf_num);
2410 	if (err)
2411 		goto err;
2412 	for (i = 0; i < max_sfs; i++) {
2413 		err = mlx5_esw_vport_alloc(esw, idx, base_sf_num + i);
2414 		if (err)
2415 			goto err;
2416 		xa_set_mark(&esw->vports, base_sf_num + i, MLX5_ESW_VPT_SF);
2417 		idx++;
2418 	}
2419 
2420 	err = mlx5_esw_spfs_init(esw);
2421 	if (err)
2422 		goto err;
2423 
2424 	for (i = 0; i < esw->esw_funcs.num_spfs; i++) {
2425 		struct mlx5_vport *vport;
2426 		u16 vport_num;
2427 
2428 		vport_num = esw->esw_funcs.spfs[i].vport_num;
2429 		err = mlx5_esw_vport_alloc(esw, idx++, vport_num);
2430 		if (err)
2431 			goto err;
2432 		vport = mlx5_eswitch_get_vport(esw, vport_num);
2433 		vport->vhca_id = esw->esw_funcs.spfs[i].vhca_id;
2434 
2435 		err = mlx5_esw_sf_max_spf_functions(dev, i,
2436 						    &max_sfs, &base_sf_num);
2437 		if (err)
2438 			goto err;
2439 		if (max_sfs)
2440 			esw->esw_funcs.has_spf_sfs = true;
2441 		for (int j = 0; j < max_sfs; j++) {
2442 			err = mlx5_esw_vport_alloc(esw, idx,
2443 						   base_sf_num + j);
2444 			if (err)
2445 				goto err;
2446 			xa_set_mark(&esw->vports, base_sf_num + j,
2447 				    MLX5_ESW_VPT_SF);
2448 			idx++;
2449 		}
2450 	}
2451 
2452 	if (mlx5_core_ec_sriov_enabled(esw->dev)) {
2453 		int ec_vf_base_num = mlx5_core_ec_vf_vport_base(dev);
2454 
2455 		for (i = 0; i < mlx5_core_max_ec_vfs(esw->dev); i++) {
2456 			err = mlx5_esw_vport_alloc(esw, idx, ec_vf_base_num + i);
2457 			if (err)
2458 				goto err;
2459 			idx++;
2460 		}
2461 	}
2462 
2463 	if (mlx5_ecpf_vport_exists(dev) ||
2464 	    mlx5_core_is_ecpf_esw_manager(dev)) {
2465 		err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_ECPF);
2466 		if (err)
2467 			goto err;
2468 		idx++;
2469 	}
2470 	err = mlx5_esw_vport_alloc(esw, idx, MLX5_VPORT_UPLINK);
2471 	if (err)
2472 		goto err;
2473 
2474 	/* Adjacent vports or other dynamically create vports will use this */
2475 	esw->last_vport_idx = ++idx;
2476 	return 0;
2477 
2478 err:
2479 	mlx5_esw_vports_cleanup(esw);
2480 	return err;
2481 }
2482 
2483 static int mlx5_devlink_esw_multiport_set(struct devlink *devlink, u32 id,
2484 					  struct devlink_param_gset_ctx *ctx,
2485 					  struct netlink_ext_ack *extack)
2486 {
2487 	struct mlx5_core_dev *dev = devlink_priv(devlink);
2488 
2489 	if (!MLX5_ESWITCH_MANAGER(dev))
2490 		return -EOPNOTSUPP;
2491 
2492 	if (ctx->val.vbool)
2493 		return mlx5_lag_mpesw_enable(dev);
2494 
2495 	mlx5_lag_mpesw_disable(dev);
2496 	return 0;
2497 }
2498 
2499 static int mlx5_devlink_esw_multiport_get(struct devlink *devlink, u32 id,
2500 					  struct devlink_param_gset_ctx *ctx,
2501 					  struct netlink_ext_ack *extack)
2502 {
2503 	struct mlx5_core_dev *dev = devlink_priv(devlink);
2504 
2505 	ctx->val.vbool = mlx5_lag_is_mpesw(dev);
2506 	return 0;
2507 }
2508 
2509 static const struct devlink_param mlx5_eswitch_params[] = {
2510 	DEVLINK_PARAM_DRIVER(MLX5_DEVLINK_PARAM_ID_ESW_MULTIPORT,
2511 			     "esw_multiport", DEVLINK_PARAM_TYPE_BOOL,
2512 			     BIT(DEVLINK_PARAM_CMODE_RUNTIME),
2513 			     mlx5_devlink_esw_multiport_get,
2514 			     mlx5_devlink_esw_multiport_set, NULL),
2515 };
2516 
2517 int mlx5_eswitch_init(struct mlx5_core_dev *dev)
2518 {
2519 	struct mlx5_eswitch *esw;
2520 	int err;
2521 
2522 	if (!MLX5_VPORT_MANAGER(dev) && !MLX5_ESWITCH_MANAGER(dev))
2523 		return 0;
2524 
2525 	esw = kzalloc_obj(*esw);
2526 	if (!esw)
2527 		return -ENOMEM;
2528 
2529 	err = devl_params_register(priv_to_devlink(dev), mlx5_eswitch_params,
2530 				   ARRAY_SIZE(mlx5_eswitch_params));
2531 	if (err)
2532 		goto free_esw;
2533 
2534 	esw->dev = dev;
2535 	dev->priv.eswitch = esw;
2536 	esw->manager_vport = mlx5_eswitch_manager_vport(dev);
2537 	esw->first_host_vport = mlx5_eswitch_first_host_vport_num(dev);
2538 
2539 	esw->debugfs_root = debugfs_create_dir("esw", mlx5_debugfs_get_dev_root(dev));
2540 	esw->work_queue = create_singlethread_workqueue("mlx5_esw_wq");
2541 	if (!esw->work_queue) {
2542 		err = -ENOMEM;
2543 		goto abort;
2544 	}
2545 
2546 	err = mlx5_esw_host_functions_enabled_query(esw);
2547 	if (err)
2548 		goto abort;
2549 
2550 	err = mlx5_esw_vports_init(esw);
2551 	if (err)
2552 		goto abort;
2553 
2554 	err = esw_offloads_init(esw);
2555 	if (err)
2556 		goto reps_err;
2557 
2558 	esw->mode = MLX5_ESWITCH_LEGACY;
2559 
2560 	mutex_init(&esw->offloads.encap_tbl_lock);
2561 	hash_init(esw->offloads.encap_tbl);
2562 	mutex_init(&esw->offloads.decap_tbl_lock);
2563 	hash_init(esw->offloads.decap_tbl);
2564 	mlx5e_mod_hdr_tbl_init(&esw->offloads.mod_hdr);
2565 	atomic64_set(&esw->offloads.num_flows, 0);
2566 	ida_init(&esw->offloads.vport_metadata_ida);
2567 	xa_init_flags(&esw->offloads.vhca_map, XA_FLAGS_ALLOC);
2568 	xa_init(&esw->vhca_type_map);
2569 	mutex_init(&esw->state_lock);
2570 	init_rwsem(&esw->mode_lock);
2571 	refcount_set(&esw->qos.refcnt, 0);
2572 	init_waitqueue_head(&esw->work_queue_wait);
2573 	atomic_set(&esw->generation, 0);
2574 
2575 	esw->enabled_vports = 0;
2576 	esw->offloads.inline_mode = MLX5_INLINE_MODE_NONE;
2577 	if (MLX5_CAP_ESW_FLOWTABLE_FDB(dev, reformat) &&
2578 	    MLX5_CAP_ESW_FLOWTABLE_FDB(dev, decap))
2579 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_BASIC;
2580 	else
2581 		esw->offloads.encap = DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2582 
2583 	esw_info(dev,
2584 		 "Total vports %d, per vport: max uc(%d) max mc(%d)\n",
2585 		 esw->total_vports,
2586 		 MLX5_MAX_UC_PER_VPORT(dev),
2587 		 MLX5_MAX_MC_PER_VPORT(dev));
2588 	return 0;
2589 
2590 reps_err:
2591 	mlx5_esw_vports_cleanup(esw);
2592 	dev->priv.eswitch = NULL;
2593 abort:
2594 	if (esw->work_queue)
2595 		destroy_workqueue(esw->work_queue);
2596 	debugfs_remove_recursive(esw->debugfs_root);
2597 	devl_params_unregister(priv_to_devlink(dev), mlx5_eswitch_params,
2598 			       ARRAY_SIZE(mlx5_eswitch_params));
2599 free_esw:
2600 	kfree(esw);
2601 	return err;
2602 }
2603 
2604 void mlx5_eswitch_cleanup(struct mlx5_eswitch *esw)
2605 {
2606 	if (!esw)
2607 		return;
2608 
2609 	esw_info(esw->dev, "cleanup\n");
2610 
2611 	mlx5_eswitch_invalidate_wq(esw);
2612 	destroy_workqueue(esw->work_queue);
2613 	WARN_ON(refcount_read(&esw->qos.refcnt));
2614 	mutex_destroy(&esw->state_lock);
2615 	WARN_ON(!xa_empty(&esw->offloads.vhca_map));
2616 	xa_destroy(&esw->offloads.vhca_map);
2617 	xa_destroy(&esw->vhca_type_map);
2618 	ida_destroy(&esw->offloads.vport_metadata_ida);
2619 	mlx5e_mod_hdr_tbl_destroy(&esw->offloads.mod_hdr);
2620 	mutex_destroy(&esw->offloads.encap_tbl_lock);
2621 	mutex_destroy(&esw->offloads.decap_tbl_lock);
2622 	esw_offloads_cleanup(esw);
2623 	esw->dev->priv.eswitch = NULL;
2624 	mlx5_esw_vports_cleanup(esw);
2625 	debugfs_remove_recursive(esw->debugfs_root);
2626 	devl_params_unregister(priv_to_devlink(esw->dev), mlx5_eswitch_params,
2627 			       ARRAY_SIZE(mlx5_eswitch_params));
2628 	kfree(esw);
2629 }
2630 
2631 /* Vport Administration */
2632 static int
2633 mlx5_esw_set_vport_mac_locked(struct mlx5_eswitch *esw,
2634 			      struct mlx5_vport *evport, const u8 *mac)
2635 {
2636 	u16 vport_num = evport->vport;
2637 	u64 node_guid;
2638 	int err = 0;
2639 
2640 	if (is_multicast_ether_addr(mac))
2641 		return -EINVAL;
2642 
2643 	if (evport->info.spoofchk && !is_valid_ether_addr(mac))
2644 		mlx5_core_warn(esw->dev,
2645 			       "Set invalid MAC while spoofchk is on, vport(%d)\n",
2646 			       vport_num);
2647 
2648 	err = mlx5_modify_nic_vport_mac_address(esw->dev, vport_num, mac);
2649 	if (err) {
2650 		mlx5_core_warn(esw->dev,
2651 			       "Failed to mlx5_modify_nic_vport_mac vport(%d) err=(%d)\n",
2652 			       vport_num, err);
2653 		return err;
2654 	}
2655 
2656 	node_guid_gen_from_mac(&node_guid, mac);
2657 	err = mlx5_modify_nic_vport_node_guid(esw->dev, vport_num, node_guid);
2658 	if (err)
2659 		mlx5_core_warn(esw->dev,
2660 			       "Failed to set vport %d node guid, err = %d. RDMA_CM will not function properly for this VF.\n",
2661 			       vport_num, err);
2662 
2663 	ether_addr_copy(evport->info.mac, mac);
2664 	evport->info.node_guid = node_guid;
2665 	if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY)
2666 		err = esw_acl_ingress_lgcy_setup(esw, evport);
2667 
2668 	return err;
2669 }
2670 
2671 int mlx5_eswitch_set_vport_mac(struct mlx5_eswitch *esw,
2672 			       u16 vport, const u8 *mac)
2673 {
2674 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2675 	int err = 0;
2676 
2677 	if (IS_ERR(evport))
2678 		return PTR_ERR(evport);
2679 
2680 	mutex_lock(&esw->state_lock);
2681 	err = mlx5_esw_set_vport_mac_locked(esw, evport, mac);
2682 	mutex_unlock(&esw->state_lock);
2683 	return err;
2684 }
2685 
2686 static bool mlx5_esw_check_port_type(struct mlx5_eswitch *esw, u16 vport_num, xa_mark_t mark)
2687 {
2688 	return xa_get_mark(&esw->vports, vport_num, mark);
2689 }
2690 
2691 bool mlx5_eswitch_is_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2692 {
2693 	return mlx5_esw_check_port_type(esw, vport_num, MLX5_ESW_VPT_VF);
2694 }
2695 
2696 int mlx5_esw_spf_vport_to_idx(struct mlx5_eswitch *esw, u16 vport_num)
2697 {
2698 	struct mlx5_esw_functions *esw_funcs = &esw->esw_funcs;
2699 	int i;
2700 
2701 	for (i = 0; i < esw_funcs->num_spfs; i++) {
2702 		if (esw_funcs->spfs[i].vport_num == vport_num)
2703 			return i;
2704 	}
2705 
2706 	return -ENOENT;
2707 }
2708 
2709 bool mlx5_esw_is_spf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2710 {
2711 	return mlx5_esw_spf_vport_to_idx(esw, vport_num) >= 0;
2712 }
2713 
2714 bool mlx5_eswitch_is_pf_vf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2715 {
2716 	return vport_num == MLX5_VPORT_HOST_PF ||
2717 		mlx5_eswitch_is_vf_vport(esw, vport_num) ||
2718 		mlx5_esw_is_spf_vport(esw, vport_num);
2719 }
2720 
2721 bool mlx5_esw_is_sf_vport(struct mlx5_eswitch *esw, u16 vport_num)
2722 {
2723 	return mlx5_esw_check_port_type(esw, vport_num, MLX5_ESW_VPT_SF);
2724 }
2725 
2726 int mlx5_eswitch_set_vport_state(struct mlx5_eswitch *esw,
2727 				 u16 vport, int link_state)
2728 {
2729 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2730 	int opmod = MLX5_VPORT_STATE_OP_MOD_ESW_VPORT;
2731 	int other_vport = 1;
2732 	int err = 0;
2733 
2734 	if (!mlx5_esw_allowed(esw))
2735 		return -EPERM;
2736 	if (IS_ERR(evport))
2737 		return PTR_ERR(evport);
2738 
2739 	if (vport == MLX5_VPORT_UPLINK) {
2740 		opmod = MLX5_VPORT_STATE_OP_MOD_UPLINK;
2741 		other_vport = 0;
2742 		vport = 0;
2743 	}
2744 	mutex_lock(&esw->state_lock);
2745 	if (esw->mode != MLX5_ESWITCH_LEGACY) {
2746 		err = -EOPNOTSUPP;
2747 		goto unlock;
2748 	}
2749 
2750 	err = mlx5_modify_vport_admin_state(esw->dev, opmod, vport, other_vport, link_state);
2751 	if (err) {
2752 		mlx5_core_warn(esw->dev, "Failed to set vport %d link state, opmod = %d, err = %d",
2753 			       vport, opmod, err);
2754 		goto unlock;
2755 	}
2756 
2757 	evport->info.link_state = link_state;
2758 
2759 unlock:
2760 	mutex_unlock(&esw->state_lock);
2761 	return err;
2762 }
2763 
2764 int mlx5_eswitch_get_vport_config(struct mlx5_eswitch *esw,
2765 				  u16 vport, struct ifla_vf_info *ivi)
2766 {
2767 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2768 	u32 max_rate, min_rate;
2769 
2770 	if (IS_ERR(evport))
2771 		return PTR_ERR(evport);
2772 
2773 	memset(ivi, 0, sizeof(*ivi));
2774 	ivi->vf = vport - 1;
2775 
2776 	mutex_lock(&esw->state_lock);
2777 
2778 	mlx5_query_nic_vport_mac_address(esw->dev, vport, true,
2779 					 evport->info.mac);
2780 	ether_addr_copy(ivi->mac, evport->info.mac);
2781 	ivi->linkstate = evport->info.link_state;
2782 	ivi->vlan = evport->info.vlan;
2783 	ivi->qos = evport->info.qos;
2784 	ivi->spoofchk = evport->info.spoofchk;
2785 	ivi->trusted = evport->info.trusted;
2786 
2787 	if (mlx5_esw_qos_get_vport_rate(evport, &max_rate, &min_rate)) {
2788 		ivi->max_tx_rate = max_rate;
2789 		ivi->min_tx_rate = min_rate;
2790 	}
2791 	mutex_unlock(&esw->state_lock);
2792 
2793 	return 0;
2794 }
2795 
2796 int __mlx5_eswitch_set_vport_vlan(struct mlx5_eswitch *esw,
2797 				  u16 vport, u16 vlan, u8 qos, u8 set_flags)
2798 {
2799 	struct mlx5_vport *evport = mlx5_eswitch_get_vport(esw, vport);
2800 	bool vst_mode_steering = esw_vst_mode_is_steering(esw);
2801 	int err = 0;
2802 
2803 	if (IS_ERR(evport))
2804 		return PTR_ERR(evport);
2805 	if (vlan > 4095 || qos > 7)
2806 		return -EINVAL;
2807 
2808 	if (esw->mode == MLX5_ESWITCH_OFFLOADS || !vst_mode_steering) {
2809 		err = modify_esw_vport_cvlan(esw->dev, vport, vlan, qos, set_flags);
2810 		if (err)
2811 			return err;
2812 	}
2813 
2814 	evport->info.vlan = vlan;
2815 	evport->info.qos = qos;
2816 	if (evport->enabled && esw->mode == MLX5_ESWITCH_LEGACY) {
2817 		err = esw_acl_ingress_lgcy_setup(esw, evport);
2818 		if (err)
2819 			return err;
2820 		err = esw_acl_egress_lgcy_setup(esw, evport);
2821 	}
2822 
2823 	return err;
2824 }
2825 
2826 int mlx5_eswitch_get_vport_stats(struct mlx5_eswitch *esw,
2827 				 u16 vport_num,
2828 				 struct ifla_vf_stats *vf_stats)
2829 {
2830 	struct mlx5_vport *vport = mlx5_eswitch_get_vport(esw, vport_num);
2831 	int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out);
2832 	u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)] = {};
2833 	struct mlx5_vport_drop_stats stats = {};
2834 	int err = 0;
2835 	u32 *out;
2836 
2837 	if (IS_ERR(vport))
2838 		return PTR_ERR(vport);
2839 
2840 	out = kvzalloc(outlen, GFP_KERNEL);
2841 	if (!out)
2842 		return -ENOMEM;
2843 
2844 	MLX5_SET(query_vport_counter_in, in, opcode,
2845 		 MLX5_CMD_OP_QUERY_VPORT_COUNTER);
2846 	MLX5_SET(query_vport_counter_in, in, op_mod, 0);
2847 	MLX5_SET(query_vport_counter_in, in, vport_number, vport->vport);
2848 	MLX5_SET(query_vport_counter_in, in, other_vport, 1);
2849 
2850 	err = mlx5_cmd_exec_inout(esw->dev, query_vport_counter, in, out);
2851 	if (err)
2852 		goto free_out;
2853 
2854 	#define MLX5_GET_CTR(p, x) \
2855 		MLX5_GET64(query_vport_counter_out, p, x)
2856 
2857 	memset(vf_stats, 0, sizeof(*vf_stats));
2858 	vf_stats->rx_packets =
2859 		MLX5_GET_CTR(out, received_eth_unicast.packets) +
2860 		MLX5_GET_CTR(out, received_ib_unicast.packets) +
2861 		MLX5_GET_CTR(out, received_eth_multicast.packets) +
2862 		MLX5_GET_CTR(out, received_ib_multicast.packets) +
2863 		MLX5_GET_CTR(out, received_eth_broadcast.packets);
2864 
2865 	vf_stats->rx_bytes =
2866 		MLX5_GET_CTR(out, received_eth_unicast.octets) +
2867 		MLX5_GET_CTR(out, received_ib_unicast.octets) +
2868 		MLX5_GET_CTR(out, received_eth_multicast.octets) +
2869 		MLX5_GET_CTR(out, received_ib_multicast.octets) +
2870 		MLX5_GET_CTR(out, received_eth_broadcast.octets);
2871 
2872 	vf_stats->tx_packets =
2873 		MLX5_GET_CTR(out, transmitted_eth_unicast.packets) +
2874 		MLX5_GET_CTR(out, transmitted_ib_unicast.packets) +
2875 		MLX5_GET_CTR(out, transmitted_eth_multicast.packets) +
2876 		MLX5_GET_CTR(out, transmitted_ib_multicast.packets) +
2877 		MLX5_GET_CTR(out, transmitted_eth_broadcast.packets);
2878 
2879 	vf_stats->tx_bytes =
2880 		MLX5_GET_CTR(out, transmitted_eth_unicast.octets) +
2881 		MLX5_GET_CTR(out, transmitted_ib_unicast.octets) +
2882 		MLX5_GET_CTR(out, transmitted_eth_multicast.octets) +
2883 		MLX5_GET_CTR(out, transmitted_ib_multicast.octets) +
2884 		MLX5_GET_CTR(out, transmitted_eth_broadcast.octets);
2885 
2886 	vf_stats->multicast =
2887 		MLX5_GET_CTR(out, received_eth_multicast.packets) +
2888 		MLX5_GET_CTR(out, received_ib_multicast.packets);
2889 
2890 	vf_stats->broadcast =
2891 		MLX5_GET_CTR(out, received_eth_broadcast.packets);
2892 
2893 	err = mlx5_esw_query_vport_drop_stats(esw->dev, vport, &stats);
2894 	if (err)
2895 		goto free_out;
2896 	vf_stats->rx_dropped = stats.rx_dropped;
2897 	vf_stats->tx_dropped = stats.tx_dropped;
2898 
2899 free_out:
2900 	kvfree(out);
2901 	return err;
2902 }
2903 
2904 u8 mlx5_eswitch_mode(const struct mlx5_core_dev *dev)
2905 {
2906 	struct mlx5_eswitch *esw = dev->priv.eswitch;
2907 
2908 	return mlx5_esw_allowed(esw) ? esw->mode : MLX5_ESWITCH_LEGACY;
2909 }
2910 EXPORT_SYMBOL_GPL(mlx5_eswitch_mode);
2911 
2912 enum devlink_eswitch_encap_mode
2913 mlx5_eswitch_get_encap_mode(const struct mlx5_core_dev *dev)
2914 {
2915 	struct mlx5_eswitch *esw;
2916 
2917 	esw = dev->priv.eswitch;
2918 	return (mlx5_eswitch_mode(dev) == MLX5_ESWITCH_OFFLOADS)  ? esw->offloads.encap :
2919 		DEVLINK_ESWITCH_ENCAP_MODE_NONE;
2920 }
2921 EXPORT_SYMBOL(mlx5_eswitch_get_encap_mode);
2922 
2923 bool mlx5_esw_multipath_prereq(struct mlx5_core_dev *dev0,
2924 			       struct mlx5_core_dev *dev1)
2925 {
2926 	return (dev0->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS &&
2927 		dev1->priv.eswitch->mode == MLX5_ESWITCH_OFFLOADS);
2928 }
2929 
2930 int mlx5_esw_event_notifier_register(struct mlx5_core_dev *dev,
2931 				     struct notifier_block *nb)
2932 {
2933 	return blocking_notifier_chain_register(&dev->priv.esw_n_head, nb);
2934 }
2935 
2936 void mlx5_esw_event_notifier_unregister(struct mlx5_core_dev *dev,
2937 					struct notifier_block *nb)
2938 {
2939 	blocking_notifier_chain_unregister(&dev->priv.esw_n_head, nb);
2940 }
2941 
2942 /**
2943  * mlx5_esw_hold() - Try to take a read lock on esw mode lock.
2944  * @mdev: mlx5 core device.
2945  *
2946  * Should be called by esw resources callers.
2947  *
2948  * Return: true on success or false.
2949  */
2950 bool mlx5_esw_hold(struct mlx5_core_dev *mdev)
2951 {
2952 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2953 
2954 	/* e.g. VF doesn't have eswitch so nothing to do */
2955 	if (!mlx5_esw_allowed(esw))
2956 		return true;
2957 
2958 	if (down_read_trylock(&esw->mode_lock) != 0) {
2959 		if (esw->eswitch_operation_in_progress) {
2960 			up_read(&esw->mode_lock);
2961 			return false;
2962 		}
2963 		return true;
2964 	}
2965 
2966 	return false;
2967 }
2968 
2969 /**
2970  * mlx5_esw_release() - Release a read lock on esw mode lock.
2971  * @mdev: mlx5 core device.
2972  */
2973 void mlx5_esw_release(struct mlx5_core_dev *mdev)
2974 {
2975 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2976 
2977 	if (mlx5_esw_allowed(esw))
2978 		up_read(&esw->mode_lock);
2979 }
2980 
2981 /**
2982  * mlx5_esw_get() - Increase esw user count.
2983  * @mdev: mlx5 core device.
2984  */
2985 void mlx5_esw_get(struct mlx5_core_dev *mdev)
2986 {
2987 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
2988 
2989 	if (mlx5_esw_allowed(esw))
2990 		atomic64_inc(&esw->user_count);
2991 }
2992 
2993 /**
2994  * mlx5_esw_put() - Decrease esw user count.
2995  * @mdev: mlx5 core device.
2996  */
2997 void mlx5_esw_put(struct mlx5_core_dev *mdev)
2998 {
2999 	struct mlx5_eswitch *esw = mdev->priv.eswitch;
3000 
3001 	if (mlx5_esw_allowed(esw))
3002 		atomic64_dec_if_positive(&esw->user_count);
3003 }
3004 
3005 /**
3006  * mlx5_esw_try_lock() - Take a write lock on esw mode lock.
3007  * @esw: eswitch device.
3008  *
3009  * Should be called by esw mode change routine.
3010  *
3011  * Return:
3012  * * 0       - esw mode if successfully locked and refcount is 0.
3013  * * -EBUSY  - refcount is not 0.
3014  * * -EINVAL - In the middle of switching mode or lock is already held.
3015  */
3016 int mlx5_esw_try_lock(struct mlx5_eswitch *esw)
3017 {
3018 	if (down_write_trylock(&esw->mode_lock) == 0)
3019 		return -EINVAL;
3020 
3021 	if (esw->eswitch_operation_in_progress ||
3022 	    atomic64_read(&esw->user_count) > 0) {
3023 		up_write(&esw->mode_lock);
3024 		return -EBUSY;
3025 	}
3026 
3027 	return esw->mode;
3028 }
3029 
3030 int mlx5_esw_lock(struct mlx5_eswitch *esw)
3031 {
3032 	down_write(&esw->mode_lock);
3033 
3034 	if (esw->eswitch_operation_in_progress) {
3035 		up_write(&esw->mode_lock);
3036 		return -EBUSY;
3037 	}
3038 
3039 	return 0;
3040 }
3041 
3042 /**
3043  * mlx5_esw_unlock() - Release write lock on esw mode lock
3044  * @esw: eswitch device.
3045  */
3046 void mlx5_esw_unlock(struct mlx5_eswitch *esw)
3047 {
3048 	up_write(&esw->mode_lock);
3049 }
3050 
3051 /**
3052  * mlx5_eswitch_get_total_vports - Get total vports of the eswitch
3053  *
3054  * @dev: Pointer to core device
3055  *
3056  * mlx5_eswitch_get_total_vports returns total number of eswitch vports.
3057  */
3058 u16 mlx5_eswitch_get_total_vports(const struct mlx5_core_dev *dev)
3059 {
3060 	struct mlx5_eswitch *esw;
3061 
3062 	esw = dev->priv.eswitch;
3063 	return mlx5_esw_allowed(esw) ? esw->total_vports : 0;
3064 }
3065 EXPORT_SYMBOL_GPL(mlx5_eswitch_get_total_vports);
3066 
3067 /**
3068  * mlx5_eswitch_get_core_dev - Get the mdev device
3069  * @esw : eswitch device.
3070  *
3071  * Return the mellanox core device which manages the eswitch.
3072  */
3073 struct mlx5_core_dev *mlx5_eswitch_get_core_dev(struct mlx5_eswitch *esw)
3074 {
3075 	return mlx5_esw_allowed(esw) ? esw->dev : NULL;
3076 }
3077 EXPORT_SYMBOL(mlx5_eswitch_get_core_dev);
3078 
3079 bool mlx5_eswitch_block_ipsec(struct mlx5_core_dev *dev)
3080 {
3081 	struct mlx5_eswitch *esw = dev->priv.eswitch;
3082 
3083 	if (!mlx5_esw_allowed(esw))
3084 		return true;
3085 
3086 	mutex_lock(&esw->state_lock);
3087 	if (esw->enabled_ipsec_vf_count) {
3088 		mutex_unlock(&esw->state_lock);
3089 		return false;
3090 	}
3091 
3092 	dev->num_ipsec_offloads++;
3093 	mutex_unlock(&esw->state_lock);
3094 	return true;
3095 }
3096 
3097 void mlx5_eswitch_unblock_ipsec(struct mlx5_core_dev *dev)
3098 {
3099 	struct mlx5_eswitch *esw = dev->priv.eswitch;
3100 
3101 	if (!mlx5_esw_allowed(esw))
3102 		/* Failure means no eswitch => core dev is not a PF */
3103 		return;
3104 
3105 	mutex_lock(&esw->state_lock);
3106 	dev->num_ipsec_offloads--;
3107 	mutex_unlock(&esw->state_lock);
3108 }
3109 
3110 bool mlx5_esw_host_functions_enabled(const struct mlx5_core_dev *dev)
3111 {
3112 	if (!dev->priv.eswitch)
3113 		return true;
3114 
3115 	return !dev->priv.eswitch->esw_funcs.host_funcs_disabled;
3116 }
3117